In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
from the segmentation process. Figure 4 shows details of the 
riparian formations. 
45°201D“W 45WW 44°40'0 , W 
Figure 2: Land-use map and location of study area. 
4***r*5W 
*t'430MV «•42’45’W 
Figure 3: Location of entire study site with the 70 plots, 35 on 
each side. 
(a) (b) 
Figure 4: Riparian forest of study site: (a) aerial view and (b) 
ground view (photos by Thais Amaral and Ivan Seixas - 
04/2010). 
2. MATERIAL AND METHODS 
2.1 Fieldwork 
The first field campaign (February 2008) was carried out to 
locate and collect allometric data. The group from the 
“Ecology and Plant Propagation Laboratory” (UNIMONTES 
University) demarcated 70 ten by ten meter plots, along both 
banks of the Pandeiros River. The plots were at a distance of 
3 meters from the Pandeiros River and are always oriented 
parallel to the Pandeiros River. A 10 m gap is always left 
between each plot except when the area was too degraded to 
be considered as riparian forest. Plots 1-35 are located on the 
left bank while plots 36-70 on the right bank. A total of 7000 
m 2 was surveyed. 
The allometric measurements of tree height and stem 
circumference at breast height (CBH) were taken for all trees 
within each plot. Shrubs and grasses were not considered. 
Hemispherical photographs were taken in each plot for later 
computation of canopy openness and leaf area index. 
Altogether seven allometric measurements were produced: (i) 
tree height, (ii) stem diameter at breast height (DBH), (iii) 
basal area, (iv) volume, (v) plot density, (vi) canopy openness 
and (vii) leaf area index (LAI). 
The second and third fieldworks were conducted in January 
2009 and April 2010, respectively, to obtain ground control 
points (GCPs). A LI geodetic GPS was employed to acquire 
GCPs, which were used for rectifying the Ikonos scene and 
precisely locating each plot. 
2.2 Image Acquisition and Pre-processing 
The Ikonos image used in this study was provided by the 
Forestry Institute of Minas Gerais. It was obtained with their 
multispectral bands (red, green and blue = 4m) and 
panchromatic (lm) already fused to a spatial resolution of 1 
m. The data, acquired in September 2007 during the dry 
season showed a good visibility with no cloud cover (Figure 
3). The image was registered to a UTM grid coordinate by 
bilinear interpolation with a root mean square error (RMSE) 
less than 1 meter. No atmospheric correction was applied to 
the image. 
2.3 Data Processing 
Data processing involved four steps: 1) cartographic 
modelling, 2) image segmentation, 3) texture feature 
calculation and 4) statistical modelling. 
2.3.1 Cartographic Modelling 
The cartographic modelling consisted in using spatial 
knowledge to “limit” the search to areas having a strong 
probability of belonging to riparian vegetation (Maillard et. 
al., 2008). This method avoids confusion between vegetation 
classes present in the study area, such as: palm swamps and 
savannah. To do so, the hydrographic network was digitized, 
overlaid on the image and used to build a buffer of 50 m 
(knowing the riparian vegetation width in the study site is 
well below that value). The buffer was used to mask parts of 
the image that fell outside of it. The Ikonos image and the 
mask are then fed to the segmentation algorithm which is 
instructed to find two classes: riparian vegetation e non 
riparian vegetation.