Aerosol and Albedo from MVIRI

Generation of a surface albedo and aerosol CDR from Meteosat MVIRI/VIS FCDR

By Yves Govaerts, Marta Luffarelli and Frank Rüthrich

The algorithm

Along with the AVHRR , MVIRI is one of the few sensors that can estimate Aerosol Optical Thickness (AOT) prior to 2000.  Moulin et al (1997) has already demonstrated the potential of MVIRI/VIS observations to monitor inter-annual variability of dust transport over the Atlantic Ocean. However, this instrument has never been used to derive aerosol properties over land surfaces. Additionally, the retrieval of aerosol optical properties over land surfaces necessitates discriminating the contribution of aerosol scattering from the surface in the observed signal which can be achieved with multi-directional observations.

Pinty et al. (2000) were the first to propose an algorithm for the joint retrieval of surface reflectance and aerosol properties to demonstrate the possibility of generating CDR from data acquired by operational weather geostationary satellites. Due to limited operational computational resources available in 1998 in the EUMETSAT reprocessing ground segment, where the data were processed, the development of this algorithm was subject to strong constraints. The Radiative Transfer Equation (RTE) solutions were precomputed and stored in lookup tables with a very coarse resolution, limiting the maximum retrieved AOT to 1, which represents a severe limitation over the Sahara region where AOT values can easily exceed such a limit. Furthermore, the radiative coupling between aerosol scattering and gaseous absorption was not taken into account. This algorithm, referred to as Geostationary Surface Albedo (GSA), has been substantially improved and generalised for the processing of any satellite data (Govaerts and Luffarelli, 2018). This new algorithm, referred to as Combined Inversion of Surface and AeRosol (CISAR) includes the following features:

  • FASTRE (Figure 1), the CISAR forward radiative transfer model, is solved online with a continuous variation of the state variables (aerosol single scattering properties, concentration and surface reflectance parameters) in the solution space;
  • The minimization of the cost function is performed within an Optimal Estimation framework, a one-dimensional variational retrieval scheme that seeks an optimal balance between the information that can be derived from the observations and the one that is derived from prior knowledge of the system;
  • Retrieval uncertainties estimated from the propagation of the radiometric and prior information uncertainties in the OE framework assuming a linear behaviour of the modelled radiative transfer processes in the vicinity of the solution.   

Figure 1: Atmospheric vertical structure of the FASTRE model. The surface is at level Z0 and radiatively coupled with the lower layer La extending from level Z0 to Za. This layer includes scattering and absorption processes. The upper layer, Lg, runs from level Za to Zs and only accounts for gas absorption processes (after Govaerts and Luffarelli, 2018).

The CISAR algorithm has been successfully applied on data acquired by radiometers on board polar and geostationary orbiting platforms (Luffarelli and Govaerts, 2019).

The scientific issues

The spectral width of the MVIRI/VIS band represents however a serious challenge for the generation of a physically-based CDR. Radiative transfer theory is formally valid only for monochromatic cases, i.e., no important spectral variations of the optical properties of the observed media are expected to take place in the processed spectral intervals.

Figure 2: Molecular transmittance in the MVIRI/VIS band spectral range for the following gases: water vapor (blue), ozone (magenta) and oxygen (orange). The grey line shows typical total (Rayleigh and aerosol) scattering transmittance. The green line illustrates the typical reflectance of a green leaf. The red line shows the pre-launch SRF of MVIRI/VIS band on board Meteosat-7. Wavelength is expressed in μm.

This assumption is clearly violated for the MVIRI/VIS band as it can be seen on Figure 2. This spectral interval contains some strong gas absorption bands and it also subject to aerosol scattering processes whose magnitude varies with the wavelength. Vegetated surface reflectance exhibits also strong and fast spectral variations over this spectral region as a consequence of the differences in the radiation transfer regimes occurring on both sides of 0.7 μm, i.e., mainly governed by absorption (scattering) at wavelengths shorter (larger) than 0.7 μm. Conversely, atmospheric radiative processes are dominated by scattering (absorption) at wavelengths shorter (larger) than 0.7 μm.  

It is therefore not possible to perform a retrieval in that spectral interval based on the radiative transfer theory assuming a monochromatic spectral interval, i.e., with invariant optical properties. As an effect, the calculation of the Top Of Atmopshere (TOA) Bidirectional Reflectance Factor (BRF) in the MVIRI/VIS band \tilde r_p =\int_\lambda r(\lambda) expressed as

\tilde r_p = \int_\lambda r(\Omega; \omega_0(\lambda), \tau(\lambda), \rho_0(\lambda), k(\lambda), \Theta(\lambda), h(\lambda))

differs from the one expressed as

\tilde r_m = r(\Omega;\int_\lambda \omega_0(\lambda),\int_\tau (\lambda),\int_\lambda\rho_0 (\lambda),\int_\lambda k(\lambda),\int_\lambda\Theta a(\lambda),\int_\lambda h(\lambda)


  • \Omega represents the illumination and viewing conditions;
  • (\omega_0(\lambda) is the spectral single scattering albedo of the FASTRE scattering layer;
  • \tau(\lambda) is the spectral aerosol optical thickness of the FASTRE scattering layer;
  • \rho_0(\lambda), k(\lambda), \Theta(\lambda), h(\lambda) are the parameters of the surface BRF model.

