2.4.4.2 Products
2.4.4.2.1 Level 2 Products Description There exists 2 forms of level
2 products. The level 2 product 2.4.4.2.1.1.
itself and the level 2 NRT / meteo
product 2.4.4.2.1.2. containing a subset of the
data contained in the relating level 2
product.
2.4.4.2.1.1 Level 2 Product
a:

Sequence
of species is given
in Specific Product Header

The level 2 product consists of
following elements (see index)
The detailed structure of the level 2
product can be found here .
2.4.4.2.1.2 Level 2 NRT /Meteo Product
The level 2 NRT / Meteo
product consists of following
elements (see index) The
detailed structure of the level 2 NRT / Meteo product
can be found here .
2.4.4.2.1.3 Product Header The product header is
divided into three parts:
2.4.4.2.1.3.1 Main Product Header The main product
header is the same for all ENVISAT products.
It specifies basic product
information such as origin of data,
processing site, processing software
version, UTC time of data
sensing and processing, orbit and
velocity parameters of ENVISAT, quality
indicators for input data, etc.
The detailed structure of
the MPH can be found
here 6.5.1. .
2.4.4.2.1.3.2 Specific Product Header The SPH is the same
for both forms of level 2 products.
It contains information
applicable to the whole level 2
product file such: The detailed structure of
the level 2 SPH can be found
here 6.5.50. .
2.4.4.2.1.3.3 Data Set Descriptors The DSD provide
information on structure and size of
included or referenced measurement
and annotation data.
2.4.4.2.1.4 Summary Quality ADS This ADS consists of 1 ADSR containing
the numbers of p,T and VMR retrievals
terminated unsuccessful because of
 excess of allowed number of
macro iterations
 excess of allowed number of
micro iterations
 excess of allowed processing time
The detailed structure of this
ADS is slightly
different for the 2 forms of level 2
products. It can be found here 6.5.45. for the full
level 2 product and here 6.5.54. for the NRT / meteo product.
2.4.4.2.1.5 Scan Geolocation ADS This ADS contains 1 ADSR per scan
providing geolocation information. It
includes: The detailed structure of this
ADS is the same for
both forms of level 2 products. It
can be found here 6.5.39. .
2.4.4.2.1.6 Structure ADS This ADS provides the
parameters which determine the size of
the DSRs of the following
MDSs and ADSs. A new ADSR is added every
time one of these parameter change
its value. Each ADSR contains:
 The number of sweeps per scan.
 The numbers of retrieved
profile, continuum and offset values.
 Indicators for existence of p,T
error propagation data.
 The numbers of microwindows.
 The numbers of spectral grid points.
 Offsets and size of first DSRs within
the following MDSs and ADSs this ADSR refers to.
The detailed structure of this
ADS can be found here 6.5.46. . It is the same
for both forms of level 2 products, but
within the NRT / meteo product
several fields are not used.
2.4.4.2.1.7 Scan Information MDS This MDS contains 1 MDSR per scan providing
geolocation information of the
individual sweeps and a subset of
the relating retrieval results. It
includes:
 The ZPD time
of each sweep of the
scan.
 The WGS84 latitude
and longitude of the tangent
points of each sweep of the
scan.
 The altitudes of the tangent
points of each sweep of the
scan.
 Flags indicating success of p,T
and VMR retrievals.
 The logical retrieval vectors
for p,T and VMR retrievals.
 The corrected altitudes for each
sweep.
 Retrieved pressure, temperature
and VMR profiles.
 Concentration and vertical
column density profiles.
 Variance data for height
correction, pressure,
temperature, VMR ,
concentration and vertical
column density.
The detailed structure of this
MDS is slightly
different for the 2 forms of level 2
products. It can be found here
for the full level 2 product and here
for the NRT / meteo
product.
2.4.4.2.1.8 PT Retrieval MDS This MDS contains 1 MDSR per scan providing the
following results of the p,T retrieval:
 The condition terminating the iterations.
 The last value of chi^{2}.
 The source of used initial guess data.
 Height correction and retrieved
pressure and temperature profiles.
 Variance/covariance matrices for
pressure, temperature and height correction.
 The pressure/temperature
covariance data.
The detailed structure of this
MDS is the same for
both forms of level 2 products. It
can be found here 6.5.48. .
2.4.4.2.1.9 VMR Retrieval MDSs For each of the target
species H_{2}O, N_{2}O,
HNO_{3}, CH_{4},
O_{3} and NO_{2} 1 VMR retrieval MDS is contained in
the full level 2 product. The
sequence of the species is given in the
specific product
header. The NRT / meteo product
contains 2 of these MDSs covering
O_{3} and H_{2}O. Each
of these MDSs contains 1 MDSR per scan providing the
following results of the VMR retrieval:
 The condition terminating the iterations.
 The last value of chi^{2}.
 The source of used initial guess data.

