Proba-V Advanced Atmospheric Correction

Minimize PV-LAC Introduction

The PV-LAC project aimed at the development, testing, and validation of two advanced algorithms for the scientific exploitation of the PROBA-V mission, one Activity focusing on Atmospheric Correction and the other on Coastal Products.


In the first Activity, an Optimal Estimation (OE) method was developed for the Atmospheric Correction of the 1 km resolution PROBA-V data. The method relies on the joint  aerosol optical depth and surface reflectance retrieval through the inversion of a physically-based coupled surface-atmosphere radiative transfer model. This approach was originally developed for geo-stationary satellite observations, which provide a very high temporal sampling, and was adapted to enable processing of PROBA-V 1 km data. PROBA-V observations are accumulated over 16 days to compose a multi-angular and multi-spectral observation vector. Within this 16-days period, surface  radiative properties are assumed invariant. Additonally, the surface reflectance retrieval and associated uncertainty of the previously processed accumulation periods is used as a priori information for the inversion.


The second  Activity within PV-LAC focused on the exploration of PROBA-V for Coastal Products generation, using only the PROBA-V central camera. This camera provides observations at 100 m spatial resolution with a 5-day global coverage. Although PROBA-V was not conceived as an Ocean Colour (OC) mission, its specifications (spectral bands and signal-to-noise ratio) are suitable to monitor some of the key OC parameters, in particular Suspended Particulate Matter (SPM). Especially in estuarine environments, the higher spatial detail of PROBA-V, combined with a sufficiently high repeat frequency, was demonstrated to be of added value to other sensors.


More information on PROBA-V can be found at http://proba-v.vgt.vito.be/


The project was executed between January 2016 and October 2017 and consisted of four tasks:

  1. Scientific review and requirements consolidation (April 2016)
  2. Algorithm definition (November 2016)
  3. Algorithm evaluation and validation (May 2017)
  4. Recommendations and roadmap (October 2017)
Minimize Methodological approach

Description of the proposed approach

The main purpose was to explore the possibility to improve surface reflectance retrieved from PROBA-V observations through a more accurate aerosol characterisation and to discuss possible methods that can be compatible with requirements established by the Global Climate Observing System (GCOS) for the corresponding ECVs. As surface reflectance and atmospheric extinction due to aerosols are tightly radiatively coupled, the most promising  retrieval approaches over land surfaces are those derived from the joint surface reflectance and aerosol retrieval. This was performed for observations of daily accumulated Meteosat Second Generation (MSG) SEVIRI (Govaerts et al., 2010). The implementation and evaluation of this method is described by Wagner et al. 2010.


The pratical implementation of such methods is based on Optimal Estimation (OE),  a method allowing for a rigorous estimation of the retrieval uncertainties or covariance error matrices, providing insights on the coupling between the various variables as a function of the observation and a priori information uncertainties. Hence, this method accounts for the observation uncertainties, propagating them into the retrieval uncertainties as required by Quality Assurance for Earth Observation (QA4EO) principles (http://www.qa4eo.org/) and International Organization for Standardization (ISO) standards, such as the  Guide to the Expression of Uncertainty in Measurement  (GUM, http://www.iso.org/sites/JCGM/GUM-introduction.htm).


The above described method has never been systematically applied to polar orbiting satellite observations with limited spectral capabilities, such as PROBA-V. Therefore it was essentially considered a demonstration activity to illustrate the benefit of such an approach, providing recommendations to pave the way towards operations.


Current methods developed for the surface reflectance retrieval were reviewed in the light of a series of criteria established according to the GCOS recommendations for ECV derivations and QA4EO principles. This review is reported in the Requirements Baseline Document (RBD).

Figure 1: Three-layer structure of the FASTRE RTM: the lowest layer represents the surface, the middle is the atmospheric scattering and absorption layer (by aerosols, water vapour and ozone), while in the upper layer only gaseous absorption occurs.


CISAR is a versatile algorithm designed to perform the joint inversion of surface reflectance and aerosol properties over both land and sea surfaces. This algorithm relies on FASTRE, a fast radiative transfer model the vertical structure of which is composed of three layers, as shown in Figure 1.

