Calibration of MERIS on ENVISAT Status at End of 2002

Transcription

1 Calibration of MERIS on ENVISAT Status at End of 2002 Bourg L. a, Delwart S. b, Huot J-P. b a ACRI-ST, 260 route du Pin Montard, BP 234, Sophia-Antipolis Cedex, France b ESA/ESTEC, P.O. Box 299, 2200AG Noordwijk, The Netherlands, ABSTRACT/RESUME ENVISAT, ESA's ENVIronmental research SATellite, has been launched successfully on 1 March 1st On-board, the MEdium Resolution Imaging Spectrometer MERIS, is responding nominally. The results from the first nine-months of in-flight verification and characterization of the instrument will be presented. The evolution of the instrument radiometric response based on the radiometric calibration results using the on-board diffusers will be discussed. The results from dedicated spectral characterization campaigns based on the spectral features of the sun (Fraunhofer), atmospheric oxygen and Erbium (doped diffuser) will be presented, and the instrument's spectral model derived will be discussed. 1 INTRODUCTION MERIS, is primarily dedicated to ocean colour observations. It has a high spectral and radiometric resolution and two spatial resolutions (1200m and 300m). The global mission (1200m) extends from +80 to -80 degree solar illumination angle and the regional mission (300m) is dedicated to coastal zone waters and land surfaces. 2 REVIEW OF THE INSTRUMENT MERIS is a programmable, medium-spectral resolution, imaging spectrometer operating in the reflective solar spectral range (390 nm to 1040 nm). Its fifteen spectral bands are programmable by ground command both in width and in position by steps of 1.25nm. The instrument scans the Earth's surface using the 'push broom' method where the spectral signal is dispersed to illuminate a 2-D detector array for each scan line. The spectral bands are constructed by first binning spectral samples directly on the array into micro-bands, and further grouping them into bands digitally before transmission to ground (See fig.1). Fig. 1.: Instrument concept The instrument has a field of view of 68.5 o and covers a swath width of 1150 km. The field of view is shared between five identical optical modules arranged in a fan shape configuration. The earth is imaged at a spatial resolution of 300 m (at nadir). The reduced resolution data (1200 m) is computed by the on-board combination of four adjacent samples across track over four successive lines. The low rate reduced resolution (RR) data will be acquired systematically, while the high rate full resolution (FR) data will be recorded in parallel according to user requests for a maximum duration of 20 minutes per orbit. Proc. of Envisat Validation Workshop, Frascati, Italy, 9 13 December 2002 (ESA SP-531, August 2003)

2 Fig. 2.: Optical concept Calibration of MERIS is performed at the orbital South Pole where diffuser plates are deployed by rotating a selection disk (See fig. 2) into any of the five positions described below: Shutter: used for dark calibration as well as for protecting the instrument from contaminants, Earth observation: a diaphragm is introduced in the field of view. Radiometric calibration: The sun illuminates a white diffuser plate inserted in the field of view Diffuser degradation monitoring: a second white diffuser plate is deployed every 3 months to monitor the degradation of the frequently used plate. Spectral calibration: An Erbium doped Pink diffuser is deployed with MERIS configured to sample its absorption features at highest spectral resolution. 3 CALIBRATION OF MERIS 3.1 Radiometric Calibration The radiometric calibration of MERIS is performed with the use of the two on-board sun-lit radiometric (white) calibration diffuser plates described above. The diffuser plates, made of Spectralon TM, provide a uniform illumination signal over the large field of view of the instrument BRDF (sr-1) Angle in FOV (degrees) Azimuth angle (degrees) Fig. 3.: Diffuser plate 1 BRDF at 410 nm for four illumination conditions corresponding to different times throughout the year. The calibration plates have been extensively characterised on-ground using a dedicated BDRF bench (See fig.3) to an absolute accuracy of better than 1%. This performance estimate was confirmed by a round-robin exercise performed with other laboratories (i.e. NIST, NASA, NPL).

