OVERVIEW OF MIPAS OPERATIONAL PRODUCTS

Piera Raspollini(1) , Ginette Aubertin(11) , Sven Bartha(10) , Manfred Birk(12) ,

Bruno Carli(1), Massimo Carlotti(2), Simone Ceccherini(1), Thomas von Clarmann(4),

Marta De Laurentis(9), Bianca Maria Dinelli(3), Anu Dudhia(6), Thorsten Fehr(9),

Herbert Fischer(4), Jean-Marie Flaud(5), Roland Gessner(10), Frank Hase(4),

Michael Höpfner(4), Anne Kleinert(4), Rob Koopman(9), Manuel López-Puertas(8),

Peter Mosner(10), Fabrizio C. Niro(9), Hermann Oelhaf(4), Gaetan Perron(11),

John J. Remedios(7), Marco Ridolfi(2), George Wagner(12) (1) Istituto di Fisica Applicata 'Nello Carrara' (IFAC) del Consiglio Nazionale

delleRicerche (CNR), Firenze (Italy), p.raspollini@ifac.cnr.it (2) Dipartimento di Fisica Chimica ed Inorganica, University of Bologna, Italy (3) Istituto di Scienza dell’Atmosfera e del Clima (ISAC) del Consiglio Nazionale

delle Ricerche (CNR), Bologna, Italy (4) Forschungszentrum Karlsruhe GmbH, Institut für Meteorologie und

Klimaforschung, Germany (5) Laboratoire Interuniversitaire des Systèmes Atmosphériques (LISA) CNRS/

Univ Paris 12 et 7, France (6) Atmospheric, Oceanic and Planetary Physics – Clarendon Laboratory – Oxford

University, UK (7) Earth Observation Science, Department of Physics and Astronomy, University

of Leicester, UK (8) Instituto de Astrofisica de Andalucia (CSIC), Granada, Spain (9) ESA-ESRIN, Frascati, Italy (10) Astrium GmbH,Friedrichshafen, Germany (11) ABB BOMEM Inc., 585 Blvd. Charest East, Quebéc, Canada (12) DLR, Institute für Meteorologie und Klimaforschnung, Oberpfaffenhofen,

Germany

ABSTRACT

After 2 years of quasi continuously operations (from July 2002 to March 2004), MIPAS on ENVISAT was stopped due to problems in the mirror drive of the interferometer. Operations with reduced spectral resolution and a new measurement scenario were resumed in January 2005. Significant modifications were performed in the ESA operational processor in both the algorithms and the auxiliary data. Performances evaluated on the basis of the first set of available MIPAS measurements in the new operation mode processed with the ESA operational processor are discussed in this paper. The new measurements are characterised by an improved vertical and horizontal resolution and a reduced standard deviation. The analysis of χ2-test statistics indicate that larger mean χ2- values are found in the new operation mode, especially for O3, CH4 and N2O. 1. INTRODUCTION MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) is operating on ENVISAT since March 2002. After two years of nearly continuous limb-scanning measurements (from July 2002 to March 2004) at the end of March 2004 the instrument was stopped due to problems with the mirror drive of the interferometer. Operations with reduced maximum path difference,

corresponding to both a lower spectral resolution (0.0625 cm-1 instead of 0.025 cm-1 ) and a shorter measurement time (1.8 sec instead of 4.5 sec), were tested for one month in August/September 2004 and then definetely resumed on January 2005. In order to exploit the reduction in measurement time a change in the measurement scenario was also implemented for measurements from January 2005 on. The new scenario adopts a finer vertical limb scanning step in the region of the troposphere and lower stratosphere. In order to process the new measurements, changes in both the algorithm and the auxiliary data were needed. While the quite extensive MIPAS database from June 2002 to March 2004 (old operation mode), processed with ESA operational processor ([1] and [2]) , has passed a co-ordinated validation process [see papers in Special Issue ‘MIPAS: potential of the experiment, data processing and validation of the results’, Atmospheric Chemistry and Physics, 2007], a preliminary set of Level 2 data relative to the new operation mode is now available for validation purposes. In this paper performances of the new code are evaluated and a comparison with the measurements in the old operation mode in term of precision and vertical resolution is performed. Ref. [2] contains the description of the performances of the old operation mode that are used as term of comparison.