FASTRE explicitly represents aerosol single scattering properties without relying on pre-computed LUTs. The inversion is performed with OE (Govaerts et al., 2010), It is assumed that the surface properties vary slowly in time and that the spectral and temporal variations of aerosol properties behave in a well-described and smooth way within the given PROBA-V spectral resolutions. In order to build a multi-angular observation vector that characterises the surface reflectance anisotropy, PROBA-V observations are accumulated over 16 days, with this period being shifted every 8 days.

The a priori information is any additional knowledge on the observed system. In the present case, the magnitude of the state variable and constraints on aerosol optical thickness (AOT) temporal and spectral variability are considered. Information on the aerosol temporal variability and spectral signature is also taken into account to build the prior information on the AOT. Aerosol layer height information is obtained from a climatology of Kinne et al. (2013).

The retrieval uncertainty is estimated from OE theory, analysing the shape of the cost function in the vicinity of the solution. In CISAR's most recent version, a Quality Indicator (QI),  evaluating among others the solution convergence and the relative contribution of the observations to the solution, was incorporated. The CISAR algorithm delivers AOT at 0.55 μm and the Rahman-Pinty-Verstraete (RPV) Bi-Directional Reflectance distribution (BRDF) inversion model parameters for the four PROBA-V bands (BLUE, RED, NIR, SWIR). From these parameters, the algorithm calculates the Bi-Hemispherical Reflectance (BHR) assuming perfectly diffuse illumination conditions.

A detailed description of the FASTRE forward model, the CISAR retrieval algorithm (including OE theory), its assumptions and limitations, as well as the input data requirements and implementation aspects are given in the Algorithm Theoretical Baseline Document (ATBD)


Minimize Algorithm Validation

For the algorithm validation, 50 AERONET stations over different landcover types and climate regions were selected, see Figure 2.


Figure 2: The location of the AERONET stations used for PROBA-V observation extraction and algorithm performance validation.


Two years of PROBA-V observations (2014 – 2015) were simulated over these stations and were used to analyse the algorithm performance, including error propagation and quality control in various pre-defined conditions. These simulations provided the algorithm's theoretical optimum performance on PROBA-V observations. Subsequently, the algorithm was applied to real PROBA-V TOA reflectance observations extracted over the same stations. For the validation of the derived surface reflectance and AOT, MODIS Bi-Hemispherical Reflectance (BHR) and AERONET AOT observations were used, respectively.


Figure 3and Figure 4show scatter plots for the AOT and BHR validation, respectively. It can be seen that AOT is overestimated for low to medium AERONET AOT (AOT < ~0.7), which is mainly the case over bright surfaces with little or no vegetation. Over these surfaces, the contrast between an aerosol layer and the underlying surface is less, thereby complicating a proper AOT retrieval. Further, the panels demonstrate the added value of the Quality Indicator, with better accuracy, precision, and uncertainty for the high-quality retrievals (QI ≥ 0.8).


Figure 3: CISAR AOT validation with AERONET for high- (QI  ≥ 0.8, left panel) and low-quality retrievals (QI ≤ 0.2, right panel). The solid black line indicates the 1:1 relation, while the dotted black lines indicate the GCOS AOT target accuracy requirement of ±10%. The solid red line indicates the orthogonal least square regression, while the colourbar indicates the colour coding for the bin density. All results were obtained for 2014 – 2015.


For BHR, the plots mainly show an overestimation of CISAR on PROBA-V relative to MODIS BHR, as is visible from the (orthogonal) least square regression being above the 1:1 line. The overestimation might be due to differences in the retrieved AOT for PROBA-V and MODIS, the spectral differences (the MODIS ‘BLUE' channel is narrower than PROBA-V BLUE), as well as differences in the absolute calibration between the two sensors. For the RED – SWIR spectral channels, results are very promising, with good accuracy and precision values. Similar as to AOT, better performance is obtained using the high-quality retrievals.


Figure 4: CISAR BHR versus MODIS BHR retrievals for QI ≥ 0.8 (upper 4 panels) and QI ≤ 0.2 (lower 4 panels).


