In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
I 
637 
the second generation ADS40-SH52 (including the NIR band) 
is superior to the summer RGB images of the first generation 
ADS40-SH40. 
The study shows that logistic regression models proved to have 
a high potential to produce meaningful tree species 
classifications with a minimum amount of effort involved in 
image acquisition, data pre-processing, derivation of 
explanatory variables and field work. Some limitations of this 
approach are briefly discussed below. 
5.1 Ground truth 
The tree samples were delineated in the field on aerial images, 
which means that well visible trees may have been preferred, or 
only the lighted parts of trees have been delineated. 
Additionally, trees may be shaded or partly hidden by others so 
that one image segment could contain more than one species. 
However, when comparing correct classification rates or kappa 
values to other studies, we emphasize that this is a qualitative 
approach. For the same reasons the model results were checked 
for plausibility by visual examination of the aerial photographs. 
These uncertainties render the statistical evaluations relative. 
5.2 Comparison with other studies 
Overall, the species accuracies obtained in this study are in the 
line or higher with those in similar studies. 
Our best result (spring 2007 data) with an overall accuracy of 
nearly 80% for seven tree species is higher to those obtained in 
other studies. Overall accuracies between 75% (based on CIR 
aerial images, Brandtberg, 2002) and 89% (based on DMC 
camera, Olofsson et al., 2006) are obtained in most studies to 
classify Norway spruce, Scots pine, birch or aspen 
Obviously, classification accuracies are lower the more tree 
species there are and if non-dominant tree species are included 
as well. Chubey et al. (2009) classified 4-6 coniferous and 4-6 
deciduous species in Canadian forests with an overall accuracy 
around 70%. 
5.3 Multispectral versus multi-temporal 
Although we found that our approach produces in general good 
results and is suitable a more detailed analysis of the 
misclassifications is needed. The full potential of a multi 
temporal approach could not be realized in this study. Due to 
differences in the flight paths and different acquisition daytimes 
(different shadows) between the 2007 and 2008 images a 
classification based on all three datasets could not be 
established. The 2008 data was therefore used separately. 
Although multi-temporal multispectral data is known as 
valuable (e.g. Key et ah, 2001), in the present study 
combinations of the two images of May and July 2007 tended 
to give lower accuracies. For the classification of Larix decidua 
and Picea abies, the single usage of multispectral information 
obtained by the August 2008 imagery was more valuable than 
multi-temporal information of the May and July 2007 imagery. 
The reason for this might be the additional usage of the NIR 
information provided by the ADS40-SH52 2008 images. 
Problems for classifying deciduous tree species are increasing 
when using summer imagery. Visual analysis of the spectral 
ranges of each species moreover revealed very similar spectral 
properties between the summer 2007 and 2008 images for 
Fagus sylvatica and Fraxinus excelsior. Even within species, 
spectral variability can be large because of illumination and 
view-angle conditions, openness of trees, natural variability, 
age of the trees, shadowing effects and differences in crown 
health. Fig. 4 illustrates this situation. 
Figure 4. Examples of the deciduous tree species Fagus 
sylvatica (blue-grey), Fraxinus excelsior (light green) and Acer 
sp. (green-blue) as they appear in the May 2007 images (left) 
and July 2007 images (right). 
5.4 Non-dominant tree species 
Generally, a relatively small sample size of non-dominant tree 
species - compared to the other species in a study area - leads 
to underestimation of these species. Tables 4-6 clearly reveal 
that most frequent failures happen in classifying the non 
dominant tree species Acer sp. Visual image inspection showed 
that Acer sp. are often short and therefore partly obscured by 
nearby large and dominant trees, or by the merging of close 
crowns. The two other non-dominant (coniferous) tree species 
Larix decidua and Pinus sylvestris in this study are classified 
with higher accuracies. 
6. OUTLOOK 
The promising results and experiences made in this study are of 
great practical interest for the Swiss National Forest Inventory. 
Actual and accurate maps of tree species and composition are 
needed by environmental agencies and land surveying offices to 
assess possible changes in species distribution or condition of 
other habitat. 
The most obvious opportunities for follow up are: The usage of 
NFI field sample plots as training data to reduce field work. 
Further development is needed for testing larger areas, which 
may consist of several image strips. BRDF-related problems or 
influences of the BRDF in terms of classification accuracy 
should also be investigated. 
7. REFERENCES 
Akaike, H.,1973. Information theory as an extension of the 
maximum likelihood principle. In: Petrov, B.N., Csaki, F. 
(eds.), Second International Symposium on Information Theory. 
Akademiai Kiado, Budapest, Hungary, pp. 267-281. 
Artuso, R., Bovet, S., and Streilein, A., 2003. Pratical Methods 
for the Verification of countrywide Terrain and Surface 
Models, In: International Archives of Photogrammetry and 
Remote Sensing, vol. XXXIV-3/W13