VMR,
concentration and vertical
column density profiles together
with the relating
variance/covariance matrices.
 The p,T error propagation
variance/covariance matrix, if available.
The detailed structure of this
MDS is the same for
both forms of level 2 products. It
can be found here 6.5.49. .
2.4.4.2.1.10 Continuum and Offset MDS This MDS contains 1 MDSR per scan providing for p,T
and VMR retrievals the
results concerning continuum and offset
fit. It includes:
 For each microwindow
the fitted instrument offset and
relating variance data.
 For all altitudes used for
continuum fit the fitted or
interpolated continuums together
with the relating microwindows,
grouping types and the
covariance with fitted VMR, resp.
fitted pressure and temperature.
This MDS is not contained
in the NRT / meteo product.
The detailed structure can be found here 6.5.47. .
2.4.4.2.1.11 PCD Information ADS This ADS contains 1 ADSR per scan
providing information on retrieval
quality for p,T and VMR retrievals. It
includes:
 The numbers of iterations performed.
 The partial chi^{2}
depending on sweep and microwindow.
 The evolution of
chi^{2}, lambda and
retrieved parameters during
macro iterations.
 Additional information produced
by the processor in textual form.
This ADS is not contained
in the NRT / meteo product.
The detailed structure can be found here 6.5.43. .
2.4.4.2.1.12 MW Occupation Matrix ADS This ADS provides
information on the microwindows selected
for p,T and VMR retrievals. A new
ADSR is added every
time the selection changes. It includes:
 For each sweep the
labels of the selected microwindows
(all retrievals).
 For each retrieval the logical
retrieval vector.
The detailed structure of this
ADS is the same for
both forms of level 2 products. It
can be found here 6.5.41. .
2.4.4.2.1.13 Residual Spectra ADS This ADS provides
information on the residual spectra. A
new ADSR is added every
time the microwindow selection
changes. It includes for each retrieval:
 The mean values and standard
deviations of residual spectra
computed from all scans covered
by current ADSR .
 The number of retrievals
contributing to the current ADSR .
 The spectral mask relating to
the current data.
This ADS is not contained
in the NRT / meteo product.
The detailed structure can be found here 6.5.44. .
2.4.4.2.1.14 Parameters ADS This ADS provides
information on instrument and processing
parameters. It contains:
 The actual elevation angles for
current scan.
 The pressure levels relating to
the retrieved profiles.
 The allowed number of macro and
micro iterations.
The detailed structure of this
ADS is slightly
different for the 2 forms of level 2
products. It can be found here 6.5.42. for the full
level 2 product and here 6.5.53. for the NRT / meteo product.
2.4.4.2.2 Extraction of Profile Data from
Level 2 Products There are 2 ways to extract
profile data from level 2 products. If
neither covariance data nor continuum and
offset data is requested, the data can
be extracted from the Scan Information MDS.
Variance/covariance data for retrieved
pressure, temperature and height correction
can be extracted from the p,T retrieval MDS.
Variance/covariance data for retrieved VMR,
concentration and vertical column density
can be extracted from the VMR retrieval MDSs.
Continuum and offset data has to be
extracted from the Continuum and Offset
MDS.
Extraction of Profile
Data from Scan Information MDS
The Scan Information MDS contains a subset
of the p,T and VMR retrieval results. Within
this MDS all fields containing profile
information are of size N^{sw}
6.5.46. . For sweeps not used for fit the
logical retrieval vectors (lfit) contain 0
and the relating profile entries are set to
NaN.
Extraction of p,T
Retrieval Data For the
extraction of the p,T retrieval data for the
ith sweep, the ith
entry of the logical
retrieval vector for p,T retrieval
(lfit) has to be checked. If it is 0,
the ith sweep has been not been
used for p,T retrieval and therefore the
ith entries of the profiles of
pressure,
variance
of pressure, corrected
altitudes, variance
of height correction, temperature
and variance
of temperature are set to NaN.
Otherwise these entries contain the
profile data related to the ith
sweep as shown in table below.
sweep index

tangent height,
geolocation, time

lfit

pressure,
temperature,
corrected
height
variance of
pressure,
temperature and
height correction

