Вы находитесь на странице: 1из 5

Microwave Radiometer (MWR) Beam-Pointing

Validation for the Aquarius/SAC-D MissionPaper


Bradley Clymer
1
, Catherine May
1
, Larry Schneider
1
, Felipe Madero
2
,
Martn Labanda
2
, Mara Marta Jacob
2
, W. Linwood Jones
1
, Life Fellow IEEE
1
Central Florida Remote Sensing Lab (CFRSL) University of Central Florida, Orlando FL
2
Comisin Nacional De Actividades Especiales (CONAE)


Abstract The Aquarius/SAC-D joint international science
mission, a collaborative effort between NASA and the Argentine
Space Agency, CONAE, was launched on June 10, 2011. The
purpose of the Aquarius/SAC-D mission is to provide
measurements of global sea surface salinity (SSS), which will be
used to understand climatic changes of the global water cycle and
how these changes influence the general ocean circulation.
This paper concerns CONAEs instrument, the Microwave
Radiometer (MWR), which is a three-channel Dicke radiometer
operating at 23.8 GHz (H-Pol) and 36.5 GHz (V- and H-Pol). The
instrument has two multi-beam parabolic reflector antennas in a
pushbroom configuration with eight beams per frequency
(looking both forward and aft). MWRs main purpose is to
provide measurements that are spatially collocated with those of
NASAs Aquarius radiometer/scatterometer. For this reason,
knowledge of the MWR antenna beam footprint geolocation is
crucial to mission success.
Results of an on-orbit validation of the MWR antenna beam
pointing - using calculated MWR instantaneous field of view
(IFOV) centers - are presented. This procedure compares
CONAE-calculated IFOV centers at land/water crossings with
high-resolution coastline maps. Also, in this paper, an algorithm
to quantify the MWR beam center geolocation error based upon
two-dimensional convolution between each beams gain pattern
and land-water transition is presented. These analysis procedures
have been validated with computer simulation and are applied to
on-orbit datasets that represent good land-water boundaries.
The goal of this research is to gain a better understanding of
satellite radiometer beam-pointing error and thereby to improve
the geolocation accuracy of past and future satellite microwave
radiometer missions.
KeywordsMicrowave Radiometry, Passive Microwave
Remote Sensing, Calibration, Geolocation
I. INTRODUCTION
The MicroWave Radiometer (MWR) is a remote sensor
developed by the Argentine Space Agency (CONAE) and
flown on the Aquarius/SAC-D satellite, a joint earth science
mission between NASA and CONAE. Aquarius/SAC-D is a
sun-synchronous polar orbiting satellite with the sensors
beams pointing away from the sun to prevent intercepted solar
radiation contamination, as shown in Figure 1. The main
purpose of this radiometer is to provide simultaneous and
collocated environmental parameter measurements to aid the
Aquarius sensors sea surface salinity retrievals. MWR
geolocation requirement is 5 km difference, between the
calculated and true latitude/longitude of the IFOV centers, for
forward and aft viewing beams (see Fig. 1).
The SAC-D satellite performs precise dynamic 3-axis
attitude control; however static biases, between the satellite
coordinate reference and the MWR instrument reference
coordinate system, can displace MWR surface IFOVs and
cause errors in geolocation. Also, three different spacecraft
attitude bias errors could occur, namely: roll, pitch, and yaw.
Each of these exhibits its own particular pattern as the
MWR IFOV traverses land/water boundaries. Roll bias errors
become evident when comparing ascending to descending
passes along an eastern or western continental coastline, while
pitch errors are evident along the northern or southern
coastlines. When perfectly geolocated, the MWR observed
boundaries will be aligned along the coast, for both ascending
and descending passes. However, when these errors exist, the
ascending and descending passes will separate nearly equal
distance on opposite sides of the coast. Yaw errors are
apparent when forward and aft beams are compared. Due to
the earths rotation, yaw effect has a larger magnitude at the
equator.
Yaw steering has been implemented to counteract this
effect, and ensure that the K- and Ka-band beams will cross a
boundary at the same location [1]. This MWR geolocation
analysis will assess the effectiveness of yaw steering at all
latitudes [2].
II. INSTRUMENT GEOMETRY
MWR it is a three channel, Dicke radiometer which
comprises the three radiometer receivers and two reflector
antennas; Ka-band, with eight beams (36.5 GHz, vertically
and horizontally polarized), looking forward and K-band, with
eight beams (23.8 GHz, horizontally polarized) looking aft, as
shown in Figure 1. As seen in the figure, the beam footprints
alternate along two conical arcs with incidence angles (52
and 58). Its instantaneous field of view (IFOV) varies from
(27 x 45 km) at 52 to (31 x 63 km) at 58, producing a swath
width of approximately 380 km which completely overlaps the
Aquarius swath. MWR takes measurements in a push-broom
fashion and has a beam sampling time of 1.92 seconds, which
leads to an along-track sampling distance of 13.1 km on the
earths surface [1].
Earth incidence angles (EIAs) of 52 and 58 were
calibrated pre-launch for the odd- and even-numbered beams,
respectively. These EIAs are the primary parameter being
validated in the current work.
Figure 1: Beam-pointing geometry
III. PREVIOUS WORK
Prior to current progress, CFRSL implemented the
difference-and-curve-fit approach illustrated here; this was
done with similar aim, namely to quantify geolocation errors
in MWR data. Similarly, previous work included 1-D
convolution of a modeled Gaussian beam with an ideal
coastline to validate the curve fitting being done. This is
illustrated in Figure 2, below.

