*S, Vol. XXXVIII, Part 7B 
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
27 
ES BASED ON 
RE ANALYSIS 
nte, Brazil 
ure of these environments, 
•esolution images, such as 
ethodology for delineating 
ed into two stages. Firstly, 
n and a river-based buffer, 
rest zones from the rest. In 
parameters of the riparian 
River, located in Northern 
leters. The forest structure 
and leaf area index which 
irian environment with an 
: best results for modeling 
Angular Second Moment, 
npte tenu de l'étroitesse de 
s produits obtenus à partir 
es. L'objectif de cet article 
aètres biophysiques à partir 
lasse de forêt riveraine et le 
îr la rivière. Le programme 
riveraines du reste. Dans la 
iur estimer les paramètres 
zone riveraine sur les deux 
:es pour créer et valider les 
r, le diamètre à hauteur de 
canopée, qui ont été acquis 
avec une précision de plus 
isation de la structure de la 
1,61) en utilisant le Second 
<11 pixels. 
•idors allowing the flow of 
in fragmented landscapes 
are impacted by logging to 
margins, livestock and others 
effective inventory of these 
mt tool for making public 
nsing and image processing 
yw-cost production of maps 
s inventories using remote 
i its narrow extends (Muller, 
1997). Mapping these areas and their biophysical parameters is a 
challenge that has motivated many authors in remote sensing 
(Nagler et al., 2001; CSIRO, 2003; Johansen and Phinn, 2006b). 
Previous studies have showed that images with medium spatial 
resolution (Landsat-TM and ETM) suffer shortcomings for 
mapping narrow environments (< 30m), such as riparian zones 
(Congalton et al., 2002; Johansen and Phinn, 2006a). Moreover, 
the alternative to use high-resolution image (Ikonos, Quickbird), 
has become affordable in recent years. The most commonly 
applied approach to map riparian areas using high spatial 
resolution is image classification. Studies by Davis et al. (2002) 
and Johansen and Phinn (2006a) showed a significant gain of 
accuracy in classification of riparian zones when using texture 
parameters in the process. 
Taking advantage of the spatial knowledge that the riparian 
vegetation accompanies the river, buffer zones can be used as a 
way of optimizing the image processing. This procedure was 
carried out successfully to map palm swamps (Maillard et. al., 
2008). Even though high resolution image data proved valuable 
for delineating riparian zones, traditional information extraction 
methods like threshold and classification (e.g. maximum 
likelihood, minimum distance) offer low accuracy. Conversely, 
image segmentation using Markov random fields (MRF) has 
produced promising results in a variety of applications, such as 
image segmentation and restoration (Tso and Mather, 2001). But 
to assess the ecological values of riparian forests, the mere 
classification is insufficient and biophysical parameters are often 
needed. 
The objective of this article is to describe a methodology for 
delineating riparian areas and extract their biophysical parameters 
from an Ikonos scene. The proposed methodology includes the 
following stages: (i) classification of the image in two classes: 
riparian zone and non-riparian zone using 50 meters buffer, (ii) 
acquisition of texture features from riparian zones segments, and 
(iii) auto-correlation of visible, near-infrared and texture bands 
with allometric measurements data from 70 field plots. The 
correlation aims at elaborating explanatory models of vegetation 
structure. 
1.1 Mapping Riparian Forest from Remote Sensing Data 
A study using Landsat-5 and photo interpretation, for mapping 
riparian forest in the Yaquina River - Oregon/USA, showed a 
success ratio of only 30% between satellite images and photo 
interpretations (Congalton et al., 2002). In another study, 
Johansen and Phinn (2006a) showed that the width of riparian 
zones is a limiting factor for their identification through products 
of medium spatial resolution, such as Landsat series. They 
pointed out that only riparian area upper than 50m could be 
accurately identified by Landsat-7 ETM+. Muller (1997) 
emphasizes the importance to develop new remote sensing 
methods for mapping riparian vegetation along rivers. 
In a study by Davis et al. (2002), the analysis of high-resolution 
aerial photographs (resolution between 11 and 100 cm) obtained 
overall accuracy of 75% to classify riparian areas using maximum 
likelihood classification and image texture. Texture features 
increased the accuracy by 20-30% in almost every case. Johansen 
and Phinn (2006b) used an Ikonos image to classify not only the 
riparian zone, but also the biophysical parameters and species of a 
riparian savannah forest in Australia. The authors highlighted the 
need to use high-resolution imagery and texture parameters for 
mapping riparian vegetation structures. They used the following 
forest parameters: canopy percentage foliage cover, leaf area 
index, tree crown size, tree height, stem diameter at breast 
height, tree species, and riparian zone width. In addition to 
the four bands of the Ikonos image (blue, green, red and near- 
infrared), eight vegetation index and measurements of texture 
(contrast, dissimilarity, entropy, homogeneity and variance) 
were used. Results showed an overall accuracy of 55% for 
species classification and a determination of 86% for the 
canopy percentage foliage using 19x19 pixels texture analysis 
window in the NDVI band. 
Alencar-Silva and Maillard (2009) compared two different 
methods of classification for palm swamps in an Ikonos 
image: traditional per-pixel classification and region-based 
segmentation and classification using MAGIC (a program 
based on Markov Random Fields). Results have shown that 
MAGIC obtained better results when compared with tradition 
classification. MAGIC was especially good in removing the 
salt and pepper effect on the classified image. 
1.2 Study Area 
The study area is situated in the margins of the Pandeiros 
River in Northern Minas Gerais, Brazil, an environmental 
protection area (Figure 2). 
The Pandeiros River is an affluent of Sao Francisco River, the 
third largest watershed of Brazil. The total area of the 
Pandeiros’ watershed is 3921.00 km 2 and its elevation varies 
from 450 m to 850 m. The study site is 1.2 km 2 along a 
slightly meandering stretch of river (Figure 3). The marginally 
climate is semiarid with about 900 mm of precipitation and an 
average temperature of over 25°C. Precipitation varies from 
124 mm per month between October and April to less than 2 
mm between May and September. A land use map of the 
region was produce from a single Landsat-7 scene acquired in 
August 2009. The classes and their respective area are 
presented in Table 1. 
Class 
% of catchment 
Open Water 
0.33 
Dry Forest 
1.08 
Savannah 
48.85 
Wetland 
3.09 
Plantation / Savannah Regeneration 
43.61 
Rock 
0.12 
Bare Soil 
2.92 
Table 1: Land-use table area of Pandeiros River 
With about 44% of plantation or degraded areas, the 
Pandeiros watershed has been strongly impacted by human 
activities. The Pandeiros also hosts the largest wetland 
complex of the State of Minas Gerais where several species of 
fish bird species reproduce, some of which are rare, endemic 
and threatened (Biodiversitas, 2005). 
Figure 3 shows the Ikonos scene of the entire study site in 
true-colour composition. The green areas located along the 
river and on the bottom right of the figure correspond to 
riparian forest and savannah formations, respectively. The 
zones in brown represent herbaceous areas. Palm swamps, 
characterised by a specific texture, can be seen on the left 
hand side of the image. The others light tone areas are bare 
soil. On the Ikonos scene riparian forest often appears similar 
to wooded savannah formation. To avoid confusion, a river 
buffer of 50 m was applied to the image to eliminate savannah