Comparison on the Performance of Selected Image Classification Techniques on Medium Resolution Data Towards Highland Bamboo Resource Mapping
No Thumbnail Available
Date
2008-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa Universty
Abstract
This study was conducted with the objective of comparing the performance of different
image classification techniques in discriminating Yushania alpina K. Shmann Lin.
(highland bamboo) from other land cover classes, with the use of medium resolution
images in three Woredas of the Bale highland. A subset of four Landsat 7 ETM+
images acquired on the 28th of November and 5th of February 2000, (path 167 / rows 55-
56 and path 168 / rows 55-56), covering the study area were used for the classification.
The image processing software employed was ENVI 4.3. After performing the
necessary pre-classification processes, Decision Tree and Maximum Likelihood
classification techniques were applied on the images and the results thereof were
examined using confusion matrix. Landsat 7 ETM+ image derived data sets such as
tasseled cap and as well the spectral values of the different bands of the images were
used as the decision rules in the Decision Tree Classifier. Analysis of the accuracy
assessment had revealed that it had an overall accuracy of 66.79% and a Kappa statistic
of agreement of about 0.56. The technique also showed user’s accuracy of 23.05% for
bamboo class. The other classification technique evaluated was maximum likelihood,
for which region of interest, based on the field data was delineated as a training guide in
the classification process. The Maximum likelihood classification showed an overall
accuracy of 87.75% and a Kappa value of about 0.84. User’s accuracy for bamboo
forest by this technique was 40.56%.
The potential highland bamboo growing areas of Ethiopia was mapped in order to
provide an indication of the suitable areas for the growth and future expansion of Y.
alpine, to guide the establishment of observation plots from which detailed ecological
requirements of the species can further be investigated and to direct future mapping of
the resource. The existing temperature, rainfall and altitudinal information were used to
identify the location of these areas and it covered an estimated 7,632,788ha representing
about 6.74% of the total country. Classification results showed that the existing
Highland bamboo land units found in the study area cover a total of 63,407.32ha. The
dominant corresponding soil types of these areas are Orthic Luvisols, Chromic Luvisols
and with not less correlation with Eutric Cambisols, Eutric Fluvisols and Eutric
Nitosols.
Keywords: Yushania alpina, Landsat 7 ETM+, remote sensing, Bale highland, bamboo
Description
Keywords
Yushania alpina, Landsat 7 ETM+, Remote sensing, Bale highland, Bamboo