The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
ground surface; b) middle - medium rough ground surface; c) 
bottom - rough ground surface (see text). 
Dataset 
Ground Roughness 
Appr. Two 
Appr. 
Three 
20m 
Pine 
Smooth 
Median rough 
Rough 
19.7±2.2m 
17.8±1.9m 
9.9±2.3m 
19.4+2.4m 
18.8±2.0m 
10.0±2.0m 
10m 
Deciduous 
Smooth 
Median rough 
Rough 
9.6+0.9m 
9.0+0.9m 
5.6+1.2m 
9.6+0.8m 
9.1±1.0m 
5.9+0.8m 
Table 2. Estimated tree height from PolSARproSim simulated 
datasets 
3.2 Results from INDREX-II Data 
In November 2004, DLR conducted an ESA-sponsored airborne 
radar campaign over Indonesian tropical forest, called 
INDREX-II (Hjansek, et. al., 2005a). Data from that campaign 
has subsequently been made available by ESA. We selected one 
of the test sites called Mawas-E as our study area because of its 
flat topography and the availability of measured tree height 
samples. The area is a tropical peat swamp forest located in 
Kalimantan, Indonesia. The L-/P-Band InSAR data were 
acquired by DLR’s E-SAR system in a quad-pol, repeat-pass 
mode along with single-pass X-Band data. 
Figure 4 shows a subset of the X-Band amplitude image 
acquired in the same campaign over the Mawas-E test site. The 
area is flat and there is a clear transition from bare area to forest 
area moving from west to east (from left to right in Figure 4). 
The left part of the image is a bare or low vegetated area, where 
we can expect the X-Band DSM (Digital Surface Model) is 
close to the ground thus we can use this area to normalize or 
validate the DEMs derived from L- or P-Band data. 
Figure 4 Selected ROI on X-Band image: The two red short 
lines on the right hand side of the image are two tree transects. 
The tree height measurements were carried out concurrently 
with the data acquisition campaign (see Figure 4, the two red 
short lines in the right hand side of the image). According to the 
ground measurement, there are more than half of the measured 
trees that are short, thin and branchless, which form a very 
dense understory with a height ranging from 2m to 8m and a 
spacing around lm. Trees with branches are about 17m high on 
average. Figure 5 also shows two of the ground photos taken in 
the forest near the tree transect. The photos show that the taller 
trees have branches and large canopy crowns while the 
understory consists of branchless, thin, and relatively shorter 
trees. 
To quantify the average forest height, the two tree measurement 
sites, each 100 meters long by 10 meters wide, are divided into 
non-overlapping lOmxlOm subplots, resulting in 10 subplots for 
north site and 10 for south site. The highest measured tree 
height in each subplot is defined as h 100 (Mette, et. al., 2004). In 
addition, as a second check in the case of h /0 o not being 
representative of the canopy, the second highest measured tree 
height in each subplots is also determined and called hj 0 o■ The 
forest in the test site can be described as “undisturbed peat swap 
forest” and is relatively dense, high, and uniform (Hajnsek and 
Hoekman, 2006). Therefore, the canopy height seen by a radar 
system over each subplot can be represented by either h t0 o or 
hjoo- Figure 6 gives h m and hj 00 as measured from tree 
transects. The average h m value is about 23m, while the 
average hj 0 o value is about 20m. The latter values show a 
smaller spread in heights. 
Figure 5 Ground photos of the forest 
Subplots (10mX10m) 
Figure 6 h/oo height (a) and hj 0 o height (b) as measured from 
tree transects (see text). 
The phase optimization and three tree height estimation 
approaches were applied to L-Band 10m baseline dataset. 
Figure 7 shows the outputs from the three approaches. The 
statistics of the estimated tree height around the tree transect 
area is given in Table 3.