Wavelength calibration of spectra measured by
the Global Ozone Monitoring Experiment by use
of a high-resolution reference spectrum

Jos H.G.M. van Geffen and Roeland F. van Oss
Applied Optics  42, 2739-2753 (2003).

1. Introduction

The Global Ozone Monitoring Experiment (GOME) aboard the ERS-2 (European Remote Sensing) satellite is a nadir-viewing ultraviolet and visible spectrometer. It measures the spectra of sunlight scattered in the atmosphere and reflected by the surface of the Earth. GOME measures in a wide wavelength range: about 240 to 790 nm, divided in four spectral channels (see Table 1). Once a day direct solar spectra are measured for radiometric calibration. The ERS-2 satellite, launched on 21 April 1995 by the European Space Agency (ESA), orbits Earth in a sun-synchronous polar orbit at an altitude of about 790 km. Each orbit takes about 100 minutes, and ERS-2 completes 14 orbits a day. The satellite crosses the equator going north to south at 10:30 a.m. local time.
Table 1 Wavelength ranges of the four detector channels of the GOME instrument, the approximate detector pixel size and the spectral resolution of GOME for each channel. The last column shows the number of data points (detector pixels) for each channel in GDP level-1 spectra.

channel wavelength range pixel size resolution #points
1 237 - 315 nm 0.11 nm 0.17 nm   695
2 313 - 405 nm 0.12 nm 0.17 nm   841
3 407 - 608 nm 0.22 nm 0.30 nm 1024
4 599 - 794 nm 0.22 nm 0.35 nm 1024
GOME is equipped with a scan mirror that redirects the backscattered light onto its optical bench. The mirror performs a 4.5-second forward scan from east to west, followed by a 1.5-second backscan from west to east. The largest off-nadir scan angle is about 30°, giving a full swath width of about 960 km at ground level. With this scanning mode, global coverage is achieved in three days. The forward scan is divided into three 1.5-second scan pixels: east, centre (or: nadir) and west (where the centre pixel is directly below the satellite and east and west are relative to centre). In the direction of flight, each ground pixel measures 40 km. For more information on the GOME instrument, the reader is referred to Ref. 1-3 and references therein.

The spectra of scattered or reflected sunlight, the earthshine spectra, contain information on the distribution of various constituents of the Earth's atmosphere. GOME's main objective is to monitor the distribution of ozone. Vertical ozone profiles, for example, can be derived from the spectra at wavelengths between 265 and 330 nm, and total ozone columns from the range 320-340 nm. Cloud fraction and cloud top pressure for the scene of the ground pixel -- information necessary for the retrieval of ozone, as clouds shield part of the ozone column from the satellite -- are derived from oxygen absorption features in the earthshine spectra, notably the oxygen-A band centred around 762 nm. Apart from ozone, the spectra contain information on a number of other trace gases, such as NO2, SO2, and BrO, as well as of aerosols. Measuring the concentration of these trace gases and their distribution, both horizontally and vertically, is essential for understanding variations in the atmospheric composition, chemical reactions and dynamical processes taking place in the atmosphere and, at a higher level, an understanding of climate change.


1.1 GOME Fast Delivery Service

Solar and earthshine spectra, the so-called level-1 data, are the official product release of the GOME Data Processor (GDP) of the ESA Processing and Archiving Facility at DLR (Deutsches Zentrum für Luft- und Raumfahrt) in Oberpfaffenhofen, Germany, together with some level-2 data, e.g. the total column of ozone and NO2, and cloud information.[1,4,5] These data are released on CD-ROMs and via the Internet between two weeks and two months after observation.

For monitoring specific phenomena that change quickly with time (such as the "ozone hole" that appears over Antarctica in the austral spring) and special short-term events (such as "ozone mini-holes"; see e.g. Ref. 6), as well as for ozone data to be useful for improving numerical weather forecasts and for validation campaigns, it is essential that the ozone data are available on a near-real time basis: within 3 to 6 hours after observation.

To this end, a GOME Fast Delivery Service (FDS) has been set up at the Royal Netherlands Meteorological Institute (KNMI), which provides total ozone columns, vertical ozone profiles, assimilated ozone fields, UV index forecasts, and cloud data to users via the World Wide Web (http://www.knmi.nl/gome_fd/). To make the FDS possible, KNMI receives via ftp the part of the raw (level-0) GOME spectra that is available in near-real time: the so-called Extracted GOME Instrument header (EGOI) data, used by ESA to monitor the status of the GOME instrument.[7,8]

The raw data in these EGOIs contain the spectral measurements of nine selected wavelength windows, listed in Table 2, as well as all instrument health parameters, such as system temperatures, and polarisation information. The nine wavelength windows are selected to monitor the detector's status and to provide data to retrieve ozone columns and vertical ozone profiles with sufficient accuracy. Note that detector channels 1 and 2 are divided into two bands ('a' and 'b') for earthshine spectra. The wavelength of the division point between bands 1a and 1b has been changed once since the launch of ERS-2: initially the division was around 307 nm, since June 1998 it is around 283 nm. Band 1a has an integration time of 12 seconds, to improve the signal-to-noise ratio of the spectra. All other bands have an integration time of 1.5 seconds. This implies that band 1a actually comprises two full scans, each with three forward and one backscan pixel, and that the measurement corresponds to a ground pixel of about 960 by 100 km.