0

valid value

1

valid value

1

valid value

1

valid value

2

valid value

1

valid value

3

valid value

0

NaN

4

valid value

1

valid value

5

valid value

1

valid value

6

valid value

1

valid value

7

valid value

1

valid value

8

valid value

1

valid value

9

valid value

1

valid value

10

valid value

1

valid value

11

valid value

1

valid value

12

valid value

0

NaN

13

valid value

1

valid value

14

valid value

1

valid value

15

valid value

0

NaN

Extraction of VMR Retrieval
Data For the extraction of the VMR
retrieval data of the ith sweep,
the ith entry of the relating
logical
retrieval vector (lfit) has to be
checked. If it is 0, the ith
sweep has been not been used for fit in
species #1 VMR retrieval and therefore the
ith entries of the profiles of
VMR,
variance
of VMR, concentration,
variance
of concentration, vertical
column density and variance
of vertical column density are set
to NaN. Otherwise these entries contain
the profile data related to the ith
sweep, as shown in the table below.
Relating altitude, pressure and temperature
information can be extracted as described above.
sweep index

lfit p,T

p,T retrieval
profiles

lfit VMR

VMR, concentration,
vertical column
density
variance of VMR,
concentration and
vertical column
density

0

1

valid value

0

NaN

1

1

valid value

1

valid value

2

1

valid value

1

valid value

3

0

NaN

0

NaN

4

1

valid value

1

valid value

5

1

valid value

1

valid value

6

1

valid value

1

valid value

7

1

valid value

1

valid value

8

1

valid value

1

valid value

9

1

valid value

1

valid value

10

1

valid value

0

NaN

11

1

valid value

0

NaN

12

0

NaN

0

NaN

13

1

valid value

0

NaN

14

1

valid value

0

NaN

15

0

NaN

0

NaN

Extraction of Profile
and VCM Data from p,T Retrieval MDS
Within the p,T retrieval MDS 6.5.48. only
the sweeps used for p,T retrieval are
considered. Therefore the size of the fields
containing profile information are of size
N_{pT}
6.5.46. . For height correction
N_{pT}1 profile are used
because no height correction is applied to
lowest used sweep. Extraction of
Profile Data For the extraction
of pressure 6.5.48. , temperature 6.5.48. and height correction 6.5.48.
profiles from the p,T retrieval MDS, the
relating logical
retrieval vector has to used to
determine the profile indices relating
to requested sweep indices, as shown in the
example below.
sweep
index

lfit p,T

index on pressure
and temperature
profile

index on height
corrections

0

1

0

0

1

1

1

1

2

1

2

2

3

0





4

1

3

3

5

1

4

4

6

1

5

5

7

1

6

6

8

1

7

7

9

1

8

8

10

1

9

9

11

1

10

10

12

0





13

1

11

11

14

1

12



15

0





Extraction of VCM Data The
mapping between sweep indices and the
indices on the variance/covariance matrices
in the p,T retrieval MDS is the same as
shown above. For the
symmetric variance covariance matrices for
pressure 6.5.48. , temperature 6.5.48. and height correction 6.5.48. only
the diagonal elements and the elements
below are included in the MDS, as shown
below.
element[0,0]





element[1,0]

element[1,1]




element[2,0]

element[2,1]

element[2,2]



...



...


element[n,0]

element[n,1]

element[n,2]

...

element[n,n]

Extraction of Profile
and VCM Data from VMR Retrieval MDS
Within the VMR retrieval MDSs 6.5.49.
only the sweeps used for relating VMR
retrieval are considered. Therefore the size
of the fields containing profile information
are of size N_{V(i)}
6.5.46. . Extraction of Profile
Data For the extraction of VMR 6.5.49. , concentration 6.5.49. and vertical column
density 6.5.49. profiles from the VMR retrieval
MDSs, the relating logical
retrieval vector has to used to
determine the profile indices relating
to requested sweep indices, as shown in the
example below.
sweep
index

