Automatic classification of land cover with high resolution data of the Rio de Janeiro City Brazil

Abstract

The city of Rio de Janeiro, with more than 5 million inhabitants, has two big mountains in its centre with natural Atlantic rain forest. The growth of the city around these two mountains puts pressure on this forest. This endangers the remaining Atlantic forest (one of the most endangered forest types of the world) and also causes landslides and mud flows. The city administration carried out a land-cover forest classification with visual interpretation using merged LANDSAT and SPOT data. This work produced a compatible thematic map in the scale 1:50,000. It took about 36 months of intensive work and high costs to produce these maps. The scale of these maps permit to have a global vision of the land change cover but unfortunately do not correspond with the geographic information system of the city, which works with a scale of 1:10,000. The city searched for options to make this work automatically and quickly to get information for planning and to propose solutions. In order to solve this problem high resolution satellite data and automatic classification of land-cover classes are needed. Consequently, images as IKONOS (multispectral with four bands and 4 meters) need to be used to produce a classification, with a scale corresponding to the GIS of the city. The automatic classification of land-cover classes provides a relatively rapid classification. Pixel based classification with high resolution data show some problems because the level of information in the data produce a lot of incorrect classified pixels. The solution to perform this classification uses the new approach that makes one "pre-classification" (classification), which transforms the pixel information in objects as well as the feature in the vector representation.