In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
634
Figure 1. Study area. © SWISSTOPO.
2.2 Ground truth
The ground truth data to validate the tree species classifications
was collected in the natural environment to be representative.
Two ground surveys were carried out in summer 2008 and 2009
focusing on the most frequent tree species (at least 10%
coverage in Switzerland) which were also visible in the aerial
images. For a total of 230 sampled trees we recorded the
species (table 1) and delineated in the field the crowns of all
visited trees on the corresponding aerial images. This
information was used as reference to digitize the corresponding
tree crowns on the ADS40 RGB images.
Scientific tree species name
(common tree species name)
Num. of
samples
Species
prop.
Acer sp. (maple)
28
< 10%
Fagus sylvatica (beech)
42
20%
Fraxinus excelsior (ash)
32
15%
Abies alba (white fir)
28
15%
Larix decidua (larch)
20
<10%
Picea abies (Norway spruce)
56
25%
Pinus sylvestris (Scots pine)
24
<10%
Table 1. Tree species sampled with number of samples.
Species proportion is based on estimates by an expert during the
field surveys
2.3 Remotely sensed data
2.3.1 Airborne Digital Sensor Data (ADS40): First
generation ADS40-SH40 and second generation ADS40-SH52
images Level 1 (Leica Geosystems AG, Switzerland) were
used in this study (for further details on the sensor see e.g.
Reulke et al. (2006). The main drawback of the first-generation
ADS40-SH40 is that the NIR line CCD is placed 18° forward
from the nadir RGB CCDs which makes it difficult to combine
all four lines. The second generation ADS40-SH52 provides the
NIR band in the same nadir position as the RGB bands. Three
Digital Surface Models (DSMs) were generated automatically
from the above images with a spatial resolution of 0.5 m using
modified strategies of NGATE of SOCET SET 5.4.1 (BAE
Systems). Prior to the DSM generation, a Wallis filter was
applied to enhance contrast, especially in shadow regions, and
to equalize radiometrically the images for matching.
Sensor
ADS40-SH40
ADS40-SH52
Acquisition date
24/05/2007 &
18/08/2008
Focal length
13/07/2007
62.8 mm
62.8 mm
Spectral
Red: 610-660
Red: 608-662
resolution (nm)
Green: 535-585
Green: 533-587
Blue: 430-490
Blue: 428-492
Ground pixel size
~25 cm
NIR: 833-887
~25 cm
Orthoimage
25 cm
25 cm
Radiometric
11 bit
11 bit
resolution
Table 2. Summary of characteristics of the image data used
2.3.2 LiDAR: National LiDAR digital terrain data (DTM)
produced by the Swiss Federal Office of Topography
(SWISSTOPO) for the study area (acquisistion date: March
2002, reflown March 2003 leaves-off) were used. The data were
acquired by Swissphoto AG / TerraPoint using a TerraPoint
ALTMS 2536 system with an average flying height above
ground of 1200 m. The DTM has an average point density of
0.8 points / m 2 height accuracy (1 sigma) of 0.5 m (Artuso et
al., 2003) and was interpolated to a regular grid with 0.25 m.
3. METHODS
3.1 Variables derived from ADS40 imagery
To extract tree area and classify tree species, several variables
(geometric and spectral signatures) were derived from the
remote sensing data using standard digital image processing
methods as described in e.g. Gonzales and Woods (2002).
Details about extraction of geometric and spectral explanatory
variables derived from airborne remote sensing data are
described in Waser et al. (2007, 2008a and 2008b). A good fit
to the given (training) data is not a sufficient condition for good
predictive models. To obtain good predictions, a small set of
powerful variables has to be selected.
Therefore stepwise variable selection (AIC, both directions,
Akaike, 1973) was applied using the defaults of R version 2.9.1.
A separate stepwise selection was performed for each tree
species. The variables were ranked according to their
contribution to the model.
The input variables used in this study consist of four commonly
used geometric parameters derived from the CHMs (slope,
curvature, and two local neighborhood functions). For further
details, see Burrough (1986) and Moore et al. (1991). Spectral
variables were derived from each of the three images. This
includes for each set of variables the mean and standard
deviations of: 3 x 3 original bands of ADS40-SH40 RGB and
ADS40-SH52 RGB and CIR images and the colour
transformation from RGB and CIR (only from the 2008 images)
to IHS into the 3 channels intensity (I), hue (H), and saturation
(S).
3.2 Image segmentation
Homogenous image segments of individual tree crowns or tree-
clusters are needed to classify tree species (see below). Both the
ADS40-SH40 and /ADS40-SH52 orthoimages were therefore
subdivided into patches by a multi-resolution segmentation
using the Definiens 7.0 software (Baatz & Schape, 2000).
Segmentation was iteratively optimized using several levels of
detail and adapted to shape and compactness parameters. The
final segmentation provided groups of trees and individual trees