Avaliação do uso do solo através de imagens Landsat : uma avaliação multitemporal a partir de um algoritmo classificador ideal

Abstract

Due to the necessity for managing of the natural resources, land use analysis has became an important task to assist decision-makers. The objective of the present work is to compare three different algorithms for image classification applied to land use evaluation. The methodology was applied to the Marmeleiro River and the Vitorino River watersheds located in the southwestern part of Paraná State, Brazil. A database was created in SPRING software for the management, manipulation and generation of geographic data based on the Army Topographic Maps. A morphometric characterization of both watersheds, as well as streams order, climate definition, geomorphology, geology, soil type, elevation and slope were carried out. Sampling points to characterize the land use were acquired by means of field surveys. The Kappa index was used to determine the performance of three algorithms: the Maximum Likelihood (MAXVER), which is a supervised algorithm, the Isoseg and Bhattacharya, which are non-supervised algorithms. Satellite images were acquired from the Landsat missions. By comparing the performance of the algorithms, for the Marmeleiro River watershed, the best classifier was the MAXVER with index of 0.7, whereas for the Vitorino River watershed, the best algorithm was the Isoseg with index of 0.92. The MAXVER alorithm also showed a good performance in the Vitorino River watershed, with an index of 0.90. The multitemporal analysis was performed by the algorithms with the highest performance indexes, thereby, identifying an increase in the arboreal vegetation from 1985 to 2000, followed by a slight fall and a new increase until 2010. Through the field surveys it was observed that reforestation areas are cultivated very close to native vegetation areas, therefore this might be a reason for the increase in the arboreal vegetation area in the watershed.