Ongoing projects
Sensitivity test for the optimization
of the spectroscopic database
The spectroscopic database adopted for the reference FM includes all the lines listed in HITRAN 2004. In order to limit the total number of spectral features to be considered in the optimised version of the KLIMA-IASI code, a sensitivity test was carried out in which we evaluated the impact of skipping, for each species, all the lines smaller than 10-5 the largest line in the selected band. The computation of the most intense feature is performed, species by species, for all 47 bands in which we sub-divided the IASI spectral range.
Figure 10 shows the difference between the spectrum simulated with and without the line skipping.
As for the previous tests, the simulations have been performed over the full IASI spectral range and the nominal instrument noise in reported in the figure for comparison.
The largest differences are observed around 700 cm-1 (CO2 band) and are due to the approximation introduced by CO2 lines skipping. Considering that in the optimised version of the KLIMA-IASI code the spectroscopic data for CO2 are taken from the line-mixing database (Niro et al, 2005a,b) and are not affected by the filtering criteria applied to HITRAN 2004, we can adopt the proposed line skipping for all other species without introducing any major source of systematic error in the forward model.
Figure 10
Difference between the spectrum simulated with and without HITRAN 2004 line skipping.
Red line is the nominal instrumental noise;
green line in the nominal instrumental noise reduced by a factor 10
Sensitivity test for the optimization
of the CO2 Line Mixing Model
In the reference forward model line-mixing effects are taken into account, by adding to the simulated radiances the contribution of CO2 line mixing modelled over the full spectral range covered by IASI. A significant reduction of the computing time is obtained in the ARM by evaluating the contribution of line-mixing on a sub-set of spectral intervals.
In order to identify the minimum sub-set compatible with the accuracy requirements of the ARM, we calculated the difference between synthetic spectra simulated over the full IASI spectral range, with and without the contribution of CO2 line-mixing. The residuals are plotted in Figure 9, along with the nominal instrumental noise.
The modelling of CO2 line-mixing can be limited to the spectral intervals, where these differences are larger than the nominal instrument noise, i.e. to the bands [650-800] cm-1 and [2000-2500] cm-1.
Figure 9
Difference between the spectrum simulated with and without the CO2 line mixing model.
The red line is the nominal instrumental noise;
the green line in the nominal instrumental noise reduced by a factor 10
Sensitivity Tests and Performances
of the KLIMA-IASI Retrieval Model
In this section, the results of sensitivity tests carried out to evaluate the performances of the KLIMA-IASI Accelerated Retrieval Model are presented. All tests have been performed using simulated IASI measurements and selecting as retrieval target the temperature profile, the water vapour profile, the total and partial columns of carbon dioxide and ozone and the Earth skin temperature.
A quantitative assessment of the quality of the retrieval products has been derived by computing:
- the retrieval error on the target parameters (including the random and the systematic components);
- the bias between the retrieved and the true value of the target parameters due the approximations of the accelerated retrieval code.
A series of five retrieval tests was conducted, by combining different options for the set up of the FMs used to generate the simulated measurements and to perform retrieval calculations.
In Table 5,, we report a summary of the configurations adopted in the individual cases, indicating:
- (a) which version of the forward model (reference FM or accelerated FM) was used;
- (b) whether the uncertainties on forward model parameters were taken into account in the retrieval process (with or without the VCM of the FM);
- (c) whether retrieval calculations are performed on the full IASI range or on a selection of spectral bands.
|
TEST |
FM of simulated measurements |
FM of the retrieval |
A priori VCM including FM errors |
Spectral band selection |
|
Test 1 |
Reference FM |
Reference FM |
NO |
NO |
|
Test 2 |
Reference FM |
Accelerated FM |
NO |
NO |
|
Test 3 |
Reference FM |
Accelerated FM |
YES |
NO |
|
Test 4 |
Reference FM |
Accelerated FM |
NO |
YES |
|
Test 5 |
Reference FM |
Accelerated FM |
YES |
YES |
Table 5
Retrieval set-up for the sensitivity tests
For each test several runs of the retrieval simulation were repeated with different realizations of the measurement noise, in order to account for the statistical variability of the retrieval products. The number of retrieval simulations performed for the individual test is 4 for the reference case (test 1) e 25 for the other cases (test 2 to 5).