_____________________________________________________

Proc. ‘Envisat Symposium 2007’, Montreux, Switzerland 23–27 April 2007 (ESA SP-636, July 2007)

Table 1 Characteristics of old and new operation modes OLD OPERATION MODE

Spectral resolution

Vertical sampling: step (km)

altitude range (km)

Horizontal sampling

(km)

Floating altitude

0.025 cm-1 3 km step 6 - 42 km

5 km step 42 - 52 km

8 km step 52 - 68 km

550 NO

Spectral resolution

NEW OPERATION MODE

NOMINAL MODE Vertical sampling Horizontal sampling

(km) Floating altitude

1.5 km step 6 - 21 km

2 km step 21-31 km

3 km step 31-46 km

4 km step 46-70 km

400 YES

UTLS-1 MODE Vertical sampling Horizontal sampling

(km)

Floating altitude

0.0625 cm-

1

1.5 km step 8-21.5 km

2 km step 21.5-27.5 km

3 km step 27.5 - 33.5 k

4.5 km step 290 YES m 33.5-51.5 km

2. DIFFERENCES BETWEEN OLD AND NEW OPERATION MODES The main features distinguishing the new operation mode from the old one are summarised in Table 1 and involve spectral resolution and vertical and horizontal sampling. In order to reduce the risk of a blockage of the mirror drive of the interferometer at the end of the slide, the maximum path difference was reduced from 20 cm to 8.2 cm with a consequent reduction of the spectral resolution from 0.025 cm-1 to 0.0625 cm-1. A reduction in the spectral resolution led to a proportional reduction in measurement time, which was exploited to increase the number of measured spectra in each scan in order to have a finer measurement grid in the upper troposphere and lower stratosphere. In the new nominal mode, the altitude range covered by MIPAS goes from 6 to 70 km, with increasing altitude steps at high altitudes, while in the so called new ‘UTLS-1’ mode the vertical sampling grid is exactly the same as the nominal range up to 27.5 km, slightly coarser above, and the altitude range is restricted to the range 6 - 50 km. The reduction in the measurement time coming from the use of a degraded spectral resolution also resulted in a reduced horizontal spacing between two contiguous limb scan measurements. The new operation modes are characterised by a floating altitude grid. This means that the lowest point of the measured vertical grid is shifted up and down according to a latitude dependent law in order to roughly follow the tropopause height along the orbit. The other points of the limb scanning grid follow the lowest point accordingly. 3. CHANGES IN THE LEVEL 1 PROCESSOR The MIPAS Level 1B algorithms for the high-resolution mode are described in details in [1]. In addition to the change in resolution for the scene measurements, the offset

and the blackbody measurements are obtained at the same resolution as the scene (0.0625 cm-1), in opposition with the high resolution mode where the offset and blackbody were obtained at a resolution of 0.25 cm-1 only. To keep the same NESR0 percentage contribution from the calibration measurements as in the high resolution mode, the L1B processing was changed to reduce by software the resolution of calibration measurements from 0.0625 cm-1 to 0.3 cm-1 (band A to C) and to 1.8 cm-1 (band D only due to noisier detector).

4. CHANGES IN THE LEVEL 2 PROCESSOR The measurements performed in the new operation mode have required changes in both the algorithm and the auxiliary data. In particular, while the use of a reduced spectral resolution has mainly affected the selection of the auxiliary data, the use of a finer vertical sampling grid has required a change in the algorithm itself.

Figure 1 The lower plot shows the major absorbing molecules in the MIPAS spectra for 21 km tangent altitude, the plot in the middle and the upper plot show the spectral location and tangent altitude range of the MWs selected respectively for the old operation mode and the new one.

20

40

60

0 2 4 6 8 0.0 0.5 1.0 1 2 3 4 5

(a)

VMR (ppmv)

Altit

ude

[km

]

(b)

Averaging Kernels

(c)

km

resolution step IFOV

4.1 Changes in the algorithm In the new operation mode the step of the measurement grid in the troposphere and lower stratosphere is smaller than the vertical IFOV (3 km), so the measurements of a limb scanning are no longer fully independent as in the case of the old measurement scenario. The use of a measurement grid significantly finer than the IFOV extension may improve the vertical resolution of the retrieved profile but, if the retrieval grid coincides with the measurement tangent altitudes, as is the case for the MIPAS operational L2 processor, the retrieval becomes ill-conditioned and a regularization is necessary to avoid instabilities in the retrieved profiles. For this reason a regularization was included in the retrieval.