3 The diffuser plates have been exposed to a post-production processing in order to reduce the degradation (browning) of its scattering characteristics to space environment. According to on-ground simulations, the degradation over the mission lifetime should be minimal. However, as a means of verification, MERIS will make use of both its on-board diffusers to monitor the degradation of the frequently used (every 15 days) diffuser-1 by comparing it with the results from diffuser-2 which will be deployed only every 3 months. Once BRDF stability of the frequently used diffuser has been assessed, comparisons of successive calibration results allows to characterise instrument degradation and hence the frequency at which calibration coefficients should be updated to ensure sufficient quality of the MERIS Level 1b products Diffuser Stability As a mean of determining the degradation of the frequently used diffuser 1, the instrument is used as a transfer radiometer to compare measurements of diffuser 1 and diffuser 2 made at consecutive orbits, hence with almost identical illumination conditions. Comparing such measurement made at different times allows to determine the degradation of the frequently used diffuser w.r.t. the reference diffuser. Fig. 4.: Relative BRDF difference between the two radiometric diffusers: from on-ground characterization (left) and from in-flight measurements on 22 March 2002 (right). Difference of Dif1/dif2 ratios [counts] from March 22 (1st Dif2 - Drift Orbit) to July 15 (2nd Dif2) Dif 1 Exposures dif. & 1 Dif 2 Exposures dif. Difference of Dif1/dif2 ratios [counts] from July 15 (2nd Dif2) to October 2 (3rd Dif2) Dif 1 Exposures dif. & 1 Dif 2 Exposures dif. 0.40% 0.20% 0.00% CamAll Cam1 Cam2 Cam3 Cam4 Cam5 0.40% 0.20% 0.00% CamAll Cam1 Cam2 Cam3 Cam4 Cam5-0.20% -0.20% -0.40% % Bands Bands Difference of Dif1/dif2 ratios [counts] from March 22 (1st Dif2 - Drift Orbit) to October 2 (3rd Dif2) Dif 1 Exposures dif. & 2 Dif 2 Exposures dif. 0.40% 0.20% 0.00% CamAll Cam1 Cam2 Cam3 Cam4 Cam5-0.20% -0.40% Bands Fig. 5.: Difference between the ratios of diffuser 1 / diffuser 2 [counts] (top left: 22 March 15 July, top right: 15 July 3 October, bottom: 22 March 3 October)

4 The comparison of the calibration results using the two diffusers showed a near identical BRDF behaviour between diffusers, as expected from on-ground characterisation. The differences seen lead us to believe that a residual nonuniformity in BRDF across the extent of the diffuser plates is involved - note that the optical axis of each cameras is centred on the diffuser plate - and that this pattern depends slightly on the scattering angle (see: Pixel in Fig 4- left) The first such comparison - March 22 nd an July 15 th - has shown no significant degradation of the frequently used diffusers 1 (see fig. 5). These results have also shown that the two diffusers have a very similar behaviour in BRDF, even for the large differences in illumination conditions (12 deg. in azimuth) encountered due to the early mission orbit manoeuvres which resulted in a BRDF[counts] difference (March-July) varying smoothly across the field of view up to 6% depending on the wavelengths. A third ageing calibration took place on October 2 nd and confirmed the above results and seems to show a slight browning effect (see fig. 5, bottom), less than 0.2 % however. Next ageing calibration, planned for late December, should give further evidence Radiometric calibration results The first on-board calibration took place on orbit 308, March the 22 nd. Raw counts are displayed on fig. 6. Calibration data have been processed as soon as retrieved and corresponding auxiliary data files were ready the next day. Fig. 6.: First radiometric calibration results [counts] Instrument response temporal evolution The evolution of the instrument response for the period April to August, based on diffuser 1 measurements and computed using a BRDF model based on on-ground characterization, has shown some degradation (<2.5%) of the blue bands for most of the cameras, and the presence of a smoothly varying (across the field of view) gain term whose curvature depends on the differences in the illumination conditions between measurements. As can be seen in fig. 7, the curvature of the smoothly varying gain difference term changes signs between the April- June and the April-September comparisons and almost vanishes for the April-end of July comparison. Right part of the figure also shows that this curvature is well correlated with the change of slope of the difference between the BRDF used for the gain computations. Fig. 8 displays the diffuser plate illumination conditions for each calibration. An obvious correspondence can be established between the solar azimuth of a given calibration relative to the orbit 846 reference and the slope of the corresponding BRDF relative difference plotted in right column of fig. 7. This leads us to believe that this effect is most likely due to residual errors in the BRDF model, and could be due to the fact that the on-ground characterisation was performed at ambient conditions and not in vacuum.

5 April 29 th to June 25 th (orbits 846 to 1667) April 29 th to July 8 th (orbits 846 to 1858) April 29 th to July 25 th (orbits 846 to 2086) April 29 th to Aug 18 th (orbits 846 to 2444) April 29 th to September 19 th (orbits 846 to 2899) Fig. 7.: Instrument gain evolution over the period May to September Figure shows, for 5 calibrations spread over 3 months, gain (left) and diffuser BRDF (right) evolution relative to a reference (April 29 th calibration). Each series number in the legends refers to the corresponding band.