Figure 2:1-D convolution of Gaussian beam and coast

IV. WINDSAT BRIGHTNESS TEMPERATURES
The geolocation analysis was performed using WindSat T
B

data for the period of October 29th-December 9th, 2012. The
WindSat instrument was selected because its wide swath
provides frequent overlap with MWR observations within a
time window of 45 minutes. The 2012 WindSat sensor data
records (SDR), obtained from the United States Naval
Research Laboratory (NRL) (http://www.nrl.navy.mil/
WindSat/), were used to produce brightness temperature
images of carefully selected sites for 23.8 GHz-H-pol, and 37
GHz V- & H-pol T
B
s, gridded on a 0.125 Lat/Long earth-
grid.
When evaluating separate tracts of the same geographical
area (e.g., Southern Australia) within any week of time, T
B

differences between 15-20 K were observed on a pixel by
pixel basis. Since it is not possible to have simultaneous
WindSat data for every pass, it was decided to use weekly
averages. Further, to minimize T
B
changes associated with the
diurnal cycle, the WindSat data were separated into ascending
and descending passes to correspond to the individual
collocations being considered.
It was important that the selected sites have contiguous
gridded brightness temperatures within the antenna pattern
surface footprint; therefore to fill missing pixels, average
brightness temperature resulting from values of the
surrounding cells, were inserted. An example of the WindSat
37 GHz V-pol T
B
image is shown in Figure 3.
Figure 3: Southern Australian coast with IFOV overlaid, and NaN
correction applied to compensate for missing WindSat T
B
s
V. GEOLOCATION APPROACH
A. Overview
Time series brightness temperatures change most rapidly
in transition between ocean and clearly defined, flat land. The
validation of beam-pointing herein is based upon this
supposition; specifically, that the maximum rate of change in
brightness temperatures indicates the point of ocean/land
transition, and that this is the 50% beam-fill point. Because of
this, the technique to estimate land/water crossing point begins
by selecting candidate data, which itself starts by selecting
sites that exhibit clear land/ocean boundaries. These sites
exhibit particular characteristics: straight line coast with the
absence of water (rivers, lakes, marshes, etc.) on the land side,
no occlusion by man-made structures (cities, large buildings),
and smooth water on the ocean side (low roughness/foam).
After sites which induce rapid T
B
changes in time series data
are found, data points which fall within range of those sites are
extracted and differenced, and a parabola is fit to the resulting
difference.
From the ensuing parabola, the vertex is extracted; this
vertex is taken to be the location of maximum T
B
change,
using MWR data observed T
B
(beam/pol) and calculated
instantaneous field of view (IFOV) boresight to calculate the
derivative of the T
B
time series, and determine the max slope
(dT
B
/dsample) location. Following that, the technique
determines the lat/long of this max-slope IFOV location, where
the land/ocean beam-fill is 50%. Subsequently, this location is
projected onto the site of interest, and the Euclidean distance
from the max-slope point to the coast is calculated to quantify
beam mispointing. This process is done for both recorded
brightness temperatures and simulated IFOVs occurring at the
locations of those data points, for validation. The error-
logging process is then repeated for all of the available,
appropriate data points, and statistics are aggregated, giving an
approximation of beam-pointing error. Appropriate data
points, in this context, are data points which cross the site in
question near its center, allowing calculation of T
B
changes
with high confidence.
A second technique uses gridded MWR T
B
data to
determine the max-slope location of a simulated Gaussian
beam pattern (convolved with a measured hi-resolution T
B

