The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
Appr. One
Appr. Two
Appr. Three
8.5+4.2m
18.7±6.1m
17.5±4.2m
Table 3. Estimated tree height from INDREX-II Data
& '
the ground roughness was increased such that dihedral return
was severely limited. In these instances, the calculated
coherence region becomes more circular and the resulting
ground phase projections become noisier. This in turn affects
the forest height estimation.
4. CONCLUSIONS
Three different approaches to forest height estimation have
been tested on both simulated and real L-Band PolInSAR data.
The results from the simulated data are quite encouraging. The
forest height estimated by the DEM differencing is normally
underestimated at a level of about 2/3 of the true forest height.
The results from the other two approaches are very similar to
each other and are within about 10% of the simulated canopy
height provided that ground return (dihedral bounce) is
adequate. It should be noted that with the simulated data set,
there are no negative effects caused by temporal decorrelation
or other factors associated with real data.
The results from real data are compared against ground
measurements. It is concluded that the estimated forest height
from 2-D search approach is quite close to the average h' W o,
roughly within 10% error. The combined approach is slightly
worse than the 2-D search approach, getting about 15% but with
a significant reduction in the computational load. On the other
hand, the height estimates from real dataset are considerably
noisier and less robust than for the simulated data. It is
suggested that the comparative noisiness and additional biases
are due to some temporal decorrelation of the repeat pass data,
and to additional attenuation of the signal in the lower part of
the canopy due to a dense understory that is not incorporated
into the RVOG model.
ACKNOWLEDGMENT
Figure 7 Estimated forest height from INDREX-II data using
three approaches: (a) DEM differencing; (b) 2-D search; (c)
Combined. The region inside the black box is used for statistics.
The two “+” signs are the locations of two tree transects.
From Figure 7 and Table 3, it is observed that, assuming the
average tree height in the region is 20m (i.e., < h'/otP), the
DEM differencing approach is estimating about 42% of <h' IO o->-
The 2-D search approach gives the best result, 95% and the
combined approach 88%. This is at a similar level to the results
from the simulated dataset. If the results, on the other hand, are
referenced to <h I0 d> rather than </?W> an additional 3 meter
underestimate is observed. At this point there is no strong
argument to prefer the one metric over the other as the more
representative canopy height.
However, the estimated tree height from INDREX-II L-Band
dataset is much noisier than those from simulated dataset
mainly due to the noisier estimation of ground elevation.
Although space precludes showing them here, the elliptical
regions estimated from this dataset are typically much smaller,
with lower coherence, rounder appearance and show an
apparent lack of dihedral bounce. Two major contributors to
this are likely: (1) temporal decorrelation, (2) the dense
understory of the forest, which may act to attenuate the dihedral
response of the larger trees. Consequently the phase separation
was reduced resulting in poorly shaped coherence regions and
inaccurate estimation of ground phase. This has been observed
in results from other PolSARproSim simulated datasets, where
We thank European Space Agency for providing the data,
Alberta Ingenuity Fund for partial financial support, and Dr.
Mark Williams for valuable comments regarding
PolSARproSim.
REFERENCE
Cloude, S.R., Papathanassiou, K.P., 1998. Polarimetric SAR
Interferometry, IEEE Transactions on Geoscience and Remote
Sensing, Vol. 36, No.5, pp.1551-1565.
Cloude, S.R., Papathanassiou, K.P., 2003. Three-stage inversion
process for polarimetric SAR interferometry, IEE Proceedings -
Radar Sonar Navigation, Vol. 150, No. 3, pp. 125-134.
Cloude, S.R., 2006. Polarization coherence tomography, Radio
Science, Vol. 41, RS4017, 2006.
Cloude, S.R., Zhang, Q., Mercer, B., 2007. Tropical Forest
Structure Estimation using L-Band Polarization Coherence
Tomography (PCT). Proceedings of 5th International
Symposium on Retrieval of Bio- and Geophysical Parameters
from SAR Data for Land Applications. September 25-28, 2007,
Bari, Italy.
Hjansek, I., Kugler, F., Papathanassiou, R., Scheiber, K., Horn,
R., Moreira, A., Hoekman, D., Davidson, M., Attema, E., 2005a.
INDREX II - Indonesian Airborne Radar Experiment
347