6 Solar Illumination on Diffuser Solar Zenith Zenith Azimuth Solar Azimuth Orbits Fig. 8.: Solar illumination conditions for the various calibrations available to date. Solar azimuth values for orbits 846 and 2086 are highlighted. Another important feature of these plots is that the curvature of the gain variation curve does not show significant spectral dependency. This property allows to derive a first estimate of the instrument degradation by normalisation with respect to one of the bands with the highest wavelength, at which the instrument degradation is expected to be minimum. This has been done on all available calibrations (from March to October 2002) using band 15 as reference and results are shown in fig. 9. The earliest calibration (March 22 nd ) has obviously been taken as starting point (i.e. no degradation), however due to later changes in the instrument band and gain settings, direct comparisons between data acquired prior and after orbit 846 are not possible. Some assumptions had to be used that increase uncertainties on the derived degradation (see e.g. the unrealistic slight negative degradations for camera 4) and may distort the overall temporal behaviour. Assessment of degradation will be improved by additional calibration acquisitions using the old band and gain settings. Even if they shall be regarded as very preliminary, present results clearly show two main points: degradation affects mainly the blue bands (412 and 442 nm) and varies highly from camera to camera. Most affected are cameras 2, 5 and, to a lesser extent, 3, with a maximum degradation of about 2.5% at 442 nm for camera 2. The BRDF model will be improved gradually as calibration data come in. This will be achieved by assuming a known degradation derived as described above, and thus considering the smoothly varying gain difference term as a correction to the BRDF model. If necessary, the process will be repeated with feedback from upgraded BRDF model to degradation determination and vice-versa. The improved model will be tested on the second year of calibration data to verify the validity of its assumptions.

7 Fig. 9.: Instrument mean temporal degradation with respect to band 15 for the five MERIS cameras Fine scale gain variations Gain evolution plots in fig. 7 also show temporal variations at high spatial frequencies, apparently randomly distributed with an amplitude increasing with wavelength and azimuth difference. These high frequency gain variations also appear on a smaller time scale (33 days instead of 100) as shown in fig. 10. This phenomenon, apparently constant with time, cannot be linked with instrument or diffuser degradation and, considering its strong correlation with the variation of solar azimuth, cannot be considered as random noise. It is very similar to a speckle effect and is believed to be a diffuser BRDF property. A possible explanation could be small surface irregularities of the Spectralon plate highly sensitive to illumination direction. Further study of this speckle effect is planned within the improvement of the BRDF model. This study will also include diffuser observations acquired in the Observation Mode of the instrument, i.e. with a time sampling step of 176 ms but with a reduced spatial resolution. Such data sets contains an almost continuous variation of the illumination angles.

8 Reference orbit: 846 (28 Apr.) Compared orbits: 1858 (8 Jul.): φ = deg (24 Jul.): φ = deg (7 Aug.): φ = deg. Reference orbit: 3068 (30 Sep.) Compared orbits: 2697 (4 Sep.): φ = deg (19 Sep.): φ = deg (7 Oct.): φ = deg. Fig. 10.: High spatial frequency relative gain variations observed on two different time scales: within 10 days (left) and 33 days (right), for 3 bands: 510 (top), 709 (centre) and 900 nm (bottom) Dark current / Offset Stability The orbital variations of the dark current are expected to be small as the detector array is maintained at a temperature of 22 +/- 0.5 by a Peltier cooler. At this temperature the theoretical number of dark current electrons is much lower than the electronic noise of the detector s output amplifier.

9 Fig. 11.: Dark current offset stability along an orbit (43.5 min) The variation in dark current / offset during observation was measured early in the mission (See fig. 11). It shows insignificant variations of the dark current /offset over the complete 43.5 minutes of nominal observation. Fig. 12.: Dark current / offset stability from March to August 2002 Fig. 12 shows dark current / offset measurements over the first six months of operations. It can be seen that although the dark current /offset is slightly different between the cameras, it has remained perfectly stable over the period Vicarious Calibration Validation of Top of Atmosphere (TOA) radiance measured by MERIS is achieved by comparison against TOA radiance obtained by the following vicarious calibration methods: Simultaneous in-situ measurements of natural target, (absolute) Stable deserts sites, (relative: multi-sensor, multi-temporal, multi-angular) Rayleigh scattering over clear water, (absolute) Sun glint (relative: inter-bands) The geographic distribution of the vicarious calibration sites can be seen in red in the fig. 13. Stable deserts sites are mainly located in North Africa and Arabia, and the Rayleigh scattering and the sun glint sites are distributed over most oceans. Both desert and ocean sites have been selected for there radiometric stability over long periods of time.