image, generated from temporally proximal WindSat data) to
calculate the observed/simulated difference from a 1km map,
and generate statistics for mispointing error.
B. Site Selection
Ideal sites for the application of the technique presented
herein are classified as such because of their similarity to a
knife-edge one-dimensional land/water crossing boundary
approximation, which has been shown to yield accurate
estimates of the crossing when convolved with a Gaussian
beam pattern, as shown in Figure 2.
To find sites for evaluation, first a latitude boundary is
determined heuristically. To avoid polar ice and to minimize
computation time by remaining largely over land, a boundary
from 60N to 50S was first imposed on candidate sites.
Next, a sample set of time series MWR brightness
temperature data for a single beam is differenced, and a
threshold to trigger investigation is set. A threshold of 15K in
T
B
difference was found to yield a good selection of candidate
sites and also, to reject outliers.
Following the discovery of a set of candidate sites,
investigation of the coastline must be performed, to eliminate
possible noise and ensure a good crossing angle between the
major axis of the beam footprint and the coast. Ideally, the
coastline is orthogonal to the footprint, ensuring fastest
transition and consequently the highest rate of T
B
change with
respect to time. In this case, investigation was done by linking
the list of sites which have met all prior criteria to Google
Earth through the KML interface, allowing rapid investigation
of controlled altitude and look angle. The aim of this visual
inspection is to filter out man-made structures which can vary
and distort brightness temperature readings in an unnatural
way; to avoid areas of high coastal foam such as inland lochs;
and to screen for inland swamps and poor coastal definition.
After a set of ideal sites is found for a single beam, a list is
aggregated and the process is repeated for all eight beams.
Because the footprints vary in location across-track by as
much as 50km, and few coasts retain the same angle across the
distance of multiple beams landfall, it was found that no sites
were suitable candidates for more than one beam. One such
site deemed usable is shown in Figure 4, on the southern coast
of Australia.
Once the list of sites is compiled on a per-beam basis, it is
then utilized in beam pointing evaluation moving forward.
Figure 5 shows sites determined to be ideal for beam 1.
C. Max Slope Location
To calculate the maximum change-of-T
B
slope point, it is
necessary to perform an interpolation. This is because time
series data points are sampled 13.1km apart on ground track,
which can result in 6.5km of quantization error.
First, the slopes of T
B
change in time must be extracted
from the data. A simple difference between datapoints is
performed, and that difference is assigned as a new
measurement, physically located midway between the
locations of the source data. Conceptually,
Slopc
+0.5
= I
B
i+1
-I
B


This offset can be seen in Figure 6. Note that the
datapoints of the slope curve are placed laterally between the
original ones.
The slope values are then fit to a least-squares parabola in
the neighborhood of the maximum value of slope, and the
vertex of that parabola is taken to be the maximum slope point
for the purpose of beam-fill calculations. An illustration of the
curve-fitting process is visible in Figure 6. In this example, the
Figure 4: Google Earth image of ideal site and 3dB IFOV

Figure 5: Sites selected for beam 1 of MWR mission
parabola maximum occurs at a location equivalent to sample
4.37.
D. Max Slope Projection on earth.
The placement of the parabola maximum is used to project
the location of the max-slope point on earth, and consequently
that on the gridded map. In the case of the example, in which
the vertex of the fit parabola occurred at 4.37, this means that
the location on the earth of the maximum slope is calculated to
be 37% of the way between datapoints 4 and 5; this distance,
along-track, is what is used to calculate the mispointing error.
E. Mispointing Distance Calculation
The mispointing error present in each assessment is simply
a vector norm of the distance between the projected point, and
the nearest coastal point, combined with a logical assessment
of whether the point leads or lags the actual crossing; if the
point follows the crossing, it is deemed to be a positive error,
and if it precedes the crossing, it is deemed a negative error.
Utilizing this, we may quantify the error as:
crror = norm_
pro]cctcJ
Iut
- coost
Iut
pro]cctcJ
Ion
- coost
Ion
] crror sign

where here we have omitted the conversion between lat/long
and earth-projected distance for clarity.
The errors for each projection are then aggregated to find a
final result.
F. Convolution for Location Validation
To validate the technique, a discrete convolution of an
ideal Gaussian beam pattern with the scene illustrated in
Figure 3 is performed. The convolution is performed as:
I
A
=
I
up
(0)F
n
(0)
360
q
2.5HPBw
0
F
n
(0)
360
q

2.5HPBw
0
.

