International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
each polarization (slant range mode), based on the 8 corner 
reflectors (placed in the field along the flight strips using 
differential GPS measurement). In this study only HH 
polarization at X and P band was analyzed. The longer 
wavelenghts (P-band) pass through the forest cover and were used 
to generated a Digital Elevation Model (DEM). The short 
wavelenghts (X band) are reflected from the top of forest 
canopies and are used to generate a Digital Surface Model 
(DSM). Both models are calibrated and have a spatial resolution 
of 2.5 m. The difference between the DSM and DEM is 
considered to represent height of vegetation cover, thus providing 
in form of an image a so-called “Interferometric Height Model” 
(DHM). This image was afterwards segmented using an 
algorithm based on hierarchical region growing. This algorithm 
start out degrading the image to pyramidical levels with 
increasing spatial resolution and performs an initial region 
growing based on a mutual nearest neighbor criterion of grey 
level values between pixels (Souza Jr. et al. 2003). On each 
sucessive level, firstly a border correction by edge detection is 
applied. Subsequently, an F-test for homogeneity provides a 
criterion for decision on whether to split polygons and grow 
regions again. The procedure works successively at Six 
pyramidical levels. In the final results, polygons below 2,500 m? 
are eliminate in order to provide a degree of segmentation that 
would sensibly reflect a minimum landscape unit in the study 
area. After the image segmentation, the interferometric height 
value from each polygon is extracted, which corresponds to the 
localization of each sample plot properly georeferenced, where 
the forest inventory was done. 
Using this procedure, interferometric height values for both 
primary and secondary forest areas were obtained, which are need 
to setup a biomass prediction model and consequently allow the 
thematic mapping, based on the adjusted equation. 
During the field survey, biometric features (DBH » 5 cm, total 
tree height) for the primary forest and secondary sucession, 
including the botanical identification were collected in several 
samples, with measurements in plots of 2,500 and 1,000 m° 
respectively (Santos et al, 2003). Secondary succession 
inventories were collected to represent three stages of 
succession (at initial, intermediate and advanced level). The 
stratification considers both the age of the natural regrowth, 
certain structural characteristics, and floristic composition. 
These data allow for the computation of structural forest 
parameters for diameter, total basal area and various height 
measures (mean height, predominant height, i.e. the arithmetic 
mean of tree height of the upper vegetation stratum), and 
biomass was derived by allometric equations from Nelson et al. 
(1999) and Chambers et al. ( 2001). 
3. RESULTS AND DISCUSSION 
3.1 Interferometric Height versus Ground Data 
Interferometric height (h ;,,) for primary forest is a measure of 
the height of some predominant collective of the tallest trees, 
excluding the under-story. The interferometric height for 
secondary growth is a measure of the height of some sub- 
emergent individuals that excludes the tallest trees (upper 
stratum), and only consists in the smaller ones. Therefore 
average height (considering DBH>5cm) was used for secondary 
succession and predominant height (mean height of the tallest 
200 individuals per hectare) was used for primary forest, 
according to Neeff et al. (2004). 
A linear function was fitted to the combination of mean and 
predominant forest heights as derived from ground data versus 
interferometric height as derived from remote sensing data 
(Figure 1). The coefficient of determination attained a value of 
R? = 0.87 on this regression analysis. Thus, the interferometric 
height is a valid measure of forest stand height from a remote 
sensing dataset. 
  
has f (m) 
  
  
  
  
ha (m) 
Figure 1. Scatterplot of forest height from ground 
measurements versus interferometric height derived from X and 
P-band SAR data. Source: Neeff et al. (2004). 
3.2 Interferometric SAR Data and Biomass 
Initially, the biomass stocks variable was logarithmically related 
to P-band backscatter data in HH polarization, showing a 
coefficient of determination of R* = 0.66 according to Santos et 
al. (2003). Using the same dataset at Tapajós region, 
considering a polynomial function, these authors reported that 
the best fit was produced using the HH polarization (R? 2 0.77) 
when comparing to HV and VV polarization. Afterwards, the 
interferometric height, i.e. the difference between DSM and 
DEM, was taken to represent vegetation height and was 
individually related to biomass, whose relation is linear and 
yields a determination coefficient of R? 20.87. 
Finally, integrating the variables mentioned, both P-band HH 
backscatter (6° yy) and interferometric height (h;,) were fitted 
to biomass levels, and the precision was slightly increased to R? 
= 0.89 for the calculation of standing biomass in primary and 
secondary forests. The intercept is not significantly different 
from 0 at level œ = 0.05, however, there is no reason to drop the 
intercept from the model. This biomass model (Neeff et al., 
2004) was established for all types of forest occurring in this 
area; ranging from initial regrowth with biomass levels below 
than 5 t/ha to primary forest with biomass levels up to ca. 350 
t/ha: 
biomass = 44.965 + 13.887 x h ;,, + 10.556 x 0° (1) 
1134 
m: 
dis 
CO