mainly
natural
"sent a
ool for
images
altitude
of the
sful for
of the
of the
| under
LAND USE EVALUATION IN INTENSIVE AGRICULTURE
AREA OF SOUTHEASTERN BRAZIL
Mónica Takako Shimabukuro
Alvaro P. Crosta
J. V. Rocha
C.A. de M. Scaramuzza
Universidade Estadual de Campinas - UNICAMP
Instituto de Biologia - Depto. de Botánica
Rua Maestro Cardim 1218/163
01323-001 - Säo Paulo - SP
ISPRS Commission VII / Working Group 3
ABSTRACT
The State of Sáo Paulo, in Southeastern Brazil has one of the most intensive agriculture, husbandry and
forestry of the country. These activities have brought about increasing environmental impacts in many
different scales of natural resources management, whose mitigation is complex and expensive. The
measurement of the agribusiness sustainability involves monitoring the spatial and temporal dynamic of
land use with remote sensing tools and methods. Since these environmental consequences of intensive
agricultural production systems are serious and growing fast, it is essential that we get more and more
automated and reliable land cover information by digital image processing systems. The aim of this work
is to analyze the digital image processing application to evaluation of land use in Brotas, State of Säo
Paulo, chosen due to its great landscape diversity and dynamic. Two main aspects of the used software
(ER-MAPPER) were analysed: 1. the user-friendliness and robustness of graphical interface and
algorithms; 2. the classification accuracy of 4 classifiers, namely ISOCLASS, minimum-distance-to-
means, maximum likelihood and Mahalanobis distance. A grid with 2124 cell of 600 by 600 meters was
overlaid on the LANDSAT images. The ground truth was checked in 46 of these points to ensure a 2,796
sampling of the whole study area. By way of assessing the image classification accuracy, it is being used
the error matrix. The preliminary results show a differential performance of each classifier to label each
one of the 15 land cover types, e.g., the maximum likelihood distinguished well sugar-cane areas and
very badly eucalyptus woods, but the otherwise minimum-distance mistook sugarcane for orange grove.
051