In this section, the conclusions of the sensitivity tests conducted for the definition of suitable trade-off between efficiency and accuracy of the KLIMA-IASI FM are drawn and the impact of the different approximations and the results of simulations performed using a fully optimised FM are presented. These will show the overall impact on the accuracy of the synthetic spectrum due to the combined approximations that we individually discussed in the previous section.

The accelerated configuration of KLIMA-IASI FM (see Table 3) has been set up by implementing the following options:

• Fine frequency grid with steps of 0.00390625 cm^-1 (equal to 1/64 the spectral sampling of the IASI instrument).
• Contribution of CO₂ line mixing modelled only in the spectral bands [650 - 800] cm^-1 and [2000 - 2500] cm^-1.
• Spectroscopic data are taken from the HITRAN 2004 catalogue for all species other than CO₂ and for each species all spectral features with intensity smaller than 10^-5 the most intense line in the selected frequency band are skipped. For CO₂ the dedicated spectroscopic database used in the line-mixing model is adopted.
• The atmospheric line shape is modelled up to ±25 cm^-1 from the line centre for the target species (H₂O, CO₂ and O₃) and ±10 cm^-1 from the line centre for all other constituents.
• An optimised vertical grid is used consisting of 44 pressure levels from 1050.0 hPa up to 0.005 hPa (21 levels between 1050 and 478.4 hPa and 23 levels between 478.4 hPa and 0.005 hPa).
• The FOV has not been modelled.

Approximations adopted for the KLIMA-IASI AFM with respect to the reference configuration

Figure 13 shows the difference between the spectrum simulated with the reference FM and the accelerated FM.

Figure 13

Difference between the spectrum simulated with the reference FM and the AFM.

The red line is the nominal instrumental noise;

the green line in the nominal instrumental noise reduced by a factor 10

Performances of the KLIMA-IASI Accelerated Forward Model

The level of performances achieved by the AFM must be compared with the requirements specified for the size of the code and for the computing time necessary to perform a single run of the FM without Jacobian calculation.

Table 4 shows that despite the significant reduction obtained with respect to the reference FM both in terms of code size (~50%) and computing time (75%), the AFM is not yet fully compliant with the requirements.

 A further reduction can be obtained by limiting the spectral band to be analysed. By restricting the spectral coverage of the FM calculations to the bands already selected for modelling the line-mixing effects ([650 -800] cm^-1 and [2000 -2500] cm^-1), a 66% reduction of the spectral band to be analyzed is achieved. As shown in Table 4, this is sufficient to satisfy the requirements for integration on G-POD.

Table 4

 Comparison between the performances of the reference and optimised FMs and G-POD integration requirements.

 The reported estimate of the computing time are based on the results