343
FOREST HEIGHT ESTIMATION FROM INDREX-II L-BAND POLARIMETRI INSAR
DATA
Q. Zhang 2 ’ *, J.B. Mercer 2 , S.R. Cloude b
2 Intermap Technologies Corp., #1200, 555 - 4 th Avenue SW, Calgary, AB, Canada T2P 3E7 - (qzhang,
bmercer)@intermap.com
b AEL Consultants, 26 Westfield Avenue, Cupar, Fife KYI5 5AA, Scotland, UK - aelc@mac.com
Commission I, WG 1/2
KEY WORDS: Forestry, SAR, Mapping, Vegetation, Estimation, DEM
ABSTRACT:
This paper presents some results of forest canopy height estimation from L-Band polarimetric InSAR data. Three approaches have
been tested using a set of PolSARproSim simulated data as well as real data from the INDREX-II campaign. The approaches are: 1)
DEM differencing, 2) 2-D search, and 3) Combined. The results show that the DEM differencing approach tends to underestimate
the forest height by one third, while the other two approaches can achieve about 90% accuracy when there is sufficient ground return.
1. INTRODUCTION
Forest canopy height is one of the important parameters that can
be utilized for purposes of indirect forest biomass estimation
allometry [Mette, et al., 2004]. Recent advancement in
Polarimetric SAR Interferometry (PolInSAR) [Cloude and
Papathanassiou, 1998; Papathanassiou and Cloude, 2001;
Cloude and Papathanassiou, 2003; Cloude, 2006] has made it
possible to estimate the forest height through the use of the
Random Volume over Ground (RVoG) model [Treuhaft and
Siqueira, 2000; Papathanassiou and Cloude, 2001]. In this paper
we will address the problem of tree height estimation using both
simulated data [Williams, 2006] and real data from the
INDREX-II campaign [Hjansek and Hoekman, 2006].
Yv =
(2)
where K z is the vertical wave number calculated from the
incidence angle (6) , the difference of two incidence angles
from two antennas (Aff) and the wavelength (A,) of the radar
system as in Equation (3).
K
z
4M 6
A sin 6
radians/meter
(3)
2. METHODOLOGIES
According to the RVoG scattering model, the complex
interferometric coherence y , can be written as [Papathanassiou
and Cloude, 2001]:
f« = exp 0)
1 + m(w)
where (f> {) is the phase related to the ground topography, m is
the effective ground-to-volume amplitude ratio (accounting for
the attenuation through the volume) and W represents the
polarization state. y v denotes the complex coherence for the
volume alone (excluding the ground component), and is a
function of the extinction coefficient a for the random volume
and its thickness h v as expressed in Equation (2).
The key point of interest for this application is the assumption
that m is polarization dependent while y v is not. Manipulating
Equation (1), it can be seen that the complex coherence values
will lie upon a straight line as a function of m within the unit
circle on the complex plane [Cloude and Papathanassiou, 2003].
In particular, for large m, the straight line intersects the unit
circle and the associated phase at this point relates directly to
the desired ground elevation. In the limit of no ground
component (m=0), the observed coherence is given by the
volume coherence y v rotated through (f)^ . A main objective of
much of PolInSAR effort has been to develop robust methods to
estimate h v through an inversion process. In this work, we will
be comparing three of these approaches.
In RVoG model inversion, the ground phase (f) Q is usually
estimated first. This can be achieved by calculating the line-
circle intersection on the complex plane [Cloude and
Papathanassiou, 2003]. The straight line can be either fitted
from a set of observed complex coherences (e.g., lexicographic
coherences, Pauli decomposition coherences, and magnitude
optimized coherences) or formed by the two ends of the
estimated coherence region resulting from phase optimization
Corresponding author.
