EXTRACTION AND CLASSIFICATION OF WETLAND FEATURES THROUGH FUSION 
OF REMOTE SENSING IMAGES IN THE OKAVANGO DELTA, BOTSWANA 
K. Talukdar 
Institute of Geodesy and Photogrammetry, ETH Honggerberg, CH-8093 Ziirich, Switzerland - 
talukdar@geod.baug.ethz.ch 
y Commission VII, WG VII/3 
KEY WORDS: Remote sensing, land cover, classification, extraction, fusion, multisensor, multiresolution 
ABSTRACT: 
The Okavango delta in northwestern Botswana is an extremely complex and dynamic wetland ecosystem. The spatial information on 
diverse wetland features of the delta is needed for hydrological modeling and water resources management. Due to large size and 
inaccessibility of the delta, satellite images provide the only viable means to reliably map and measure these features. For better 
identification and delineation of these features in the Okavango delta, efficient image analysis techniques are needed. The synergistic 
use of images from different sensors with varied spatial and spectral resolutions have the potential for better extraction and 
classification of features. This paper focuses on extraction and classification of landscape and land cover features through fusion of 
different resolution images acquired by Landsat 7 ETM+ and SPOT 5 HRG sensors over the Okavango delta. Both multispectral and 
panchromatic images from these two sensors are used. Different image fusion approaches are examined and used to increase 
reliability in feature interpretation. The effects of data fusion in recognition and extraction features are examined and illustrated. 
Thematic information extraction was carried out by means of supervised and unsupervised classification to produce landscape/land 
cover classes for different spatial resolution data set. The results indicate that as spatial resolution increases, high spatial frequency 
landscape/land cover features are extracted in increasing detail. However, spatial heterogeneity also increases with increasing spatial 
resolution. 
viable means to reliably map and measure these features. Since 
the wetland is composed of heterogeneous objects it is difficult 
1. INTRODUCTION 
The spatial information on wetland features (i.e. landscape and to reliably identify and characterize features from satellite 
land cover) of the Okavango delta is needed for hydrological images. For better identification and delineation of these 
modeling and water resources management. Wetlands are by features in the Okavango delta, efficient image analysis 
definition lands with water-tables at or near the surface, either 
seasonally or permanently (Hughes, 1996). Wetlands and the 
issue of their management and preservation now engage the 
increasing attention of public. A goal is to provide better 
information as wetland ecosystems are influenced or exposed to 
environmental risks. There are a number of types and 
classification systems of wetlands to describe them (Anderson 
et al., 1976). They were categorised into different types and 
classes based broadly on hydrological, geomorphological, soil 
and vegetation characteristics of wetlands. There are a variety 
of information needs in wetland ecosystem management and 
landscape characterisation. 
techniques are needed. The synergistic use of images from 
The Okavango River, which originates in central Angola, after 
flowing through over a thousand kilometer branching out to 
form the Okavango delta, one of the largest inland wetland 
ecosystems in the world, comprising an area between 16,000 to 
22.000 km?. The whole catchment area of the river plus the 
delta together known as the Okavango River Basin (Figure 1), 
comprises an area of nearly 429,400 km”. The Okavango sub- 
basin is part of the larger Makgadikgadi basin, which covers an 
area of approximately 725,300 km” (Ashton and Neal, 2003). 
Figure 1. Location of the Okavango River Basin and the Delta, 
Southern Africa 
different sensors with varied spatial and spectral resolutions 
have the potential for better extraction and classification of 
features. The principal aim of this study is to examine the effect 
of fusion of remotely sensed images from different sensors for 
extracting and classifying wetland features. Wetland features 
The delta exhibits great variations in areal extent, spatial ve. 3 rtl 
are represented as a collection of pixels in an image. For the 
complexity and temporal dynamics. Due to large size and 
inaccessibility of the delta, satellite images provide the only 
296 
  
  
Internation 
purpose of 
pixels or g 
The Okava 
complex ar 
demand fr 
Botswana) 
complex 1 
important 1 
demand fo 
and associ: 
The Okava 
of the Kal 
(Bailey, 19 
great Mak 
sands abou 
large alluvi 
a series of 
channels, l: 
separate ag 
has a very 
distance of 
every 4 km 
* Mohen 
  
Figure 2. T 
(5,4,2 comb 
The Delta i 
wide variati 
swamps anc 
swamp cov 
swamp vari 
the permane 
the seasonal 
al., 1998).