TITLE

Teledetección y redes neuronales aplicadas al mapeo de coberturas del suelo de la cuenca del Matanza-Riachuelo, Buenos Aires, Argentina

AUTHOR(S)
Serial, Rodrigo Martín Becerra; Czibener, Daniela; Nabel, Paulina Esther
PUB. DATE
July 2009
SOURCE
Revista Geográfica;jul-dic2009, Issue 146, p125
SOURCE TYPE
Academic Journal
DOC. TYPE
Article
ABSTRACT
The Matanza-Riachuelo river basin is a particularly complex area in terms of both its current environmental situation and the diversity of land use and of materials covering its surface. By analyzing satellite images, it is possible to produce up-to-date base cartography, essential for the environmental management of the basin. We used a multitemporal series of Landsat 5 (TM) images for this work, which correspond to an annual cycle, as well as a set of derived bands, including NDVI, brightness, greenness, and humidity (tasseled cap transformation), and bands of scatter measurements of data in the annual cycle (e.g. variance), amongst others. After the bands had been selected, two neural networks (multilayer perceptrons) were trained using the Backpropagation algorithm. The first of these was used to classify the image into three main groups of ground cover: water, waterproofed, and vegetation. The second network was used to classify the vegetation into subclasses: evergreen forest, deciduous forest, grass, winter plant association, and summer plant association. The overall exactitude of the resulting map was 95.13%. The most common types of ground cover were grasses and waterproofed surfaces. The latter are mainly associated with urbanized zones and take up almost a quarter of the river basin area. Two types of vegetation related with agricultural activity are also found in large proportions. Along the urban-rural axis, the diversity of ground cover changes: there is minimal diversity (maximum uniformity) in urban areas, maximum heterogeneity in periurban areas, and low values in rural areas, without reaching the level of uniformity shown in the city. The results of this work encourage the use of satellite images and nonparametric classifiers like neural networks to produce cartography in complex environments such as that of the river basin in question. This cartography could be useful as a tool for the environmental management of said basin.
ACCESSION #
47371889

 

Related Articles

Share

Read the Article

Courtesy of THE LIBRARY OF VIRGINIA

Sorry, but this item is not currently available from your library.

Try another library?
Sign out of this library

Other Topics