Bush Encroachment Mapping Using Supervised Classification and Spectral Mixture Analysis in Borana Rangelands: A Case Study in Yabello Woreda
No Thumbnail Available
Date
2009-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa Universty
Abstract
Bush encroachment has been among the major threats to the livelihood of Borana
pastoralists and their ecosystem. A study was conducted in Yabello district to map bush
encroachment using supervised classification and SMA (spectral mixture analysis).
Landsat Thematic Mapper and Enhanced Thematic Mapper Plus images that were taken in
January 1986 and 2003 respectively, 33333were used to assess the land-use/land-cover
dynamics, to determine rate of bush encroachment and to spatially locate bush encroached
areas. Six land-use/land-cover types were defined for supervised classification. They were
forest, woodland, bushland, grassland, crop/cultivated land and bareland. Unconstrained
linear SMA was run on the original DN image using woodland, bushland, Savannah
grassland, tilled soil and bare surface endmembers. Endmembers collected from the field
were tested for their purity by applying pixel purity index and minimum noise fraction
transform. Classification result and unmixing were assessed for their accuracy using
random samples collected from the field. Overall accuracy and Kappa coefficient for
supervised classification of the 2003 image were 82.66% and 0.778, respectively. For the
bush fraction image of 2003, they were 72.28% and 0.640, respectively. The study showed
that bushland, woodland and crop/cultivated land were increased and forest, grassland
and bareland were decreased between the study period. The rate of bush encroachment
was 2.54% per year. Bush fraction image showed that in 2003 about 20% of the land was
covered by bushy species that had > 40% canopy cover. In Borana rangelands, where
heterogeneous vegetation exists, spectral mixture analysis using woodland, bushland,
savannah grassland, bare surface, tilled soil and shade endmembers based on Landsat
images was found less promising to unmix the land-covers. However, it gave better
measure on the status of bush encroachment during 2003 than supervised classification
method. Among others, multiple endmember spectral mixture analysis based on high
spectral resolution images is recommended to better unmix the land-covers.
Key Words: Bush encroachment, supervised classification, SMA, endmember & Landsat
Description
Keywords
Bush encroachment, Supervised classification, SMA, Endmember & Landsat