Ingen bildrubrik

1
Accurate Real-time Navigation of AVHRR data at
high latitudes
Adam Dybbroe, Pascal Brunel, Anne Marsouin,
and Anke Thoss Swedish Meteorological and
Hydrological institute (SMHI), S-60176
Norrköping, Sweden Centre de Météorologie
Spatiale, Météo-France, BP 147, 22302 Lannion
Cedex, France,
Adam.Dybbroe_at_smhi.se
Introduction
Norrköping, SMHI
Smidbjerg, DMI
NOAA14
The position of an AVHRR footprint depends on
time, satellite position and velocity, satellite
attitude (its orientation) and radiometer viewing
geometry. The radiometer geometry is known before
launch. Time is usually available, e.g. through
the satellite time corrected from the satellite
clock error, or from an independent clock.
Satellite position and velocity may be calculated
by an orbit prediction model ingesting daily
bulletins (e.g. TBUS). The remaining unknown is
the satellite attitude, or in fact how the actual
attitude deviates from its nominal value. It is
well known that without any kind of special
post-processing locally received NOAA AVHRR data
may have navigational errors causing the AVHRR
footprints to be displaced several kilometres
(not seldom around 10 or more). These errors
fluctuates in time (significant changes from day
to day is the normal) as can be seen from figure
1, and are caused by deviations between the
nominal satellite attitude and the actual,
unknown, attitude. Exact navigation may not be
crucial when data are used only for image display
at the forecasters desk. But inaccurate
navigation may impose serious problems when data
are used as input to a physical retrieval, like
e.g. an objective cloud classification scheme
(see figure 2). Most modern geophysical satellite
retrievals rely on other data sources like e.g.
high-resolution land
Lannion, CMS
Distance (km)
For the material discussed in this poster we used
NOAA HRPT data from these three receiving
stations. The DMI station served as a backup for
the station at Norrköping during temporary
antenna failures in April and May 2001.
Figure 2 Nowcasting SAF Cloud Mask assuming zero
attitude (left) and with navigation adjustment
using ANA (right) over the Norwegean Sea and
Lofoten (Norway) April 8, 2001, 1105 UTC,
NOAA16. Blue and green are cloud free sea and
land respectively, and white is either snow cover
or sea ice. Clouds are either grey (cloud
contaminated pixels) or orange (cloud filled
pixels). Notice that the poor navigation causes
an artificial increase in cloud cover along the
coast, and that snow cover on land may be taken
as sea-ice.
NOAA16
ANA at SMHI - Preliminary results
In January 2001 a first ANA users workshop was
held at the Centre de Météorologie Spatiale,
Météo-France, in Lannion, with the participation
of SMHI, as well as the Danish and Norwegian
Meteorological institutes. In spring 2001 the ANA
software was installed along side AAPP (version
2.8) in a development environment at SMHI, and
tested on locally received HRPT data. In our
first tests we have been running ANA without the
MAIA cloud mask. This set-up proved still to be
able to eliminate cloudy landmarks (no land-sea
mask or bad correlation) and the results showed
to be in good agreement with those at CMS. Except
for the pitch error in the beginning of April,
there are no significant deviations in the
attitude angles as estimated at SMHI and CMS. The
solar activity was very high in early April 2001
causing large errors in the orbit prediction
which show up as large pitch errors. From April 4
till April 8 the TBUS messages were not updated
at SMHI, probably explaining the large deviations
between the CMS and SMHI estimates. It can be
seen that there seem to be a bias in the roll of
NOAA 16 of around 0.5 to 1 mrad. The pitch
fluctuates a lot even during normal solar
activity and it is quite large for NOAA 16 (5-10
mrad) during this period. It is, however,
important to notice that the pitch error also
include other errors (orbit prediction and clock
errors). ANA is now installed in a
semi-operational environment at SMHI for the
production of the prototype cloud and
precipitation products of the Satellite
Application Facility (SAF) to support Nowcasting.
See www.smhi.se/saf, Dybbroe et al. (2000), Thoss
et al. (2001) and Dybbroe et al. (2001) for a
description of the AVHRR AMSU/MHS based cloud
and precipitation products and algorithms
developed at SMHI.
Distance (km)
use, and without precise navigation of the
satellite data it is difficult to make full use
of such data. Some retrievals rely on data from
several satellite overpasses (e.g. snow cover
mapping or SSTs), and are thus particular
sensitive to fluctuations in navigation
errors. The actual satellite attitude can be
estimated if an adjustment is performed on the
raw data (navigated using the nominal attitude)
using known landmarks, as has been done
operationally at CMS, Météo-France since 1990,
using their Automatic Navigation Adjustment (ANA)
technique.
Figure 1 RMS distance error in km for NOAA14
(top panel) and NOAA 16 (lower panel) during
April and May 2001, using nominal attitude (red
dashed) and after having applied the navigation
adjustment (blue).