10 Fig. 13.: Location of vicarious calibration sites. Data acquired over these calibration sites are systematically processed to monitor continuously the evolution of the instrument response, this will allow for an independent verification of the calibration results based on the on-board diffusers. Preliminary results will presented during the ENVISAT Validation Workshop by the AO PIs and will not be reported here Inter-comparison of MERIS and AATSR MERIS and AATSR, both on board ENVISAT, have three bands with sufficiently close central wavelengths to allow direct comparisons of top of atmosphere reflectance. The Nadir view of AATSR can be co-located with MERIS data if restricted to few kilometres at centre of swath to avoid viewing geometry discrepancy due to the conical scanning of AATSR. To account for the slightly different resolutions of the two instrument, data have been aggregated in macropixels maximising the overlap and compared on a statistical basis. Each macro-pixel is centred on the satellite nadir and sizes 20 by 29 km (across-track by along-track). The top of the atmosphere reflectance measured simultaneously by MERIS and AATSR can be seen in fig. 15 for orbits 2136 (July 28 th, top) and 3357 (October 21 st, bottom). Geographic coverage of both comparisons are shown on fig. 14. Fig. 14.: geographical coverage of MERIS and AATSR for orbits 2136 (left) and 3357 (right). AATSR swath is shown in light blue.

11 Fig. 15.: Comparison of MERIS and AATSR Top of Atmosphere Reflectance on July 28th (top) and October 21st (bottom). Plots left to right for each orbit show scatter plots of reflectance at 560 (left), 665 (centre) and 865 nm (right). Error bars denotes standard deviation of data within each macro-pixel. These results show that both instruments, which are calibrated using different diffusers types, compare very well (negligible bias, deviation < 6%). It must however be noted that these are only preliminary results, that need to be confirmed statistically over a much larger data set. Furthermore, greater attention should be paid to calibration coefficients used for both instruments for each comparison. Note: data points neatly departing from the overall trend in bottom right plot are due to saturation over very bright clouds in MERIS: the 865 nm band has been tuned with a relatively high gain to fulfil the Ocean Colour algorithms requirements. Those pixels will be filtered out in future comparisons. However, due to limited amount, they should not have a significant impact on the statistics. 3.2 Spectral Calibration Knowledge of the central wavelength of the MERIS bands for each pixel is necessary for the correct estimation of the in-band solar irradiance used within the radiometric calibration process. In addition, band 11 dedicated to O2 absorption measurement has even more constraining spectral characterisation requirements to fulfil the atmospheric pressure estimations algorithms needs. The on-board calibration mechanism includes an Erbium doped SpectralonTM diffuser plate dedicated to spectral characterisation, which offers a number of spectral absorption features in the spectral range of MERIS. The absorption peaks selected are those centred at 409 & 522 nm (see Fig. 16).The spectral calibration is performed in two steps. First, the instrument is configured to have 15 adjacent bands centred on the spectral feature, the pixels are calibrated using the radiometric "white" calibration diffuser, the following orbit the pink diffuser is deployed and the signal from this spectral feature acquired. Processing on ground determines the position on the detector array of the centre of the Erbium spectral feature. This is repeated periodically for both spectral features in order to determine whether the spectrometer s dispersion law is varying throughout the mission.

12 Reflectance Erbium doped diffuser spectra Wavelength (nm) Fig. 16.: Erbium doped Spectralon TM Diffuser reflection spectrum. In addition, a number of specific Earth observation campaigns have been dedicated to spectral calibration at the mission beginning (March and April) and with particular attention given to wavelengths around the molecular oxygen absorption band (760 nm). These measurements include: Observation of the O2-A absorption feature at 760 nm using two independent wavelength derivation methods (pressure homogenisation at LISE, neural net at FUB). Observation of Sun s Fraunhofer lines circa 393, 484 and 853 nm. The resulting data set, combined with the results of the on-board spectral calibration hardware, proved to be fairly coherent and has been used to derive the instrument spectral model described below. A second set of spectral calibration campaigns took place in November including O2-A and Fraunhofer observations, using optimised band settings and including additional Fraunhofer lines observations circa 585, 656 and 866 nm. Analysis of corresponding data confirms the validity of the spectral model mathematical expression. The first set of parameters, while under revision to account for the news data, remains for all applications except O2 absorption. Further investigations are planned however to better understand the temporal variations of the O2 band spectral characterisation, believed to be due to improved radiometric calibration of the observations. As a matter of fact, due to operational constraints at this early stage of the mission March 2002, it was not possible to operate an on-board radiometric calibration sequence with the O2 dedicated band setting Spectral Calibration Results The combination of the results from all the methods mentioned above allows us to precisely characterize the spectral dispersion law of each of the 5 MERIS spectrometers over almost all the useful spectral range and confirms the expected spectral slope (smile) across the field of view for 4 of the 5 optical modules (see fig. 17). wavelength In-flight spectral calibration detector line 15 (ca 409nm) line 297 (ca 760nm) shift (nm) Spectral shift within FOV: deviation wrt mean, module detector Fig. 17.: Spectral calibration for all camera s field of view. The spectrometer's smile has a very linear behaviour across the field of view of all cameras and this variation appears to be independent of the absolute wavelength (See fig.17). Therefore a relatively simple and robust model could be derived for each spectrometers, considering separately a mean dispersion law and an across-track deviation term.

