LAND COVER CLASSIFICATION USING MULTI-SOURCE DATA FUSION OF 
ENVISAT-ASAR AND 1RS P6 LISS-III SATELLITE DATA - A CASE STUDY OVER 
TROPICAL MOIST DECIDUOUS FORESTED REGIONS OF KARNATAKA, INDIA 
Vyjayanthi Nizalapur 
Forestry and Ecology Division, LRG, RS & GIS-AA, National Remote Sensing Agency 
e-mail: n.wiayanthi@gmail.com 
KEY WORDS: Remote sensing, Land cover, Classification, Multisensor, Fusion, ASAR 
ABSTRACT: 
The present study addresses the potential of Synthetic Aperture Radar (SAR) data for land cover classification in parts of Dandeli 
forested regions, Karnataka, India, a FCC has been generated from coherence and backscattering co-efficient images of ENVISAT- 
ASAR data (HH polarizations) of 25 th Sep 2006 and 30 th Oct 2006. Similarly, ENVISAT-ASAR data (HH polarization) of 25 th Sep 
2006 along with IRS -P6 LISS-III of 11 th Jan 2005 were subjected to data fusion to generate a False Colored Composite (FCC) using 
multi-source Intensity Hue Saturation (IHS) fusion technique. The two FCCs were subjected to maximum-likelihood classification 
technique separately and classification accuracy from the two methods is computed. Results suggested that SAR data is capable of 
discriminating major land cover types viz., forests, agriculture, water bodies, barren/fallow, urban settlements. Composition of 
coherence information given by the ASAR along with backscatter images enhanced the delineation capabilities of SAR data. The 
over all classification accuracy and kappa coefficient of the False Colored Composite (FCC) were observed to be 78% and 0.75 
respectively. Further, an attempt has been made to discriminate different forest types by merging the optical LISS-III data with HH 
polarized ASAR data. The merged output has been found to better delineate the forest types apart from other land-cover classes and 
minimize the shadow effect. The overall classification accuracy and kappa coefficient of merged data was observed to be 82% and 
0.80 respectively. Results of the study suggest the significance of SAR data towards better classification of the land cover classes, 
when used in conjunction with optical RS data. 
1. INTRODUCTION: 
Land cover classification is a primary requirement for 
management and planning of various resources. Remote 
sensing techniques aided with ground information provide a 
reliable source of land cover classification in a cost and time- 
effective way. While the utility of optical data in land cover 
classification is well known, the potential utilization of 
information given by space-borne and airborne RADAR 
systems in land cover classification is successfully attempted in 
several studies. Remote sensing using SAR data is useful for 
mapping and monitoring land cover over tropical regions, 
where continuous cloud cover hinders optical imagery 
acquisition. It is important to develop tools to obtain useful 
thematic information from radar data in terms of landscape 
features and patterns (Simard et al. 2000). RADAR systems are 
of immense use in deriving forest structural parameters such as 
timber volume, basal area, dominant height, biomass, etc. at 
plot and stand-level (Manninen et al. 2005). 
Radar sensors operating with different wavelengths and 
polarizations can be widely used for large-scale land cover 
mapping and monitoring using backscatter coefficients of 
different polarizations. Further, interferometric coherence 
derived from complex SAR data provides valuable information, 
which can be used as an additional tool along with backscatter 
data to enhance the application potential of microwave remote 
sensing in discriminating land-cover classes. Apart from this, 
studies on combining microwave and optical data have also 
been done in various works, which suggest enhanced 
discrimination capabilities of merged data and the same can be 
attempted to yield better classified outputs. Image fusion 
techniques deal with integration of complementary and 
redundant information from multiple images to create a 
composite image that contains a better description of the scene 
(Saraf, 1999). Data fusion can reduce the uncertainty associated 
with the data acquired by different sensors or by same sensor 
with temporal variation. Further, the fusion techniques may 
improve interpretation capabilities with respect to subsequent 
tasks by using complementary information sources (Wen and 
Chen, 2004). 
The fusion of two data sets can be done in order to obtain one 
single data set with the qualities of both (Saraf, 1999). The low- 
resolution multispectral satellite imagery can be combined with 
the higher resolution radar imagery by fusion technique to 
improve the interpretability of the fused/merged image. The 
resultant data product has the advantages of high spatial 
resolution, structural information (from radar image), and 
spectral resolution (from optical and infrared bands). Thus, the 
merged image provides faster interpretation (Simone et al., 
2002), and can help in extracting more features (Wen and Chen, 
2004).Various image fusion techniques are available in 
published literature (Li et al., 2002; Tu et al., 2001). 
In the present study, we attempt to analyze the potential of SAR 
data in the discrimination of different land-cover classes 
2. STUDY AREA: 
The study area is in parts of Dandeli sub division, Uttara 
Kannada district, Western Ghats, Karnataka, India (Fig 1). 
Geographically it is a transitional zone between the younger 
rocks of Deccan trap formation and the older crystalline rocks 
of Archean shield of the Indian Peninsula. The soil on the 
331