Modeling of the ageing effects on Meteosat First Generation Visible Band

Transcription

1 on on Meteosat First Generation Visible Band Ilse Decoster, N. Clerbaux, J. Cornelis, P.-J. Baeck, E. Baudrez, S. Dewitte, A. Ipe, S. Nevens, K. J. Priestley, A. Velazquez Royal Meteorological Institute of Belgium (RMIB) Satellite Application Facility on Climate Monitoring (CM-SAF) Vrije Universiteit Brussel (VUB) MeteoClim PhD Symposium on Meteorology and Climatology November 5, 2010, RMI

2 Outline on

3 Background on PhD connected with the Department of Electronics and Informatics (ETRO) of the Faculty of Engineering Sciences at the Vrije Universiteit Brussel (VUB) Part of the GERB team at the RMI Belgium, supported by Climate Monitoring SAF (CM SAF) Meteosat satellites in geostationary orbit at 0 longitude Meteosat-1 to -7 are part of Generation (MFG), containing only a narrow band (NB) imager, called MVIRI Since 2004, the Meteosat Second Generation (MSG) satellites carry next to the narrow band imager SEVIRI also a broad band (BB) instrument called GERB

4 Background (2) on In a geostationary orbit, GERB measures the Earth Radiation Budget through two broad band channels GERB team at RMI does operational work on the data coming from the GERB instrument Next to doing this operational work, the GERB team here also has experience in creating GERB-like data from SEVIRI through a NB-to-BB conversion for periods in time where the GERB instrument failed Use this experience to work on the generation of GERB-like data from MVIRI from Generation satellites

5 Background (3) on Only a narrow band imager onboard of the satellites: Meteosat Visible and Infrared Imager (MVIRI) Visible channel Infra red channel Water vapour channel Normalised radiance Wavelength (micrometer) Figure: Normalised spectral response curves for MVIRI channels.

6 Background (4) on Figure: Operational time for Meteosat satellites. To create GERB-like data for MFG satellites, we will use overlap between MVIRI data from last MFG satellite without GERB instrument (Meteosat-7) and SEVIRI data from the first MSG satellite with GERB instrument (Meteosat-8).

7 Background (5) on However, MFG satellites seem to have bigger ageing problem than MSG satellites First need to try to correct for this Only visible (VIS) band data used from: (1982/02/ /05/12) Meteosat-7 (1998/06/ /06/14)

8 Outline on

9 MFG detectors and optics on Degradation of the silicon detectors and the mirror optics leads to decrease in spectral response of the radiometer which can be seen as a decreasing trend in the time Need to remove the trend by correcting for the ageing effect Reflectance ratio Clouds / / / / / / / /01 Time Figure: Cloud time for Meteosat-7.

10 MFG detectors and optics (2) on EUMETSAT uses a constantly increasing calibration coefficient in time to correct for ageing. For Meteosat-7 this leads to an increase of about GOVAERTS et al.: OPERATIONAL 2.2% CALIBRATION per year OF THE METEOSAT (Results RADIOMETER VIS from BAND Y.Govaerts et al. 2004) 1907 Figure: Calibration coefficient Meteosat-7 (Govaerts et al. 2004). Fig. 7. (Top) Meteosat-7 VIS band calibration coefficients derived during each application cycle of SSCC over desert target ( red symbols) and sea ( blue symbols). The estimated error is shown with the vertical bars. The red (blue) dashed line indicates the estimated linear sensor drift over desert and sea targets. The black symbol is for the calibration coefficients estimated at the launch time, with the associated error shown with the black vertical bars. (Middle) Mean offset value in digital count ( red symbols) and associated error (red vertical bars). The retrieved offset value is shown with the green symbol. (Bottom) The probability is shown with the blue symbol, the probability with the red symbol, and the overall quality indicator is shown with the green symbol. In my work the calibration coefficient will be kept constant (value at launch) and the possibility of a non constant decrease in time will be investigated. surface and atmosphere state variable error. The residual random error is close to 1.5%, and the total error is now dominated by RTM and NSR uncertainties. For each processed ten-day period, calibration coefficients are derived with an estimated error of about 6% for Meteosat-7. The Meteosat-5 displacement from 0 to 63 had no significant impact on the butis similar inmagnitude to thecontribution of the NSRuncertainty for Meteosat-7. As a result, the total error after the temporal averaging is principally controlled by this latter contribution. Sea surfaces being very uniform, whatever the search area, the spatial averaging minimally reduce the total error. For Meteosat-7, the calibration coefficient derived at