Attitude (Yaw, Roll and Pitch) errors during
April and May 2001, as estimated by ANA at SMHI
Norrköping (red) and at CMS Lannion (blue)
NOAA 14 pitch error
NOAA 14 yaw error
NOAA 14 roll error
Yaw (mrad)
Future plans
Roll (mrad)
Pitch (mrad)
The preliminary results during spring summer and
autumn at high latitudes are promising. However,
a clear weakness of ANA is the nighttime land-sea
discrimination method (also used for twilight
conditions) which depends on a single Tb4-Tb5
histogram. Depending on the season, geographical
area and hour in the day, this method may fail to
build the land-sea mask of a viewed landmark,
which reduce the number of images where the
navigation can be successfully adjusted. The
high-latitude winter season with little or no
daylight, snow cover on the ground and sea ice
along the coasts, and cold land and water
surfaces, is particularly problematic. With the
technique as implemented today ANA is expected to
fail on a major part of the satellite passes
received at SMHI Norrköping during the winter
season (November-March). Therefore, a small
project has been set up at SMHI with support from
the Swedish National Space Board (SNSB) in order
to improve on the performance of ANA in general
and at high latitudes in particular. Our focus
will be on trying to improve on the nighttime
algorithm, making use of more image features than
just the current Tb4-Tb5.
NOAA 16 yaw error
NOAA 16 pitch error
NOAA 16 roll error
Yaw (mrad)
Roll (mrad)
Pitch (mrad)
Acknowledgement
The NOAA reception at SMHI Norrköping failed
during major parts (27-30 April and 17 31 May)
of the study period discussed, due to severe
problems with the antenna. But thanks to DMI, and
in particular Søren Andersen who made their
locally received NOAA HRPT level 0 data
accessible to us we managed to keep a continuous
data series for research and development. This
work is part of a project sponsored by the
Swedish National Space Board and SMHI.
What is ANA?
4) For each cloud free (cloud free or nealy
cloud free as determined by the cloud mask, e.g.
MAIA) window a binary land-sea image is
constructed, using AVHRR channel 1 and 2 for
daytime conditions, and channel 4 and 5 at
night. 5) A similarity coefficient is calculated
for all possible displacements. The displacement
corresponding to the maximum of the similarity
coefficient gives the landmark navigation error
expressed in line and pixel numbers. 6) Attitude
estimation Assuming the attitude error is
constant over the whole image, it can be
estimated by the rms resolution of a system
involving the measured landmarks position in the
image and their true latitudes and longitudes.
The Automatic Navigation Adjustment (ANA)
technique was developed at CMS Météo-France in
the 1980ties, to provide highly accurate
geo-located AVHRR data. ANA combines a physical
image deformation model and automatic adjustment
on coastal landmarks. The navigation adjustment
is done in satellite coordinates allowing
interpreting the landmark navigation errors in
terms of satellite attitude yaw, pitch and roll.
In short the adjustment involves the following
six steps
References
1) A generation of a set of coastal landmarks
from the World Vector Shoreline database
(obtained freely with the Generic Mapping Tools
http//gmt.soest.hawaii.edu/). Done only once for
your local receiving station. 2) Landmark
location Using the physical deformation model
and the nominal attitude the landmarks inside the
swath are converted to binary land-sea images
expressed in satellite coordinates. Each such
reference window is associated with an actual
window in the AVHRR image, which is centred on
the calculated landmark position and fully
containing the reference window. 3) Cloud mask A
cloud mask may be derived over every actual
window and those that are too cloudy are
rejected. This step is presently omitted at SMHI.
Bordes, P., Brunel, P. and Marsouin, A., 1992
Automatic Adjustment of AVHRR Navigation. Journal
of Atmospheric and Oceanic Technology, Vol. 9,
No. 1, 15-27. Brunel, P. and Marsouin, A., 2000
Operational AVHRR navigation results. Int. J. of
Remote Sensing, vol. 21, no. 5, 951-972. Dybbroe,
A., Thoss, A. and Karlsson K.-G., 2000 The AVHRR
AMSU/MHS products of the Nowcasting SAF.
Proceedings of the 2000 Eumetsat Meteorological
Stallite Data Users Conference, Bologna, Italy,
pp. 729-736. ISBN 92-9110-037-4. Dybbroe, A.,
Thoss, A. and Karlsson, K.-G., 2001 Validation
of Nowcasting SAF Polar platform products.
Proceedings of the 2001 Eumetsat Meteorological
Satellite Data Users Conference, October 1-5,
Antalya, Turkey, pp. 344-451. ISBN
92-9110-044-7. Thoss, A., Dybbroe, A. and
Bennartz, R., 2001 The Nowcasting SAF
Precipitating Clouds Product. Proceedings of the
2001 Eumetsat Meteorological Satellite Data
Users Conference, October 1-5, Antalya, Turkey,
pp. 399-406. ISBN 92-9110-044-7.
See Bordes et al. (1992) and Brunel and Marsouin
(2000) for a detailed description of ANA. The ANA
software has recently been made compatible with
the EUMETSAT ATOVS and AVHRR processing package,
AAPP, version 3.0, which is becoming a standard
also for processing AVHRR data. The software is
available on request, as such, without any
commitment to further maintenance. It is
described in a User's Manual. Contact at CMS
pascal.brunel_at_meteo.fr