Figure 2 Retrieved ozone profile with retrieval errors (a), averaging kernels (b), vertical resolution (black line) compared with the measurement vertical grid step (red line) and the vertical IFOV (green line) (c). Table 2 Percentage of superposition of the microwindows

PT 5.4% H2O 16.9% O3 11.4%

HNO3 12.8% CH4 9.0% N2O 31.7% NO2 25.5%

Regularization with the ‘error consistency method’ [3, 4] was adopted. It is based on the classical Tikhonov regularisation but uses a retrieval dependent regularization strength calculated with an analytical relationship. Regularization is applied a-posteriori. First the χ2-function is minimized using the Levenberg-Marquardt method, secondly, when convergence has been reached, the regularization is applied. A dedicated Averaging Kernel, computed for each profile, is provided in the product files.

5. IMPROVEMENTS IN VERTICAL RESOLUTION As an example of the retrieval in the new operation mode, Fig. 2 shows the results for O3. The retrieved ozone profile with its retrieval errors (as obtained from the measurement noise) is shown in panel (a), the averaging kernels are shown in panel (b), the vertical resolution (defined as the full width at half maximum of the averaging kernel) compared with the measurement vertical grid step and the vertical IFOV is shown in panel (c).

4.2 Changes in the microwindows The measurements performed at reduced resolution, as well as the change in the measurement scenario, has required a new microwindow (MW) selection. Figure 1 shows the location of the MWs used for the operational retrievals for both the old and the new operation modes and compares them with the contribution of each species to the observed spectrum.

From panel (a) we can see that the code is able to retrieve a non oscillating profile with acceptable retrieval errors. Panel (b) shows that the Levenberg-Marquardt method and the a-posteriori regularization modify the averaging kernels only in the vertical range where the grid step is smaller than the IFOV (as desired). Panel (c) shows that the vertical resolution is about equal to the step of the retrieval grid when this step is larger than the IFOV, while it is smaller than the IFOV when a finer retrieval grid is adopted. This is possible because a weak regularization is applied. Stronger regularizations, as e. g. that produced by the L-curve method, provide a vertical resolution always larger than the IFOV. In the new operation mode the vertical resolution is significantly improved with respect to the old operation mode, for which the vertical resolution is about equal to the step of the retrieval grid, i.e. 3 km up to 42 km and even worst above.

The MWs selected for the new operation mode are significantly different from those selected for the old operation mode. In particular, for pT retrievals some microwindows in the CO2 laser band, that can be affected by Non-LTE, had to be selected in order to have enough information at low altitudes. Table 2 reports the percentage of superposition of the mws for the different species. The mws selected for pT retrieval constitute the extreme case with only 5.4 % of superposition. The maximum of superposition occurs for N2O mws, with 31.7%. Furthermore, the spectral range analysed in the new operation mode is 3-4 times larger than in the old operation mode. The increased spectral range compensates the loss of information content caused by the reduced spectral resolution.

Figure 3 Maps of the retrieval errors for the ozone retrieval of the orbit #7090 (left side), acquired with the old operation mode, and of the orbit #17540 (right side), acquired with the new operation mode.

6. IMPROVEMENTS IN RETRIEVAL ERROR In order to evaluate the performances of the code in analysing measurements corresponding to the new operation mode, Fig. 3 shows the retrieval errors for the ozone retrieval, as a function of altitude and orbital coordinate, for both the orbit #17540, acquired with the new operation mode (plot on the right), and the orbit #7090 (measured on 9th July 2003), acquired with the old operation mode (plot on the left) . The orbital coordinates are linked to latitude and are equal to 0° and 180° at the equator, to 90° at the North pole and to 270° at the South pole. In the new operation mode the retrieval errors are smaller for all the species. This result has been obtained despite the improved vertical and horizontal resolution. Indeed, the distribution of the coloured spots along the orbital coordinate shows that also the horizontal sampling has improved The increased number of spectral points to be simulated, together with the increased number of sweeps for each

