Satellite Imagery Classification with ER Mapper in the Mountains of Kyrgyz Republic.

Galina N. Fet, Geobiophysical Modeling Program, Marshall University, fet2@marshall.edu

James O. Brumfield, Geobiophysical Modeling Program, Marshall University, brumfiel@marshall.edu

The goal of this study is testing classification techniques of the Landsat7 ETM+ data with ER Mapper software. A representative altitudinal profile of the mountain habitats was studied within the Tien Shan Mountains of Kyrgyz Republic. It is a newly independent country with population 4.6 million located in the very center of Asia, north of China, within the mountain systems of Tien Shan and Pamiro-Alai which has fragile mountain ecosystems are notable for their biodiversity. The study area was Ala-Archa State National Park where different types of coniferous forests and semi-desert environments were described and georeferenced with GPS unit.

The National Park territory includes meadow-steppe, forest-meadow, subalpine and alpine vegetation zones. The following groups of vegetation associations and correlating spectral signatures were distinguished within the forests: juniper-shrub forest (Juniperus semiglobosa – Berberis integerrima – Rosa spinosissima), juniper-herbaceous forest (Juniperus semiglobosa – herbaceous vegetation), juniper-medow forest (Juniperus semiglobosa – Phlomis oreophila – Alchemilla sibirica), Spruce forests, formed by Schrenk’s spruce (Picea schrenkiana).

Higher resolution satellite imagery (USGS, Landsat 7 +ETM, 2002) was used for classification. The use of Image processing classification techniques such as ISOCLASS with ER MAPPER software proved to provide an opportunity for computer modeling and evaluation of the forest community types. Satellite imagery of 30 m resolution (red, green, blue, bands 3, 2, 1) was data fused with intensity layer (band 8) of 15 m resolution. Satellite imagery of high resolution allowed for spatial analysis that provided correlation between the species and habitats. Using image processing, we are able to outline compatible habitats in the forest communities. As a result, map representation of forest types distribution was obtained. The geobiophysical modeling of the environmental parameters with spatial analysis such as neighborhood and overlay functions and statistical analysis such as principle components and cluster analysis algorithms in an Image/GIS environment provided correlation needed for classification of the habitats.