The visible channel of the Geostationary MVIRI instrument needed some special care to be fit for climate studies. Here is how we approached the Problem.
The Measurements
MVIRI’s channel for visible light measures a broad band of wavelengths roughly between 0.4 and 1.0 µm. Covering the blue, green and red spectral regions at once, the images, that are obtained every 30 minutes, are essentially grey (Figure 1): Dark surfaces, such as oceans, appear dark and bright surfaces, such as deserts or clouds, appear bright. Despite the lack of color, a series of seven MVIRI Instruments now has served the weather community for 35 years with valuable information on the state of the atmosphere (eg. cloud fraction) and surface (solar irradiation, surface albedo). The image pixel received from the satellites are dimensionless counts, but they can be transformed into top-of the atmosphere reflectance by applying the measurement equation, which includes the application of the calibration:
\overline{C}_{E} is the georectified earth-pixel count
\overline{C}_{S} is the mean space count
{E}_{O,sun} is the SRF-convoluted solar irradiance
Y is time since launch in fractional years
a_{0-2} are the calibration parameters
and \theta is the solar zenith angle in radians.
Figure 1: Example image of top-of-atmosphere reflectance from the VIS band of MVIRI onboard Meteosat-7 taken on 30th January at 12:30 UTC (courtesy: frank Rüthrich, EUMETSAT).
The Problem
The sensitivity of the MVIRI visible channel differs for every wavelength. The sensitivity for blue light, for example, is generally less than for red light – making the blue oceans even darker and the red deserts even brighter. This sensitivity pattern of the sensor is described by the spectral response functions (SRFs) which are different for every individual MVIRI instrument. The previously considered SRFs were determined before the launch of each satellite, with techniques and requirements regarding the accuracy evolving over time. Among other issues, previous studies (Govaerts 2010,Decoster et al. 2013) have indicated problems of the instrument calibration that are caused by the inaccurate knowledge of the SRFs. Moreover, it was found that the SRF of one instrument changes its shape over time (Decoster et al. 2013,Decoster et al 2014). Without properly accounting for the true shape of the SRFs (Figure 2), as well as for other effects in the instrument that cause uncertainty, the MVIRI instruments were not fit for climate studies.
Figure 2:The spectral response functions of MVIRI onboard Meteosat-2-7. Dotted lines represent the SRF measured before launch. The blue line indicates the MET7 SRF used as a first guess in the reconstruction methodology for all satellites. Other colors show the reconstructed SRFs at different degradation stages(courtesy: Ralf Quast, FastOpt).
The Methods
To reconstruct the true shape of the SRF of each satellite at each aging state, calibration targets were used with dominant spectral contributions in different regions of the visible spectrum. There are ocean targets (blue), Desert targets (red) and deep convective clouds (white). A methodology was developed to accurately simulate the satellite signal above those target types for instances, where also satellite observations are available. Large numbers of collocated pairs of observations and simulations were then used to apply a new optimal estimation technique designed to retrieve the best fitting shape of the SRF (Figure 2). The reconstructed SRFs were then used for recalibrating each instrument. In this context also new methods for propagating uncertainties through the entire processing chain were applied.
The Results
Figure 3: Recalibrated (harmonised) time series of the MVIRI FCDR top-of-atmosphere reflectance extracted above Algeria, along with uncertainties divided based on the error correlation properties (independent and structured) (courtesy: Frank Rüthrich, EUMETSAT).
The recalibration of the MVIRI instruments has led to an uncertainty-quantified harmonised data record that has proven to correctly account for the different SRFs. While differences of the instruments are maintained in the dataset (Figure 3), those are fully attributable to the now well known SRFs and can be considered and used by the users.
Of course the accounting for differing, and time-variant SRFs requires additional caution during the application. But the benefit of those additional steps in the processing is a very stable data-record as illustrated with the anomaly time series of homogenised reflectances above Algeria in https://doi.org/10.3390/rs11101165.
Figure 4: Time series of homogenised reflectance above Algeria after seasonality was removed by considering only anomalies from the mean annual cycle. SCIAMACHY spectra from that site were used for deriving the homogenisation adjustments. Details are provided here. (courtesy: Frank Rüthrich, EUMETSAT).