Table 2 Definition of the nine EGOI windows since June 1998, used for the near-real time Fast Delivery Service at KNMI as outlined in the Introduction. The fourth column lists the number of data points (detector pixels) in each window, which totals to 445.

window wavelength range width #points band
1 272.16 - 275.91 nm   3.75 nm   35 1a
2 282.93 - 285.55 nm   2.62 nm   25 1b
3 292.51 - 302.96 nm 10.45 nm   98 1b
4 305.31 - 307.87 nm   2.56 nm   25 1b
5 311.92 - 314.46 nm   2.54 nm   25 1b
6 323.13 - 336.22 nm 13.09 nm 116 2b
7 351.64 - 352.76 nm   1.12 nm   11 2b
8 372.32 - 373.42 nm   1.10 nm   11 2b
9 758.00 - 778.50 nm 20.50 nm   99 4  
EGOI window 6 is the "ozone DOAS window" (Differential Optical Absorption Spectroscopy) from which the total ozone column can be retrieved.[7,9] Windows 1 through 6 are used for the retrieval of ozone profiles.[10] The measurements in windows 7 and 8 are used for the correction of the polarisation sensitivity of the instrument. Window 9 contains the oxygen-A band, which is used for the retrieval of cloud fraction and cloud top height. In FDS, the latter is done with FRESCO (Fast Retrieval Scheme for Cloud Observables[11,12]).


1.2 Wavelength calibration

Starting from the raw data sent to Earth by GOME, a number of steps have to be taken to produce spectral data (level-1) from which useful information can be retrieved.[13] One of these steps is a wavelength calibration, linking each detector pixel to a specific wavelength. The standard wavelength calibration of the GDP uses spectral lines of an onboard PrCr/Ne hollow cathode lamp to correct for shifts in the wavelengths associated with the detector pixels, for example caused by temperature effects, with respect to a pre-flight calibration. The calibration lamp provides a large set of emission lines throughout GOME's wavelength range[1,2], though the coverage is sparse in some parts of the spectral range.

The calibration method has proved to be insufficiently accurate for the retrieval of ozone profiles, which requires an accuracy in the wavelength calibration of about 0.002 nm, and the accuracy of the DOAS-column retrieval of trace gases also improves with a more accurate wavelength calibration, as shown by sensitivity studies done for the Ozone Monitoring Instrument (OMI).[14] Similar sensitivity studies have not been performed for GOME, but the results of the studies for OMI and some simple test-case studies have shown that the quality of most retrieved level-2 products can be expected to improve with an improved wavelength calibration. Since the retrieval depends on several factors, it is not possible to give an quantitative statement on the improvement a good wavelength calibration makes.

Apart from the observed insufficient accuracy of the GDP calibration method, there are indications that the calibration lamp is malfunctioning, and it may therefore be necessary to switch it off permanently. For these reasons, a new calibration method has been developed for the FDS. The method -- which is described in Section 2 -- uses as reference spectrum a high-resolution solar spectrum, with irradiance values given at 0.01-nm intervals. This spectrum[15] has been obtained from ground-based and balloon-based measurements with an accuracy of 0.001 nm above 300 nm and 0.002 nm below that.

To better match GOME observations against this reference spectrum, the latter is convolved with GOME's slit function and subsequently integrated over the spectral bins of the detector. The location and width of these bins are allowed to vary along the detector: both a shift and a squeeze are applied to the measurements (the GDP's calibration method with the lamp lines applies only a shift). Starting from an initial guess of the wavelength grid, each of the nine EGOI windows is calibrated separately. The method is akin to the chi-square minimisation of a merit function investigated by Caspar and Chance[16], who used a general least-square fitting procedure. The method described in the present paper employs a tailor-made chi-square minimisation, and applies both a shift and a squeeze.

Section 3 discusses some important issues regarding the calibration, such as the accuracy of the calibration and the required signal-to-noise ratio of the spectra, and looks briefly at the variation of the calibration results along an orbit and from orbit to orbit.

The calibration method was initially specifically designed for use with the EGOI windows in the level 0-to-1 processor of the FDS. It can, however, be used also to re-calibrate the wavelength grid of GDP level-1 spectra, as the method can be applied to any wavelength window within the range of the reference spectrum. For this reason, the method has been implemented as part of a software package called GomeCal, which also includes an improved polarisation correction as well as a re-calibration of the reflectivity and a correction for the degradation of the GOME instrument. The package, which is introduced in Section 4, is made available via the World Wide Web (http://www.knmi.nl/gome_fd/gomecal/).

Finally, some concluding remarks are given in Section 5.

<=== contents of this paper
===> references

<=== Publications page
<=== Post-doc. research at KNMI page

Jos van Geffen -- Home  |  Site Map  |  Contact Me

created: 11 February 2003
last modified: 3 February 2008