MIPAS Microwindow Error Analysis
Last Updated: 08OCT07
Revised SPREAD error (from 2% down to 0.2%) and GAIN error for C,D bands
(from 2% to 1%) for OFL data.
The following table shows the error analysis for the
nominal sets of microwindows used in both
Near Real Time (NRT) and
Off-Line (OFL) processing in
normal MIPAS operations, ie about 95% of the time.
(The main difference between the two sets is that that the OFL processing uses
an extended altitude range for most species).
These errors have been evaluated for 5 different atmospheric conditions.
- DAY
- Mid-Latitude day-time (similar to US Standard Atmosphere)
- NGT
- Mid-Latitude night-time
- SUM
- Polar Summer day-time
- WIN
- Polar Winter night-time
- EQU
- Equatorial day-time
- GLW
- Not actually an atmospheric profile but a global
composite of results for the five
atmospheres, with twice the weight given to results from the polar winter case.
Click on the 'NRT' or 'OFL' labels
in the following table for plots of the error analysis for
each combination of atmosphere/retrieved species, and on 'Data' for the
numerical data that was plotted (OFL retrievals only).
In the plots, the same symbols are used for each
error source (explained below)
throughout, and listed in the key in the approximate order of
significance for that plot. Only the most significant errors are plotted.
Click on the atmosphere
or species for plots of the atmospheric profiles assumed.
List of errors considered
- TOT
- Total Error. Root sum square of SYS and RND components
- SYS
- Systematic Error. Root sum square of individual systematic error
sources (ie everything except random error)
- RND
- Random Error. Due to the propagation of instrument noise through the
retrieval. Based on in-flight values of NESR from orbit 2081.
- (Sep 05) in pT retrieval, this includes a modified a priori pointing
covariance which more closely simulates that used in the retrieval.
- NB: A more accurate assessment of this component is included in the L2
product
- NONLTE
- Non-LTE error. Due to assumption of local thermodynamic equilibrium when
modelling emission in the MIPAS forward model. Based on calculations
using vibrational temperatures supplied by M.Lopez-Puertas, IAA, Granada.
- SPECDB (formerly referred to as HITRAN)
- Spectroscopic database errors. Due to uncertainties in the strength,
position and width of infrared emission lines.
Based on estimates supplied for each molecule/band by J.M.Flaud, LPM,
Paris.
- GAIN
- Radiometric Gain Uncertainty. Due mostly to non-linearity correction in
bands A, AB and B. A uniform value of ±2% has been assumed for bands
A, AB and B, and ±1% for bands C and D
- SPREAD (replaces previous ILS error)
- Uncertainty in width of apodised instrument line shape (AILS).
A value of 0.2% has been assumed based on likely variations in apodised
instrument line shape from modelled.
- SHIFT
- Uncertainty in the spectral calibration. The design specification of
±0.001cm-1 has been used, and is consistent with the 1st derivatives signatures
in the residual spectra.
- CO2MIX
- CO2 line-mixing. Due to neglecting line-mixing effects in the retrieval
forward model (only affects strong CO2 Q branches in the MIPAS A and D
bands)
- CTMERR
- Uncertainty in gaseous continua. Assumes an uncertainty of ±25% in
the modelling of continuum features of H2O (mostly), CO2, O2 and N2.
- GRA
- Horizontal gradient effects. Due to retrieval assuming a horizontally
homogeneous atmosphere for each profile. Error is calculated assuming
a ± 1K/100km horizontal temperature gradient.
- HIALT
- Uncertainty in high-altitude column. Retrieval assumes a fixed-shape of
atmospheric profile above the top retrieval level. Effect is calculated
assuming `true' profile can deviate by climatological variability.
- PT (OFL data only, added Sep05)
- Propagation of pT random covariance into VMR retrieval
- TEM (NRT data only)
- Temperature propagation error. Temperature and pressure are retrieved
first, this represents the contribution of a nominal 1K temperature error
into the constituent retrievals.
- NB: A more accurate assessment of this component is included in the L2
product and is typically 50% larger
- PRE (NRT data only)
- Pressure propagation error. As with temperature, effect of a nominal
2% pressure retrieval uncertainty
- NB: A more accurate assessment of this component is included in the L2
product and is typically 50% larger
- [species]
- Uncertainties in assumed profiles of contaminant species.
For most species this is the climatological 1-sigma variability
(profiles supplied by J.Remedios, U.Leicester). However,
for contaminant species which are also retrieved by MIPAS
(ie CH4,H2O,HNO3,N2O,NO2,O3) the retrieval total error is assumed where this is
smaller than the climatological variability.
Use of Systematic Errors
The definition of 'systematic error' here includes everything which
is not propagation of the random instrument noise through the retrieval.
However, to use these errors in a statistically correct manner for
comparisons with other measurements is not straightforward.
Each systematic error has its own length/time scale:
on shorter scales it contributes to the Bias and on longer scales contributes
to the SD of the comparison.
Fortunately, two of the larger systematic errors (PT and SPECDB)
can be treated properly:
The pT propagation error (PT)
is uncorrelated between any two MIPAS profiles
(since it is just the propagation of the
random component of the pT retrieval error through the VMR retrieval)
so contributes to the SD of any profile comparison
Spectroscopic database errors (SPECDB) are constant but of unknown sign,
so will always contribute to the Bias of any comparison, but note that
the magnitude of these errors is very uncertain.
Of the other significant errors, the calibration-related errors
(GAIN, SHIFT, SPREAD)
should, in principle, be uncorrelated between calibration cycles however
analysis of the residuals suggests that these errors are almost constant
so could be included in the Bias.
The horizontal gradient (GRA), high altitude column (HIALT)
and contaminant gas errors ([species])
are likely to be correlated over small areas (1000km) or times (weeks), hence
contribute to the Bias for localised comparisons,
but as the comparison datasets are extended
these errors will contribute more to the SD.
Line mixing errors (CO2MIX)
are also contribute towards the Bias but in principle the
sign of these errors is known (unlike spectroscopic errors) so this bias
could be removed.
Non-LTE errors (NONLTE)
should also, in principle, contribute a known Bias but these
are highly variable (especially diurnally) so care has to be taken to make
sure that representative conditions for the comparison are used.
Reference
- Microwindow
Selection for High-Spectral-Resolution Sounders
- App. Optics, 41, 3665, 2002
- Dudhia, A., V. L. Jay and C. D. Rodgers.