11 Outline on

12 on Requirements: Both cloudy and clear sky time Clear sky images were created every 10 days through a pixel to pixel analysis of a of 30 images before and 30 images after the original one Clear sky time for different scene types Scene types used: bright vegetation, dark vegetation, bright desert, dark desert and ocean Time as constant as possible Look for stable sites: stable clear sky sites have lowest standard deviation in the total of images stable cloudy sites are chosen amoungst the highly convective clouds, so the highest reflectance values Averageing was done in space

13 (2) on Time are expressed in reflectance ratio: Value of original images in counts [DC] Radiance obtained using a constant calibration: rad = calibration * (value - offset) [W /(m 2 sr)] Reflectance obtained as: refl = rad / (irr * cos(sza) * π * (dist) 2 ) Reflectance ratio obtained by dividing the reflectance with a simulated reflectance and multiplying with a narrow band (NB) to broad band (BB) correction NB to BB correction through a NB to BB fit of simulated reflectance values: refl BB = a + b refl NB

14 (2) on Time are expressed in reflectance ratio: Value of original images in counts [DC] Radiance obtained using a constant calibration: rad = calibration * (value - offset) [W /(m 2 sr)] Reflectance obtained as: refl = rad / (irr * cos(sza) * π * (dist) 2 ) Reflectance ratio obtained by dividing the reflectance with a simulated reflectance and multiplying with a narrow band (NB) to broad band (BB) correction NB to BB correction through a NB to BB fit of simulated reflectance values: refl BB = a + b refl NB

15 (2) on Time are expressed in reflectance ratio: Value of original images in counts [DC] Radiance obtained using a constant calibration: rad = calibration * (value - offset) [W /(m 2 sr)] Reflectance obtained as: refl = rad / (irr * cos(sza) * π * (dist) 2 ) Reflectance ratio obtained by dividing the reflectance with a simulated reflectance and multiplying with a narrow band (NB) to broad band (BB) correction NB to BB correction through a NB to BB fit of simulated reflectance values: refl BB = a + b refl NB

16 (2) on Time are expressed in reflectance ratio: Value of original images in counts [DC] Radiance obtained using a constant calibration: rad = calibration * (value - offset) [W /(m 2 sr)] Reflectance obtained as: refl = rad / (irr * cos(sza) * π * (dist) 2 ) Reflectance ratio obtained by dividing the reflectance with a simulated reflectance and multiplying with a narrow band (NB) to broad band (BB) correction NB to BB correction through a NB to BB fit of simulated reflectance values: refl BB = a + b refl NB

17 (3) on Obtain 6 time : 1 for cloudy sky and 5 for clear sky with obvious decrease in reflectance ratio Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Reflectance ratio / / / / / / / /01 Time Figure: Original time for Meteosat-7.

18 on Modeling decrease of spectral response curve φ(λ, t) At time t 0, spectral response curve φ(λ, t 0 ) = φ 0 (λ) Linear or exponential decrease of φ(λ, t) in time? Is decrease of φ(λ, t) in time wavelength dependent? Normalised radiance time = launch time = 2900 days Wavelength (micrometer) Figure: Spectral Response curve for Meteosat-7 at launch and how it could look after 2900 days.