lfit VMR

index on VMR,
concentration and
vertical column
density profile

0

0



1

1

0

2

1

1

3

0



4

1

2

5

1

3

6

1

4

7

1

5

8

1

6

9

1

7

10

0



11

0



12

0



13

0



14

0



15

0



Extraction of VCM Data The
mapping between sweep indices and the
indices on the variance/covariance matrices
in the VMR retrieval MDSs is the same as
shown above. For the
symmetric variance covariance matrices for
VMR 6.5.48. , concentration 6.5.48. and vertical column
density 6.5.48. only the diagonal elements and
the elements below are included in the
MDS, as shown above.
Extraction of Data from
Continuum and Offset MDS Within
the Continuum and Offset MDS the fitted
continuum and offsets of all retrievals are
reported. Extraction of Offset
Data For each retrieval
fields of size N_{offset(pT)}
6.5.46. , respectively N_{offset(V(i))}
6.5.46. are contained in the Continuum
and Offset MDS containing the offset values 6.5.47. itself,
the relating variance data 6.5.47. and the
labels of the relating
microwindows 6.5.47. . Extraction of
Continuum Data For the
extraction of continuum the parameters
number of altitudes used for continuum fit
(N_{cgid(pT)}
6.5.46. , N_{cgid(V(i))}
6.5.46. ) and max. number of microwindows per
height (N_{MWPT}
6.5.46. , N_{MWV(i)}
6.5.46. ) are needed. For each retrieval the indices of sweeps 6.5.47. used
for continuum fit is given in the MDS.
For each of these sweeps the labels of the used
microwindows 6.5.47. , the grouping types 6.5.47. of
these microwindows and the continuum 6.5.47. at these
microwindows is given. For all microwindows
with grouping types 1 .. 4 additionally the
variance of the fitted
continuum 6.5.47. and the covariance with the
fitted profile (i.e. for p,T retrieval covariance of continuum
and pressure 6.5.47. and covariance of continuum
and temperature 6.5.47. , for VMR retrieval
covariance of continuum
and VMR 6.5.47. ) at this sweep is given. All
of these fields are of size N_{MWPT}
6.5.46. , respectively N_{MWV(i)}
6.5.46. with unused fields set to default
values. The following table shows an
example.
sweep index