13 λ( kl, ) = λ( l) + λ( k) where: k and l stand for the spatial and spectral co-ordinates of a given detector, respectively and, λ () l λ( k), mean dispersion law mainly linear is a polynomial of order 5 (best fit),, the across-track variation term, is a linear fit of the data at 760 nm expressed relative to its mean value. The mean dispertion law is shown on fig. 18 with a) spectral slope and intercepts for each module (right) and b) the deviation from the linear trend as a function of wavelength (left). Spectral models: linear trends Spectral model: mean deviation from linear slope (nm) module slope intercept 390 intercept (nm) nm CCD rows Fig. 18.: Spectrometer model: linear slope and intercept, deviation from linear. m1 m2 m3 m4 m5 The model of the grating's theoretical dispersion law was used to fit the deviation from the linear trend. This model was selected to better constrain the fit in the blue part of the spectrum. As can be seen from fig. 19, the main contribution to the residual error of the spectral model is due to the deviation between the model and the measurements in the blue. residual (nm) spectral models: residual error detector Fig. 19.: Residual error (measured data model) Fig. 19 show the residual error of the model versus the measurements for all the different derivation methods used. Results from the Fall campaigns, not yet used within the modelling process, show a very good quality, thanks to O2 data radiometric calibration and band setting optimisation. Data are now available for 6 wavelengths within MERIS range acquired over a very short period. Fig. 20 show two examples of the new campaign results: wavelength of the O2 band circa 760 nm derived by the two methods (so-called LISE and FUB), that show an excellent agreement and very low noise, and results from a Fraunhofer line around 485 nm.

14 485 (row 76) O2 (row 297) wavelength (nm) shift (nm) LISE FUB pixel pixel Fig. 20.: examples of results from the Fall spectral campaigns (left: Fraunhofer line around 485 nm, right: O2 band around 760 nm derived by the 2 methods) The entire set of results, expressed as spectral shift with respect to the module mean, is plotted on fig. 21. It confirms the coherency between the different methods and the approach used for the spectral model derivation. Spectral shift wrt module means shift (nm) detector Fig. 21.: Fall spectral campaigns results expressed as spectral shift with respect to per module means Spectral Stability The spectral behaviour of the instrument is monitored regularly by on-board spectral calibrations using the Erbium doped diffuser plate. Fig. 22 show the location on the CCD of the 522 nm Erbium spectral feature as measured in April and July. It shows a stability better than 0.02 nm as average and 0.1 nm peak. Fig. 22.: Erbium doped diffuser measurements (522 nm) April and July. Analysis of the common wavelengths from the Spring and Fall spectral campaign data sets globally confirms this level of stability, as shown on fig. 23, once measurement quality upgrade through spectral observation window enlargement

15 (393 and 485 nm) is taken into account. However some features must be further analysed like the almost systematic shift of module 2, seen at 393 and 760 nm as well as in Erbium doped diffuser data at 522 nm. 393 (row 3) 485 (row 76) wavelength (nm) Fall Spring wavelength (nm) Fall Spring pixel pixel 761 (row 297) 853 (row 371) wavelength (nm) Fall Spring wavelength (nm) Fall Spring pixel pixel Fig. 23.: Spectral campaigns measurements (393, 485, 761 & 853 nm), Spring and Fall. 4 CONCLUSION AND RECOMMENDATIONS The calibration of the instrument is expected to yield results of unprecedented accuracy. The highly stable design of the instrument, coupled with the system described above will guarantee users data of the highest quality for the complete lifetime of MERIS. However, a number of tasks have been identified that are necessary to achieve these goals. Among those are: Improvement of diffuser BRDF modelling, including diffuser ageing Study speckle effect Repeat spectral characterisation campaigns at least every year Include a pixel-wise degradation model in the Level 1b processing Continue systematic acquisition over vicarious calibration sites Hold regular Cal/Val workshops ACKNOWLEDGEGMENTS A very special acknowledgement is made to all members of the ENVISAT team involved in this complex task.