Finally, to get an indication on the added value of CISAR relative to the current operational TOC reflectances, both TOC reflectance datasets were compared against a "reference" dataset of TOA reflectances atmospherically corrected using the aerosol provided by AERONET. The results of this comparison are shown in Figure 5. Most striking is that the CISAR TOC reflectances agree considerably better with the AERONET reference TOC reflectances for the BLUE channel than the operational ones, while for the RED – SWIR differences between CISAR and operational TOC reflectances relative to AERONET TOC reflectances are smaller. As aerosols contribute most to the observed TOA reflectance in the BLUE channel, a difference between the CISAR and operational AOT will result in larger TOC reflectance differences in this channel.


An extensive description of the validation methodology and additional results for the AOT and BHR validation are described in the Validation Report.    


Figure 5: TOC reflectances for BLUE (upper left), RED (upper right), NIR (lower left), and SWIR (lower right) calculated using SMAC, but with different AOT inputs: the dark-coloured dots were obtained with CISAR v2 AOT (QI > 0.8), while bright-coloured dots denote the current operational (OP) TOC reflectance. The least square regression lines are drawn in similar coloured dashed lines. Values are assessed against TOC reflectances obtained with SMAC, using AERONET Level 2.0 AOT, on the x-axis.


Minimize References
  • Berthelot B. and G. Dedieu, Correction of atmospheric effects for VEGETATION data. Physical Measurements and Signatures in Remote Sensing, p. 19 – 25, 1997.
  • Govaerts, Y. and M. Luffarelli, Joint retrieval of surface reflectance and aerosol properties with continuous variations of the state variables in the solution space: Part 1: theoretical concept, Atmos. Meas. Tech. Discuss., https://doi.org/10.5194/amt-2017-29, in review, 2017.
  • Govaerts, Y. M., S. Wagner, A. Lattanzio, and P. Watts, Joint retrieval of surface reflectance and aerosol optical depth from MSG/SEVIRI observations with an optimal estimation approach: 1. Theory, J. Geophys. Res., 115, D02203, https://doi.org/10.1029/2009JD011780, 2010.
  • Holben, B.N., T.F. Eck, I. Slutsker, D. Tanré, J.P. Buis, E. Setzer, E. Vermote, J.A. Reagan, Y.J. Kaufman, T. Nakajima, F. Lavenu, I. Jankowiak, and A. Smirnov, "AERONET—A federated instrument network and data archive for aerosol characterization." Remote Sens. Environ., 66(1), 1-16, 1998.
  • Kinne, S., D. O'Donnel, P. Stier, S. Kloster, K. Zhang, H. Schmidt, S. Rast, M. Giorgetta, T.F. Eck, and B. Stevens. 2013. ‘MAC-v1: A New Global Aerosol Climatology for Climate Studies'. J.  Adv. Mod. Earth Sy. 5 (4): 704–40. https://oi.org10.1002/jame.20035, 2013.
  • Lucht, W., C.B. Schaaf, and A.H. Strahler, An algorithm for the retrieval of albedo from space using semiempirical BRDF models. IEEE Transact. Geosci. Remote, 38(2), 977-998, 2000.
  • Maisongrande P., B. Duchemin, B. Berthelot, C. Dubegny, G. Dedieu, and M. Leroy, A new algorithmic concept for atmospheric correction of surface reflectances delivered by the VEGETATION system. Proceedings of the VEGETATION 2000 conference, Belgirate, 2001.
  • Rahman, H., and G. Dedieu, SMAC: a simplified method for the atmospheric correction of satellite measurements in the solar spectrum. Remote Sensing, 15(1), 123-143, 1994.
  • Sterckx, S., I. Benhadj, G. Duhoux, S. Livens, W. Dierckx, E. Goor, S. Adriaensen, W. Heyns, K. Van Hoof, G. Strackx, K. Nackaerts, I. Reusen, T. Van Achteren, J. Dries, T. Van Roey, K. Mellab, R. Duca, J. and Zender, The PROBA-V mission: image processing and calibration. Int. J. Remote Sens., 35(7), 2565 – 2588, 2014.
  • Wagner, S. C., Y. M. Govaerts, and A. Lattanzio (2010), Joint retrieval of surface reflectance and aerosol optical depth from MSG/SEVIRI observations with an optimal estimation approach: 2. Implementation and evaluation, J. Geophys. Res., 115, D02204, https://doi.org 10.1029/2009JD011780.