This results in a T
B
for each location given, which is then
processed as above to calculate the maximum slope. In each
case, it was seen that the maximum slope point for the
convolved T
B
led to results similar to those of the measured
T
B
, which is taken to as an indication of the techniques
validity.
VI. RESULTS
An example of mispointing error for the Beam 1,
36.5GHz, Vertical and Horizontal Polarization is shown
below. The results indicate a difference between the V- and H-
pols, which is attributed to a correction done to the data
product prior to its use, and remains to be further investigated.
Figure 7: Errors in land-water crossing for beam one over the South-
Australia site
Each beam exhibits its own particular mispointing error;
automation of the evaluation process allows the aggregation of
statistics for each beam and polarization. Table 1 shows a
summary of mispointing error statistics, by beam and
polarization, for the sample period of October 29th-December
9th, 2012.
Table 1 Error for Each Beam, Ascending
37V 37H
Beam 1
Mean Error, km p = 2.88 p = -4.S1
Error Std. Dev, km o = 1.67 o = 2.61
Beam 2
Mean Error, km p = -8.22 p = -1u.28
Error Std. Dev, km o = 2.u6 o = S.7S
Beam 3
Mean Error, km p = -S.uu p = -.7u
Error Std. Dev, km o = 2.SS o = .8S
Beam 4
Mean Error, km p = -8.94 p = -S.24
Error Std. Dev, km o = 2.u8 o = 1.49
Beam 5
Mean Error, km p = 8.28 p = u.u2
Error Std. Dev, km o = 1.u7 o = 2.6u
Beam 6
Mean Error, km p = 7.86 p = 4.S9
Error Std. Dev, km o = 4.S7 o = S.86
Beam 7
Mean Error, km p = -1.S4 p = -S.1S
Error Std. Dev, km o = S.14 o = 1.4S
Beam 8
Mean Error, km p = 4.61 p = 9.S1
Error Std. Dev, km o = 1u.76 o = S.SS
Observe that the mean error for Beam 1, 37H is negative;
as the technique dictates, this is an indication that the
Figure 6: T
B
s and slopes, with fit parabola and its maximum
geolocation represents an early crossing, which is visible in
Figure 7.
VII. CONCLUSIONS AND FUTURE WORK
Initial indications show that this technique yields statistics
which may be used to correct mispointing errors in MWR data
product, beam-by-beam, prior to its delivery. The validation of
this constitutes a next step: apply the results a posteriori to the
datasets already recorded, evaluate the error results, and
observe the reduction of error inherent in geolocation.
Following such a large-scale evaluation, an iterative
approach in which the data are evaluated with mispointing
corrections repeatedly, to converge to a minimum error may
be justified to find the optimal correction, in a least-squares
sense. One possible result is that the mispointing changes over
time, in which case a trend line would need to be fit to the
data, and continually adjusted to evaluate new measurements.
Additionally, it is planned to evaluate the mispointing
error with respect to latitude to determine if spacecraft
position relative to the equator bears any impact upon the
mispointing error (or upon the efficacy of the technique), in
which case a model would then be constructed to correct for
mispointing based upon latitude.
ACKNOWLEDGMENT
This research was sponsored under a grant from NASA
Headquarters Earth Sciences Division.
REFERENCES
1]
S. S. Khan, "Simulation of Brightness Temperatures
for The Microwave Radiometer on the Aquarius/SAC-D
Mission," M.S. thesis," Orlando, FL, 2009.
2]
Catherine May and W. Linwood Jones, "Engineering
Evaluation of Multi-beam Satellite Antenna Boresight
Pointing using Land/Water Crossings," in Southeastcon,
2012 Proceedings of IEEE, Orlando, FL, 2012, pp. 1-5.

Вам также может понравиться