Basic features of the Accelerated Retrieval Model
The basic features of the inverse model implemented in all retrieval tests can be summarised as follows:
- Multi Target Retrieval;
- Pressure retrieval grid for temperature: equal to the retrieval grid of IASI L2 operational products for the first 21 levels from 1050.0 hPa to 478.54 hPa (0-2 km); for the levels above only one every third pressure level is maintained. The total number of levels of the retrieval grid for temperature is 44;
- Pressure retrieval grid for H2O: equal to the retrieval grid of IASI L2 operational products for the first 21 levels from 1050.0 hPa to 478.54 hPa (0-2 km); for the levels above only one every third pressure level is maintained. The total number of levels of the retrieval grid for temperature is 44;
- Pressure retrieval grid for CO2: equal to the retrieval grid of IASI L2 operational products for the first 4 levels from 1050.0 hPa to 978.98 hPa (~0.3 m); for the levels above only one every second pressure level is maintained up to 499.54 hPa (~ 5.6 km). The total number of levels of the retrieval grid for temperature is 12;
- Pressure retrieval grid for O3: one every third pressure level of the retrieval grid of IASI L2 operational products is taken from 1050.0 hPa to 759.16 hPa (~2.4 km). The total number of levels of the retrieval grid for temperature is 4;
- Marquardt parameters: 0.01 for all targets;
- Marquardt and Gauss dumping factor: 3;
- Use of the optimal estimation.
The target parameter of the retrieval are:
- The profile of temperature;
- The profiles of water vapour, carbon dioxide, and ozone;
- The band dependent earth surface temperature.
The atmospheric scenario (true atmospheric state and a priori values) selected for all the retrieval tests were taken from the IG2 climatology developed by University of Leicester for the operational analysis of MIPAS-ENVISAT data (Remedios et al, 2007.). In particular:
- True atmospheric state - IG2 profiles, October 2008 at Northern mid-latitude [20°N - 65°N]
- A priori atmospheric state - IG2 profiles, October 2007 at Northern mid-latitude [20°N - 65°N]
These IG2 profiles were used for the retrieved gases and temperature and pressure profiles.
The thermal contrast between the Earth surface and lowest atmospheric layer (defined as the difference between Earth skin temperature and the air temperature at ground level) was taken equal to zero for the true atmospheric state and equal to +10 K for the a priori state.
The a priori uncertainties that we have assumed on the retrieval targets are reported in Table 6.
|
Target Parameter |
A Priori Error |
Error Correlation Length
|
|
Temperature |
4 K |
3 km |
|
Water Vapour |
100 - 120 % |
5 km |
|
Carbon Dioxide |
10 % |
5 km |
|
Ozone |
100 % |
5 km |
|
Surface Temperature |
20 K |
N.A. |
Table 6
A priori errors and correlation length used for the retrieval tests
Sensitivity test for the optimization
of the fine frequency grid
In order to reduce the overall size of the KLIMA-IASI code and the computing time requested for the simulation of a single spectrum, the sampling interval of the fine frequency grid was reduced from the reference value of 0.001953125 cm-1 (equal to 1/128 the spectral sampling of IASI instrument) to the coarser value of 0.00390625 cm-1 (equal to 1/64 the frequency sampling of IASI instrument).
Figure 8 shows the difference between the spectrum simulated using the downgraded fine wave number grid and the reference fine wave number grid. The impact of this approximation can be evaluated by comparison with the nominal instrumental noise also reported in the figure.
The simulation has been performed on the full IASI spectral range. The larger differences are observed in the CO2 band between 650-800 cm-1 and in the O3 band close to 1100 cm-1, where the contribution of carbon dioxide and ozone concentration in the uppermost levels is important.
Figure 8
Difference between the spectral radiances simulated using the downgraded fine wave number grid
and the reference fine wave number grid. The red line represents the nominal instrumental noise;
the green line in the nominal instrumental noise reduced by a factor 10
.
Sensitivity Tests and Performances
of the KLIMA-IASI Forward Model
In this section we describe the Reference Forward Model that was used to evaluate the impact of different approximations investigated in our sensitivity study along with the outcome of the individual sensitivity tests performed for the optimization of the KLIMA-IASI forward model and with the Accelerated Forward Model (ARM) obtained from the implementation of the resulting trade-offs :