19 - linear decrease - Example on Meteosat-7 time for ocean, clouds, dark vegetation, bright vegetation, dark desert and bright desert without narrow band to broad band correction: Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Reflectance ratio / / / / / / / /01 Time Figure: Original time for Meteosat-7.

20 - linear decrease - Example (2) on Meteosat-7 time after linear ageing correction of the spectral response curve: φ(λ, t) = φ 0 (λ) (1 αt) Reflectance ratio Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert 1999/ / / / / / / /01 Best α = days 1, which corresponds to a linear decrease of the spectral response curve of 2.0% per year. RMS error of the residual drift is 0.95% over the full period. Time Figure: Time with linear ageing model for Meteosat-7 (ageing parameter α = days 1 ).

21 - linear decrease - Example (3) on Reflectance ratio Met-7 time after ageing correction of the spectral response curve: φ(λ, t) = φ 0 (λ) (1 αt) (1 + βt (λ λ 0 )) Ocean Clouds Dark Vegetation 1999/ / / / / / / /01 Time Bright Vegetation Dark Desert Bright Desert Best α = days 1 and β = days 1 µm 1 which corresponds to a decrease of the spectral response curve of 1.8% per year. RMS error of the residual drift is 0.49% over the full period. Figure: Time with linear ageing model for Meteosat-7 (ageing parameters α = days 1, β = days 1 µm 1 ).

22 - exp decrease - Example on Meteosat-7 time after linear ageing correction of the spectral response curve: φ(λ, t) = φ 0 (λ) exp( αt) (1 + βt (λ λ 0 )) Reflectance ratio Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Best α = days 1 and β = days 1 µm 1. RMS error of the residual drift is 0.50% over the full period / / / / / / / /01 Time Figure: Time with linear ageing model for Meteosat-7 (ageing parameter α = days 1, β = days 1 µm 1 ).

23 - spectral change on Studies showed that for CERES FM1, 2, 3 and 4, the shape of the spectral degradation looks like this: Figure: Spectral degradation of CERES instruments (by K. J. Priestley). Similar studies for other instruments (GOME, MERIS) show this same spectral shape.

24 - exp decrease - Example (2) on Meteosat-7 time after linear ageing correction of the spectral response curve: φ(λ, t) = φ 0 (λ) exp( αt) ( 1 exp ( )) β λ t Reflectance ratio Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Best α = days 1 and β = days µm 1. RMS error of the residual drift is 0.50% over the full period / / / / / / / /01 Time Figure: Time with linear ageing model for Meteosat-7 (ageing parameter α = days 1, β = days µm 1 ).

25 - Original time on time for ocean, clouds, dark vegetation, bright vegetatiaon, dark desert and bright desert without narrow band to broad band correction: Reflectance ratio Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert / / / / /01 Time Figure: Original time for.

26 - Linear ageing model on time after linear ageing correction of the spectral response curve Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Ocean Clouds Dark Vegetation Bright Vegetation Dark Desert Bright Desert Reflectance ratio Reflectance ratio / / / / /01 Time / / / / /01 Time Figure: Time with linear ageing model for. Left figure ageing parameter α = days 1. Right figure ageing parameters α = days 1, β = days 1.

27 - Ocean curve on Bump in ocean curve could be caused by El Chichon eruptions in Mexico during end of 1982/03 and beginning of 1982/04. Reflectance ratio Ocean / / / / /01 Figure: Meteosat-7 ocean time after linear ageing correction (ageing parameter α = days 1 ). Time

28 Outline on

29 on Goal to create GERB-like data for MFG First creating ageing models to correct for the degradation of the instruments Meteosat-7 data can be corrected well using a linear wavelength dependent model (stability of 0.49%), except for a slight curve in the time which still needs to be removed (perhaps exponential model instead) data correction is probably hindered by the volcanic eruptions of El Chichon Stronger wavelength dependent ageing in Meteosat-7 than due to different shape of the spectral response curve