The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
345 
document and algorithm specifications can be found in 
(Williams, 2006). PolSARproSim is capable of generating 
PolInSAR images with different wavelengths, imaging 
geometries, ground surface properties, forest types, etc. It is 
well suited for performing sensitivity analyses with respect to 
various parameters although it should be noted that is has yet to 
be validated over a large range of conditions. We have used the 
simulator to create a number of datasets with different input 
parameters. For illustration purpose, the cases of 20m Pine tree 
and 10m deciduous trees over a ground surface with three 
different smoothness levels are selected as our simulated 
datasets. Table 1 summarizes the simulation configurations. 
Note the surface property value is only a scalar number with 
“0” as smoothest and “10” as roughest. Similarly the ground 
moisture content value uses “0” for driest case and “10” for 
wettest case. The outputs of the simulator consist of co 
registered SLCs at different polarizations, a flat-earth phase file, 
and a vertical wave number (Xz) file. The Kz value for this 
simulated data is 0.13 Rad/m. 
Platform Altitude (m) 
3000 
Incidence Angle (°) 
45 
Baseline H / V (m) 
10/0 
Ground Surface Properties 
0,5, 10 
Ground Moisture Content 
4 
Trees Species 
Pine 1/ Deciduous 
Mean Tree Height (m) 
20/10 
Forest Stand Density (Stems/Ha) 
300/150 
Forest Stand Area (Ha) 
1 
Table 1. PolSARproSim simulation configuration 
The results of tree height estimation are shown in Figure 2 for 
20m pine tree and Figure 3 for 10m deciduous trees. In both 
cases, we have three outputs for the three types of ground 
surface. In Figure 2 and 3, the green and blue solid lines are the 
phase centers corresponding to the two ends of the coherence 
region; the red solid line is the estimated ground phase center, 
and the green and blue dashed lines are the height from: Ground 
plus Estimated h v using approach two and three respectively. 
These profiles are in the range direction with illumination from 
the left. There appears to be an edge effect - probably from 
layover - that causes the anomaly at the leading edge of the 
profiles. We ignore it in this work. 
The results show that when the ground surface is smooth, 
permitting strong dihedral return, (or in another words, when 
the ground contribution is large enough), the phase optimization 
algorithm works well in estimating the ground elevation for 
both types of trees. Subsequently, three tree height estimation 
algorithms gave a quite encouraging result. The DEM 
differencing approach estimated about 70% of the designed h v , 
while the other two estimated about 90% with very similar 
performance (See Table 2). 
When the ground contribution becomes less, the estimation of 
ground elevation becomes worse, as expected. This in turn 
reduces the tree height estimation accuracy. In case (c) where 
there is almost no ground return, the topographic phase estimate 
is biased and noisy. Without ‘a priori’ ground information in 
this case, the derived canopy height will be severely 
underestimated and noisy. 
simulated dataset for 20m pine tree: a) top - smooth ground 
surface; b) middle - medium rough ground surface; c) bottom - 
rough ground surface (See text). 
C*«m Prof-K- •- Center» 
Cross Profile Phgse 
simulated dataset for 10m deciduous tree: a) top - smooth