This difference is illustrated in Figure 3 where the blue curve represents the TOA BRF \tilde{r}_p calculated for one day of MVIRI/VIS band observation over a tropical forest and the red curve shows the corresponding BRF \tilde{r}_m value.

Figure 3: TOA BRF daily cycle over a tropical forest expressed as \tilde{r}_p    (blue curve) and \tilde{r}_m    (red curve).

A workaround has been implemented in the CISAR algorithm to address this issue that relies on two successive corrections. A black surface is first assumed and the VIS band single scattering albedo

\tilde{\omega}_0 = \int_\lambda \omega_0(\lambda)

of the FASTRE scattering layer is adjusted to minimize the difference between \tilde{r}_m and \tilde{r}_p .

The correction factor \gamma of the \tilde{\omega}_0 value adjustment is estimated for different values of the water vapour total column concentration and aerosol optical thickness \tilde{\tau} .

The second correction step concerns the magnitude of the surface BRF \tilde{\rho}_0 . This correction is also estimated so that \tilde{r}_m \rightarrow \tilde{r}_p . As it is not possible to know a priori the surface spectral variations within the MVIRI/VIS band, the correction was determined for a large number of pre-calculated surface reflectance spectral variations.


MVIRI/VIS FCDR data are accumulated during the course of one day to form a multi-angular observation vector. The CISAR algorithm is applied on this observation vector to derive:

  • \tilde{\tau} the aerosol optical thickness in the MVIRI/VIS band and associated uncertainty;
  • \tilde{\rho}_0, \tilde{k}, \tilde{\Theta}, \tilde{h}   the values of the surface BRF model in the MVIRI VIS band from which the BiHemispherical Reflectance (BHR) is derived.

These values are derived on a daily basis at the native MVIRI/VIS band pixel resolution.

A level-3 CDR is generated from these values aggregating them on a monthly basis at one-degree resolution (Figure 4). Over ocean surface, the MVIRI AOT compare favourably with the one derived from MODIS observations to the exception of Antartic and Groenland where very high optical thickness are derived. Over land surfaces, the AOT derived in the MVIRI/VIS band exceeds the MODIS one.

Figure 4: Monthly mean over August 2003 of the AOT in the MVIRI VIS band retrieved with the CISAR algorithm (left panel) and the AOT at 550 nm retrieved from MODIS with the combined Dark Target Deep Blue algorithm (right panel).  Despite the different magnitude of the retrieved AOT, some patterns are visible in both datasets, such as the biomass burning in central Africa and the dust moving from the Sahara region towards Latin America.

FCDR benefits for CDR generation

The MVIRI/VIS band FCDR presents decisive advantages for the generation of a CDR with respect to the original Meteosat First Generation level 1.5 data in the so-called RECT2LP format. This original format contains only the digital count values. The quantitative use of these data required significant efforts to generate the corresponding TOA BRF from the digital count values. Viewing geometry is also missing is that format. The new MVIRI/VIS band FCDR provides to the users directly TOA BRF values accounting for the reconstructed sensor spectral response, together with the illumination and viewing geometries and acquisition time. The availability of detailed pixel-based information on the various types of radiometric uncertainties represents a significant breakthrough of this new FCDR. Such information is indeed of critical importance for retrievals performed in an Optimal Estimation framework as is the case with the CISAR algorithm.

Figure 5 AOT scatterplot between CISAR retrieval (y axis) and AERONET product (x axis) over water. The retrieval is performed with the original SSR (left panel) and with the ageing SSR (right panel).

The use of reconstructed VIS band spectral responses accounting for its spectral ageing is undoubtedly the most important benefit of this FCDR. Figure 5 and 6 illustrate the difference in AOT retrieval based on original RECT2LP and FCDR data. The CISAR algorithm has been applied on two time series of MVIRI observations acquired over AERONET stations corresponding to water and vegetated surface types. The first time series is directly based on the original RECT2LP format where data have been calibrated with the original pre-launch sensor spectral function. The second time series is based on FCDR relying on the reconstructed sensor spectral function accounting  for its spectral ageing. Over water surface the correlation coefficient between the MVIRI AOT values and the AERONET ones increases from 0.38 with the original sensor spectral function up to 0.72 with the reconstructed ones. There is also a benefit over vegetated  surfaced thought it is less pronounced as over water surfaces.

Figure 6 Same as Figure 5 but over vegetation