Abstract

Landsat 5 TM and Landsat 7 ETM Imagery. Applications to Supervised and non-Supervised Classifications in the Eastern Region of Chaco Province, Argentina (South America)

Roberto Torra
Departamento de Geociencias, Facultad de Ingenieria
Universidad Nacional del Nordeste, Argentina
roberto_torra@arnet.com.ar


Abstract :
This paper deals with satellite images. They were employed in supervised and non-supervised digital automatized classifications. Several algorithms were used as well as standard programs, commonly related to many digital image processing tasks.

The study area is located at the east side of the intertropical latitude of Chaco Province, which was early deforested on XIX and XX centuries. The study area embraces about of 7.000 km2. Large sectors of soya and cotton commodities are principal resources for men.

Training sites were defined and differentiated. Ten classes were established. The images employed were Landsat 5-7 226/79, 227/79. They were acquired at the wet period (summer) in order to improve contrast between landscape elements. A previous field recognition of the area resulted of a great utility during the computational tasks. The algorithms used were K-means, IsoData (non-supervised), Maximum Likelihood, Improved Maximum Lakelihood and Neural Net Analysis (supervised). Also, we used automatic post-processes algorithms.

Special discrimination was addressed to cattle raising activities and forest units as well as deforested sectors. An incipient erosion processes of aridification and soil degradation also was determined. Water bodies as well as the incipient reorganization drainage system (3,000 to 6,000 years B.P.) were clearly remarked. Mangrove vegetations that follows joint to small rivers were outstandingly outlined. Main elements of geomorphology were appropriately appointed. Special discrimination was performed into floodable areas related to rainy periods. Forest units were perfectly synthetized. The final chartography were plotted and compared with false coloured compositions 4/5/3.
  1. The method results are of great utility to peer mapping land use and rural land management.
  2. The scales that varying about of 1:75.000 up to 1:000.000 -according to the pixelar spatial resolution of the Landsat sensors- showed be well appropriate for use in supervised and non-supervised classifications at east intertropical Chaco province.
  3. All the algorithms used showed adequate resolution and classes discrimination.
  4. The neural net analysis algorithm gave the best results.