microwindow

grouping type

continuum

variance, covariance

10

PT__0001

3

valid value

valid values

PT__0002

5

valid value

NaN ^{b}

PT__0010

1

valid value

valid values

blank ^{a}

1 ^{a}

NaN ^{a}

NaN ^{a}

11

PT__0002

1

valid value

valid values

PT__0010

2

valid value

valid values

PT__0011

6

valid value

NaN ^{b}

PT__0012

2

valid value

valid values

13

PT__0002

1

valid value

valid values

PT__0012

1

valid value

valid values

PT__0020

1

valid value

valid values

blank ^{a}

1 ^{a}

NaN ^{a}

NaN ^{a}

14

PT__0012

1

valid value

valid values

PT__0020

1

valid value

valid values

blank ^{a}

1 ^{a}

NaN ^{a}

NaN ^{a}

blank ^{a}

1 ^{a}

NaN ^{a}

NaN ^{a}

a 
unused field 
b 
relating continuum value
is not fitted, but
interpolated (grouping
type > 4) 
2.4.4.3 Appendix A: Mapping of temperature
error on to the retrieved VMR profiles
At a generic iteration, the VMR profile
y is obtained applying to
the profile y
_{0} of the previous iteration the correction:
  eq 2.43 
where n is
the residuals vector, is the VCM of the observed spectra
and K is the jacobian of
the VMR retrieval. An
uncertainty on the assumed tangent
pressures and temperatures, translates into
an error on the simulated spectra and
therefore into an error on the retrieved profile equal to:
  eq 2.44 
where C is
the matrix accounting for p,T error propagation
in the simulated spectra of VMR retrieval and contains the
derivatives:
  eq 2.45 
The index
'i' identifies the fitted
spectral points (as a function of frequency for
all the microwindows and all the tangent
altitudes) and the index 'j'
identifies the retrieved tangent altitudes. In
equation (A2) D is assumed
locally independent on p,T (always true for
small errors ). As the error on the
retrieved p,T is described by a VCM , the corresponding VCM
relating to y and
due to p,T error is given by:
  eq 2.46 
where we have defined
E = DC. E
is the matrix transforming p,T error into VMR
error. Matrix E depends on:
 current atmospheric status (p,T and VMR)
 observation geometry
 set of adopted MWs in VMR retrieval
(Occupation Matrix)
In MIPAS Level 2 processor each OM has
a pT error propagation matrix E
attached to it. Since OMs are tabulated for
latitude bands, it is possible to use different
E matrices for different
latitude bands (of course this does not prevent
from using the same OM throughout the whole orbit
and different E matrices
depending on latitude). Matrix
E and are both outputs of MIPAS
Level 2 processor. The users should use Equ.
(A4) to derive the VMR error component due
pT error propagation.
The dependencies of the pT error propagation
on latitude, atmospheric model and
acquisition scenario are analyzed,
quantified and discussed in [39]
Ref. [1.57 ]
.
2.4.4.4 Appendix B: Algorithm for generation
of the optimized microwindow databases
Microwindow selection is performed by
an algorithm which simulates the propagation of
random and systematic errors through a retrieval
and attempts to maximise the information
content
Bennett, V. L., A. Dudhia and C. D. Rodgers
Ref. [1.69 ]
. The information content of a
microwindow increases as the log of the
determinant of the total covariance
decreases, total covariance being the sum of the
random and various systematic error
covariances. Broadly speaking, 1
`bit' of information is equivalent to a
factor 2 reduction in the uncertainty at one
profile altitude. Microwindows are
created by first selecting a number of single
measurements, identified by location in the
spectral and tangent altitude grids, as
starting points. Adjacent measurements are
added to each until the information content no
longer improves or a maximum width of
3cm^{1} is reached. The best of these
trial microwindows is selected, the retrieval
covariance modified, and the process
repeated for a new set of measurements as
starting points. The procedure of growing
microwindows also allows for measurements within
microwindows to be `masked', i.e., excluded
from the retrieval. This usually applies to
measurements where the associated systematic
errors such as the uncertainty in modelling a
contaminant, outweigh any benefit in the
reduction of the random error when considering
the total covariance. Initially,
a set of typically 10 microwindows, or 10000
measurements (whichever occurs first) is
selected based on the assumption that
spectra for all MIPAS bands are available.
Further microwindows are then selected to
maximise information retrieved in situations
where data from different bnds
may be unavailable. This set of 2030
microwindows constitutes the database.
Occupation Matrices (OM) represent subsets of
microwindows to be used under different
retrieval circumstances, and these are
constructed using the same approach:
selecting the microwindows from the database
(rather than growing new microwindows) in the
sequence which maximises the retrieved
information. A number of these OMs are
precomputed, corresponding to different
bandavailabilities, and associated with
each of these is a single figureofmerit
representing the information content.
References: see
Bennett, V. L., A. Dudhia and C. D. Rodgers
Ref. [1.69 ]
2.4.4.5 Appendix C: Handling of initial guess profiles
For the analysis of a given scan, the
retrieval code needs the following
atmospheric profiles as input:
 Pressure and temperature profiles
 Continuum profiles for
microwindows used in p,T retrieval
 H_{2}O VMR profile
 Continuum profiles for
microwindows used in H_{2}O retrieval
 O_{3} VMR profile
 Continuum profiles for
microwindows used in O_{3} retrieval
 HNO_{3} VMR profile
 Continuum profiles for
microwindows used in HNO_{3} retrieval
 CH_{4} VMR profile
 Continuum profiles for
microwindows used in CH_{4} retrieval
 N_{2}O VMR profile
 Continuum profiles for
microwindows used in N_{2}O retrieval
 NO_{2} VMR profile
 Continuum profiles for
microwindows used in NO_{2} retrieval
 VMR profiles for other contaminants.
These profiles are used in the
different retrievals either as a first guess of
the profiles that are going to be retrieved or
as assumed profiles of the atmospheric model
(profiles of interfering species and p,T
profiles in the case of VMR retrievals).
For each of these profiles both an
apriori estimate and the result of
the 'most recent measurement'
(obtained either from the retrieval of
the previous scans or from a previous
retrieval of the same scan) are available.
The apriori estimate of a
profile is obtained either from climatology
or from the ECMWF.
On the light of the fact that in some cases
the errors of the retrieved profiles may be
very large (VMR retrieval at very low
altitudes, continuum retrieval), using
the 'most recently' retrieved
profiles as initial guess for the retrieval
may not be a good strategy. In MIPAS
Level 2 processor, the initial guess /
assumed profiles are obtained using the
'best estimate' of the profiles
available at the time when the retrieval
is started.
For pressure and temperature profiles the
'best estimate' coincides with the
'most recent measurement'. For
continuum profiles the 'best
estimate' coincides with their
apriori estimate. For all the
other profiles, the 'best
estimate' is obtained by calculating
the optimal estimation (weighted average)
between the most recently measured profiles
and the apriori ones.
The optimal estimation consists in weighting
the retrieved profile, with its covariance
matrix, with the apriori profile,
which is also characterized by a covariance
matrix with eventually large diagonal values.
The optimal estimation of the profiles is
determined not only at the beginning of each
scan analysis, but also after each VMR
retrieval, because the retrieved VMR
profile is used as assumed profile in the
subsequent retrievals (from the same
scan).