scan, has an impact on the computing time needed to process each orbit, that is a factor 5 longer than in the old operation mode. 7. ERROR BUDGET Figure 4 reports the a-priori estimation of the total error budget, including the contributions of both the measurement error and the forward model errors, of the retrieved profiles obtained with the new operation mode, for a retrieval without regularization. We see that the forward model error contribution is comparable with the measurement error for most species. The estimated total error budget for temperature at low altitudes is larger in the new operation mode than in the old operation mode, and this affects the total error budget of the other species, since the pT error propagation is the main contribution to the forward model component of the total error budget. The total error budget of the new operation mode is somewhat larger than the one of the old operation mode, however, when regularization is applied an improvement is observed in the retrieval (see Fig. 3).

Figure 4. Error budgets for the MIPAS retrievals of temperature, pressure, H2O, O3, HNO3, CH4, N2O and NO2 computed for midlatitude daytime conditions. The solid line is the Total Error, represented as the rootmeansquare of the Random Error, shown as the dotted line, and the forward model error, shown as the dashed line. The forward model error is itself the rootmeansquare of the various components shown by different symbols.

8. CHI-SQUARE-TEST STATISTICS The χ2-test of each retrieval, given by the ratio between the square summation of the difference between the observations and the simulations weighted by the measurement error at convergence and the difference between the number of observation and the number of retrieved parameters, is a good diagnostic tool to identify possible problems in the retrieval. pT H2O O3 HNO3 CH4 N2O NO2 Mean χ2- test

2.19 1.34 2.89 1.59 2.36 2.15 1.39

% succ. retrievals

83.2 82.9 82.5 82.7 83.0 82.4 82.9

All measurements in the new operation mode that have been processed so far (both in the nominal and UTLS-1

modes) have been used to compute χ2-test statistics. In particular, the average χ2-test is reported in Table 3. The retrievals from reduced resolution spectra produce χ2 values larger than those from full resolution spectra especially for O3, CH4 and N2O. From Table 3 it also results that a significant percentage of pT retrievals are not successful in the new operation mode. Investigations are on-going. The histograms of the χ2-test values for the different species are shown in Fig.5. The χ2-test of all species are distributed around a constant average value with the only exception of O3, whose anomalous behaviour is under investigation.

0 1 2 3 4 5 6 70

200

400

600

800

pT retrieval

# of

seq

uenc

es

X2-test0 1 2 3 4 5 6 7

0

500

1000

1500

2000

H2O

# of

seq

uenc

es

X2-test

0 1 2 3 4 5 6 70

100

200

300

400

500

O3

# of

seq

uenc

es

X2-test

0 1 2 3 4 5 6 70

200

400

600

800

1000

HNO3

# of

seq

uenc

esX2-test

0 1 2 3 4 5 6 70

200

400

600

CH4

# of

seq

uenc

es

X2-test0 1 2 3 4 5 6 7

0

200

400

600

N2O

# of

seq

uenc

es

X2-test

0 1 2 3 4 5 6 70

200

400

600

800

1000

1200

1400

NO2

# of

seq

uenc

es

X2-test

Figure 5 Histograms of χ2-test distribution for pT, H2O, O3, HNO3, CH4, N2O and NO2 retrievals. 9. CONCLUSIONS The analysis of the preliminary set of level 2 measuremens relative to MIPAS new operational mode confirms that these measurements are characterised by an improved vertical and horizontal resolution and a reduced estimated standard deviation . The mean χ2 values in the new operation mode are generally larger, especially for O3, CH4 and N2O. 10. REFERENCES

1. Raspollini, P. et al., ‘MIPAS level 2 operational analysis’, ACP, 6, 5605-5630, 2006.

2. Kleinert, A. et al, MIPAS Level 1B algorithm

overview: operational processing and characterization’, ACP, 7, 1395-1406, 2007

3. Ceccherini, S., Analytical determination of the regularization parameter in the retrieval of atmospheric vertical profiles, Opt. Lett., 30, 2554-2556, 2005

4. S. Ceccherini et al., Technical Note; Regularization performances with the error consistency method in the case of retrieved atmospheric profiles, ACP, 7, 1435-1440, 2007.