POTENCIALITIES OF USING TEXTURE MEASURES FOR MAPPING LAND USE DYNAMICS BASED ON RADAR 
IMAGERY 
Milton Cezar Ribeiro '? 
Diogenes Salas Alves ! 
Joào Vianei Soares ! 
Corina C. F. Yanasse ! 
Fernando Mitsuo li ! 
| INPE--Instituto Nacional de Pesquisas Espaciais 
Caixa Postal 515, 12201- 097 Sao José dos Campos, SP, Brazil 
{milton,dalves,corina, fii} @dpi.inpe.br, {vianei}@ltid.inpe.br 
2 University of Säo Paulo, Forest Science Department, 
Av. Padua Dias 11, 13418-900 — Piracicaba — Säo Paulo — Brazil 
Commission VII, Working Group 4 
KEY WORDS: SAR, window size, texture measurements, Amazonia, dynamic change 
ABSTRACT 
This paper presents a methodology based on the selection of texture measures obtained from JERS-1 and Radarsat SAR data We 
studied 6 cover classes: recent deforestation, pasture and bare soil, young, intermediate and old secondary forest and mature forest.. 
Our interest is to find how texture measures can be used for land cover/land use classification. Were generated 140 Haralick's texture 
images combining SAR image and window sizes. The best window sizes were 11x11 and 15x15 pixels for JERS-1 and Radasat data, 
respectively. For recent deforestation or pasture cover classes the best discrimination were found when using entropy and mean of 
sum vector, for JERS-1 data, and mean, contrast and variance of difference vector for Radarsat data. Our results show that JERS-1 
was better than Radarsat for recent deforestation discrimination. 
1. Introduction 
Tropical forest has been the focus for many studies, mainly due 
to their importance of climate, biogeochemical, and 
hydrological cycles. Large areas of the Brazilian Amazon have 
been clear cut, allowing agriculture and ranching establishment 
(INPE, 1998). Few years after pasture or crop fields 
establishment, the soil looses its fertility and the area is 
abandoned (Uhl et al., 1988). Then, a natural succession leads 
to forest regeneration. 
Because of the logistics and the size of the Amazon, satellite 
imagery is the only feasible way for forest monitoring. Optical 
data have been contributing on this studies (Tardin et al., 1980, 
Alves et al., 1992, Lucas et al., 1993, Moran et al, 1994). The 
presence of high cloud cover is an impediment to collect optical 
images, resulting in some uncovered regions. Synthethic 
Aperture Radar (SAR) data can be obtained independent of 
weather conditions. SAR data could also help to improve 
knowledge of forest regeneration. For instance, Luckman et al. 
(1997a) and Luckman et al. (1997b) tested the relationship 
between for both radar backscatter and texture measurements 
with regenerating stages. The results were shown to be good for 
deforestation assessment, as well as to identify some of 
secondary forest classes. 
This paper presents a methodology based on the selection of 
texture measures obtained from JERS-1 and Radarsat SAR data. 
We defined six cover classes as follows: RD: recent 
deforestation; PA: pasture or bare soil; Y2F: young secondary 
forest; I2F: intermediate secondary forest; O2F: old secondary 
forest and MF: mature forest Fourteen second-order texture 
images were generated for each SAR image, using different 
window sizes. Our goal was to investigate how texture measures 
can be used for land cover/land use classification. 
2. Study site 
The studied site is located in Rondónia State, western Brazilian 
Amazon, central coordinates S09*50'/W62?30" (Figure 1). The 
region presents a slightly undulated terrain, with an average 
rainfall of about 2,200mm, and mean temperature of 23,6°C 
(H.Schimiz, unpublished data), with a dry season from late 
April to late August. 
Deforestation has increased in Rondónia at very high rates in 
the past two/three decades (INPE, 1998; Tardin er al., 1980). 
Figure 1 illustrate a fish-bone pattern that characterize the local 
landscape. 
3. Field Survey 
During June/96 and September/96 we carried out two field 
surveys. At the first survey we performed aerial reconnaissance 
and took about a hundred 1:15,000-scale aerial photos. On 
September we visited an amount of interest targets by road, 
registering GPS coordinates and local cover information. We 
also made local people interview on both stages. 
392 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
=F (708. pM Q4 7$ A 
ne 
Ou 
im: