MIPAS Orbital Retrieval using Sequential Estimation (MORSE)
01MAY03
Introduction
The Michelson Interferometer for Passive Atmospheric Sounding is one of the
on the European Space Agency's Envisat satellite launched in
March 2002. MIPAS measures atmospheric limb emission spectra from 685-2410
cm-1 (14.5 to 4.1um) over an altitude range 6-68 km.
Since MIPAS is one of the core instruments, ESA have undertaken to process and
distribute MIPAS data operationally. This includes both atmospheric spectra (L1
data) and derived profiles of atmospheric pressure, temperature and
concentrations of CH4, H2O, HNO3, N2O, NO2 and O3 (L2 data). The ESA processing
algorithms were designed before launch so are intentionally relatively simple
and robust. However the software is complex and not publicly available.
In addition to the ESA L2 products, MIPAS has the potential to retrieve
numerous additional species or parameters, or the same species using more
sophisticated algorithms, and a number of laboratories around Europe have been
developing their own codes to do this, coordinated under the EU Framework V
project AMIL2DA). However, these codes are generally designed only for local
use by experts and often require large, dedicated machines.
The aims of this project are
- to generate a retrieval algorithm MORSE which can
either replicate or improve upon the ESA L2 processing, and
- to make the software available for NERC users to download and install using
only basic computing facilities and background knowledge.
The code will be written in FORTRAN and designed for a Unix/Linux operating
system. Documentation and tutorials will be in HTML and available on the web.
The goal is to provide the resources to allow a postgraduate student
or postdoctoral researcher working independently to perform their own
retrievals from MIPAS spectra while leaving sufficient time to concentrate
mainly on the scientific aspects of the results rather than the details
of the processing.
Schedule
Since MIPAS is already operational it is considered desirable to generate
usable software in as short a period of time as possible.
An additional advantage is to bring users early experience of the code
during the development in order to generate feedback, help with testing and
refine the user-requirements.
The project is split into 4 phases, each of 6 months, with software releases at
the end of each phase. The start date is 1st January 2004.
-
6months - Phase I - MORSE v1 (1st July 2004)
-
12months - Phase II - MORSE v2 (1st January 2005)
-
18months - Phase III - MORSE v3 (1st July 2005)
-
24months - Phase IV - MORSE v4 (1st January 2006)
The additional capabilities of each phase of development are outlined in the
science case below.
Science Case
In the following, limitations of the ESA L2 products are assessed together with
the additional capabilities offered by the MORSE code at each phase of
development.
These comments are based on the current state of the ESA processor and it
should be noted that upgrades are anticipated in the time scale of the order
of a year.
Phase I
- MORSE v1 Features:
- Retrieval of single profile from nominal MIPAS scan
- ESA or user-supplied microwindows
- Optimal estimation retrieval
- Reprocessing
- ESA's processing is fully automated so that if retrievals fail for some
reason, e.g., due to convergence or cloud contamination problems, the profiles
are simply unavailable until possibly the next reprocessing run.
The code will allow a user to hand-generate missing data, e.g., critical to
a particular campaign, using the same spectral data as the ESA processor
(although not an identical retrieval algorithm).
- Aims: recover missing ESA profiles where L1 data is available
- Microwindow selection
- The ESA retrieval uses a fixed set of microwindows and spectroscopic data.
A more flexible approach would allow the sensitivity to spectroscopic
parameters to be assessed, e.g. assessing the impact of using a different
set of line parameters for a particular molecule
- Aims: spectroscopic validation
- Improved ESA Products
- ESA retrieve profiles of pT and key-species in near-real-time using a limited
set of microwindows - typically only 2 or 3 per species. Additional accuracy
can be obtained using more microwindows, at the expense of additional cpu time.
- Aims: improved accuracy of standard products
- Lower altitude limit
-
ESA's current processing is limited to 12km and above, ignoring spectra from
6 and 9km in order to minimise problems with clouds and convergence issues
caused by large tropospheric water vapour. The option will be available to
retrieve at these altitudes where clouds do not occur.
- Aims: Upper-troposphere/lower stratosphere retrievals
- Additional Products
- Many more species can be retrieved, albeit with lower
signal/noise or under enhanced conditions, using essentially the same algorithm
e.g. CFCs, N2O5, Sulphur budget (SO2, SF6, OCS),
Chlorine budget (ClONO2, HOCl, CCl4, ClO), NH3, HCN, C2H6 and H2O2
- Aims: additional species
- Retrieval algorithm
-
There are several possible methods for inverting a set of radiance measurements
to retrieve a profile. ESA use the least squares fit approach which has its
advantages in terms of being unbiased and maximising vertical resolution but
suffers from instabilities and forces a specific set of retrieval
levels to match the available measurements.
It is intended to provide alternative approaches
based on optimal estimation or regularisation in order to improve
the robustness and flexibility of the retrieval.
- Aims: more uniform retrievals with well-defined averaging kernels
Phase II
- MORSE v2 Features:
- Simultaneous retrievals of multiple species
- Flexible retrieval/measurement spacing
- Choice of optimal estimation or regularisation
- Retrieval sequence
-
A limitation on the ESA processing is that first pressure and temperature
are retrieved, followed by each species individually. This necessitates the
use of microwindows where the emission is primarily from a single species.
Allowing for joint retrievals (e.g., pT plus H2O simultaneousl) allows use
of spectral intervals where there is overlap between different molecules and
should improve retrievals of the marginal species.
- Aims: improved accuracy
- Special Modes
-
MIPAS has a nominal scan mode of 17 elevation steps from 68-6km but also a
number of so-called "special modes" where the scan pattern, and sometimes
spectral resolution, are changed, e.g, concentrating on the UTLS region
and improving the horizontal sampling. These special modes will be operated for
a few days each during the first part of 2003 but ESA currently plan only on
processing these data to L1, ie generating spectra but not retrieving
profiles. The intention is for the science community to establish the
usefulness of such observations so processing of this data will be included
here.
-
Aims: retrievals of profiles during MIPAS Special Modes (eg polar vortex,
strat/trop exchange, high horizontal resolution)
- Retrieval algorithm
-
There are several possible methods for inverting a set of radiance measurements
to retrieve a profile. ESA use the least squares fit approach which has its
advantages in terms of being unbiased and maximising vertical resolution but
suffers from instabilities and forces a specific set of retrieval
levels to match the available measurements.
It is intended to provide alternative approaches
based on optimal estimation or regularisation in order to improve
the robustness and flexibility of the retrieval.
- Aims: more uniform retrievals with well-defined averaging kernels, eg for data
assimilation studies
Phase III
- MORSE v3 Features:
- Choice of continuum constraints
- Handling of horizontal gradients
- Orbital processing
- Retrieval of vibrational temperature and isotopic ratios
- Isotopic ratios
- Isotopic ratios give additional information on the history and source of
atmospheric species. With MIPAS spectral resolution it is possible to isolate
spectral lines from individual isotopes. Although these are generally much
weaker than the main isotopic lines (through a lower concentration) a retrieval
formulated in terms of determining the ratio of minor/main isotopes rather than
the absolute concentration should remove many sources of inaccuracy.
- Aims: Water vapour isotope distributions are the most likely target.
- Vibrational Temperatures
- The ESA retrieval assume local thermodynamic equilibrium, i.e. that all
spectral emissions are characterised by the Planck function for the local
temperature. At high altitudes, or short wavelengths, especially during
day-time conditions the emissions of particular molecules deviate from this
and are typically characterised by an internal `vibrational' temperature.
The proposed retrieval will either allow vibrational temperature profiles
to be externally supplied or retrieved directly. This may allow the retrieval
of CO.
- Aims: CO retrievals, non-LTE modelling
- Continuum constraints
- The ESA retrieval is built around the approach of simultaneously fitting an
independent atmospheric continuum profile in each microwindow, effectively
the molecular signature becomes the difference between the line peak and the
continuum background rather than the zero radiance level. This is an advantage
for molecules with distinct emission line features, e.g., in eliminating
contributions from aerosol or thin clouds, but restricts the
sensitivity to retrieving heavy molecules such as N2O5 and the CFCs.
Two alternatives will be provided: either using broader microwindows or
retrieving continuum absorption
independently, e.g. from adjacent spectral regions, which is a useful
product in itself.
- Aims: cloud/aerosol studies, additional continuum-like species
- Horizontal Gradients
- The ESA retrieval assumes a horizontally homogeneous atmosphere when
retrieving each profile, e.g., that the atmospheric parameters retrieved
from the 21km tangent point are the same as those encountered by the ray
passing through the 18km tangent point as it crosses the 21km surface,
even though these are displaced some 75km horizontally either side.
One approach being explored by several groups is a 2-dimensional retrieval
where a two dimensional field is retrieved in the vertical plane
(e.g., a full orbit) rather than a single profile. However, this greatly
increases the processing requirements. The approach adopted here will be
a two-pass sequential estimator (Kalman smoother)
where profiles are first retrieved assuming
no gradient, then subsequently reretrieved using gradient information from
the first pass.
-
Aims: improved accuracy, particularly in regions of sharp gradients, generation
of larger datasets
Phase IV
- MORSE v4
- The last phase is intended only to allow for a period of support including
code maintenance, bug-fixes, improvements to documentation and minor
algorithm refinements based largely on user-feedback.
This will culminate in a `final' release of MORSE v4.