Next: Key Assumptions and
Up: Vegetation Studies of the
Previous: The SASS Data
In generating their regional maps, Kennett and Li used a ``binning'' technique
for mapping 
and . In this technique, the study area was divided into
rectangular ``bins'' (
or 
on a side). Each 
measurement was assigned to the bin in which the cell center fell. 
and 
were estimated for each bin using linear regression. The ``binning'' technique
must use resolution ``bins'' no smaller than the largest measurement cell,
limiting the resolution to no better than 50 km.
Unfortunately, the 50 km intrinsic resolution of the scatterometer is too coarse
for many studies. To ameliorate this problem, we developed a new technique
for generating enhanced resolution images of 
and 
from the low resolution
scatterometer measurements of . Our technique is based primarily on
ground-based signal processing. The method takes advantage of the areal overlap
in the backscatter measurements taken at different times of a given region, and,
using an indirect measurement (i.e., reconstruction) formulation, extends the
effective resolution. However, the improved
resolution is not without cost: the noise in the images of 
and 
increases
as we attempt to improve the resolution. Hence, a tradeoff must be made between
resolution and the noise level. In addition, certain conditions on the data are
required and various assumptions must be made. These will be described below.
Because of the spatial sampling characteristics of the SASS measurements, we must
use data from multiple orbits to obtain sufficient measurement overlap to
apply our reconstruction technique
. To accurately estimate
and , we must assume that the target does not change appreciably during the
imaging time interval. We define the ``imaging time interval'' as the length
of time required to accumulate all the measurements used to reconstruct an
enhanced resolution image.
The ultimate obtainable resolution with our technique is a
function of the measurement cell overlap and the noise in the measurements.
Arbitrarily reducing the size of the resolution elements will not increase the
effective resolution of the resulting estimate since, as the resolution element
size is decreased, the noise in the estimates increases. To minimize the
estimate noise for a given resolution element size, the number of measurement
cells should be maximized. In turn, however, increasing the number of
measurements requires increasing the imaging time interval. Thus, the
resolution element size must be a tradeoff between the imaging time interval and
the image estimate noise level.
Our imaging technique provides estimates of 
and 
for each element of a
rectilinear grid of small resolution elements from the lower resolution
measurements. In this paper, the iterative SIRF (scatterometer image
reconstruction with filtering at each stage) algorithm was used to generate
enhanced resolution images of 
and 
over the Amazon region of South America
from three months of SASS data. For the images shown in this paper, the
resolution element (pixel) size is 
(approximately
km). When applied to SASS, this represents about the best
resolution enhancement possible. As described in LHW, modifications to future
scatterometers could result in an enhanced resolution of 1-2 km. These
resolutions are similar to the resolution of local area coverage (LAC) from the
Advanced Visible High Resolution Radiometer.
Although the enhanced images are only medium-scale, our results demonstrate
that they can be successfully applied to studies of tropical vegetation. Thus,
this technique can be used to augment the data from existing and planned
instruments, leading to their more effective use. The wider
coverage of the enhanced scatterometer data may be more effective for large-scale
monitoring than higher resolution radar sensors such as SARs because they are
lower cost. Using the technique, scatterometer data can be used to extend the
results of focused studies by high resolution sensors to much larger
(continental) areas. While we have limited ourselves to using
Ku-band SASS data in this paper, the technique can also be applied to the C-band
ERS-1 scatterometer. A later paper will present ERS-1 results.
Next: Key Assumptions and
Up: Vegetation Studies of the
Previous: The SASS Data