004 
of 
cof 
2 
Is”; 
ing 
port 
one 
we 
our 
TROPICAL FOREST BIOMASS MAPPING F ROM DUAL F REQUENCY SAR 
INTERFEROMETRY (X AND P- BANDS) 
J. R. Santos * , T. Neeff ^ |. y. Dutra*, L. S. Araujo *, F. F. Gama“, M. A. T. Elmiro “ 
" INPE - National Institute for Space Research . Av. dos Astronautas, 1758 Caixa Postal 515 
12.227-010. .. Sáo José dos Campos (SP), Brasil roberto @ltid.inpe.br 
e University of Freiburg, Tennenbacher Straße 4, 79085 Freiburg Brsg.; Germany, üllneeffG fullbriehtweb.ore 
* UFMG-Federal University of Minas Gerais, Av. António Carlos, 6627, B.Horizonte, mtimbo (? ufmg.br 
Commission VII, WG VII/3 
KEY WORDS: Interferometer, SAR, Forestry, Land cover, Inventory, Mapping, Modelling. 
ABSTRACT: 
Radar sensors operating with different wavelengths and polarizations have been widely used for large-scale forest mapping and 
monitoring. The interferometric phase obtained by microwave sensors contains additional information on the three-dimensional structure 
of the scattering targets in the image. An experiment was performed in the Brazilian Amazon (Tapajós National Forest and surroundings) 
to provide airborne SAR data at X- and P- bands over tropical rain forest. In a first step of the presented research the regular radar 
backscatter results are joined with an interferometric height model to establish a statistical relationship to forest biomass (primary and 
secondary vegetation). Subsequently, that model is applied for generation of a thematic land cover map. Backscattering of P-band waves 
mainly occurs on the ground surface, and can be used for interferometric generation of a Digital Elevation Model. The X-band is 
reflected by dossel, and thus relates to the forest canopy in a Digital Surface Model. The difference between both models has been shown 
to represent height of vegetation. Care was taken in establishing statistical models that relate dendrometric parameters from forest types 
to both P-band backscatter and interferometric height. A best biomass model [biomass = 44.965 + 13.887 x h ; + 10.556 x O?^nmi ] was 
established after comprehensive testing of a range of specific allometric equations to achieve statistically high precision in biomass 
prediction. A segmentation algorithm (hierarchical région growth) was applied to the remote sensing dataset to provide means for 
application of the biomass model to homogeneous landscape units with similar biophysical characteristics and site histories. A final 
mapping result displays forest biomass, and accounts for different successional stages and primary forest in intervals. 
1. INTRODUCTION highway BR - 167 Cuiabá - Santarém. The yearly rainfall varies 
between 1,750 and 2,000mm. The distrofic yellow latossol 
With the advance of remote sensing technology, SAR data are (oxisol) soil type predominates in two textural classes: clay and 
available to supply and/or to complement the information level medium clay. These are normally deep soils, found over hilly to 
obtained by optical Sensors, referring to monitoring land cover strong hilly terrain, covered by dense forest of lowlands and 
in Amazonia region. For certain specific studies, such as e.g. the sub-montane. Human Occupation is related mostly to 
estimation of biophysical parameters of vegetation cover, radar subsistence agriculture (rice, cassava, maize, beans, pepper) and 
data present limitations inherent to the frequencies used specially to huge areas for extensive cattle raising. 
(Wegmüller and Werner, 1997). In this context, a scientific 
experiment was performed in the Brazilian Tapajós region, to 
provide airborne interferometric X and P-band SAR data over 
tropical rainforest. The objective of this study is to analyze and 2.2 Remote Sensing and Ground Data 
map the biomass variation of primary forest and secondary 
succession using dual frequency SAR interferometry. This The airborne SAR images were obtained by a system developed 
interferometric approach demonstrate a potential to improve by AeroSensing RadarSysteme GmbH Company, that acquires 
knowledge of forest Structure and the estimation of its’ both P (polarimetric) and X- band interferometric data. This radar 
biophysical parameters (Cloude et al., 2000; Balzter, 2001). System provides P band (A= 72 cm), obtained with a middle 
frequency of 415 MHz, band-width of 70 MHz, depression angle 
2. MATERIALS AND METHODS of 435° range resolution of 1.5 m and azimuth resolution of ().7 m, 
for 1 look slant range image. The X band imagery have a pixel up 
2.1 Test-Site Description to 0.5 m of ground range resolution, with HH polarization, middle 
frequency of 9.6 GHz and 400 MHz band-width. 
The area under study is located at the lower Tapajés River ; 
region (Pará State, Brazil), between W 24^ 53! 10 55? 06' and The radar tracks were radiometrically corrected according to the 
e , ’ 7 = NS * ~ . 
S 03°03 to 03°12 close to the Säo Jorge village, along antenna pattern using a function based on homogeneous extended 
areas; and afterwards, the polarimetric calibration was done for