Methods

The following steps were taken to produce the final QA4ECV spectral albedo products:

1. Satellite Top-of-Atmosphere 1km reflectances, corrected for Sun-Earth distance and using the same solar irradiance model are used as input from the ESA MERIS, SPOT-VEGETATION and Proba-V instruments

2. Every input Pixel is identified as land, water, cloud, snow with a given probability

3. Spectral Directional Reflectance, SDR (sometimes known as Top-of-Canopy reflectances or Bidirectional Reflectance Factors, BRFs) are retrieved correcting for the effects of the atmosphere using aerosol optical depth estimates produced from the same input data

4. SDRs are mapped into the same MODIS spectral responsivities Big Data analytics and radiative transfer models

5. Every orbit is binned into the MODIS SIN 10° x 10° tiles

6. Using a DAILY MODIS spectral BRDF climatology, Bi-directional Reflectance Distribution Functions (BRDFs) in netCDF are derived from the MODIS-like BRFs using an optimal estimation model and gaps are filled in (due to persistent cloud cover or low solar elevations) from a 16 year climatology derived from a DAILY MODIS BRDF time series

7. Bi-Hemispherical Reflectances (BHR, sometimes known as "white-sky" albedo) and Direct Hemispherical Reflectances (DHR, sometimes known as "black-sky" albedo) are then integrated from the BRDF for a particular solar angle range every day in netCDF

8. 1km MODIS SIN tiles are then upscaled, mosaiced and projected into 0.05° and 0.5° and into monthly time-steps in netCDF

9. Browse products are generated from each albedo product at 1km, 0.05° and 0.5° mosaics in PNG and animations with annotations created in MPEG2 and placed on the right area of the website

The traceability chain for albedo can be found here



BHR-TIP LAI and FAPAR processing:

(1) The separate snow and no snow QA4ECV albedo are used individually. Their projection is the one of the final product.

(2) The albedo inputs are checked for physical consistency and completeness on a per grid cell basis. Inconsistencies are flagged, apparently flawed data is rejected.

(2) Look-up tables of inversions from the Two-stream Inversion Package (TIP) are used, matching the albedo uncertainties and the uncertainty correlation. Inversion results for LAI, FAPAR, and their a posteriori uncertainties are retrieved by simple table look-up.

(3) The inversion results from snow and no snow are combined, weighted by the relative Weighted_Number_of_Samples variable from the respective albedo observations.

(4) The a posteriori uncertainties of the inversions are propagated to the combined result.

(5) The product is stored in CCI and CF-1.6 compliant netCDF4 classic format, containing one day per file. Processing and quality flags allow for the exact reconstruction of the processing path of each individual grid cell.

The traceability chain for BHR-TIP LAI and FAPAR can be found here



DHR-FAPAR processing:

(1) Thresholds were appllied to the daily surface reflectances in red and near-infrared spectral bands, i.e. Band 1 and Band 2, for separating bare soil, vegetated and cloud pixels occurrence (JRC Flag).

(2) The surface reflectances were then converted to normalized reflectances in Band 1 and Band 2 using pre-defined values of RPV model parameters over bare soils and vegetated regions, respectively. They were optimized for each NOAA platform spectral band response.

(3) Using optimized coefficients polynomial formulae daily FAPAR and its uncertainties were calculated.

(4) Time-composite was then applied to provide 10-days and monthly products.

(5) Regridding method was applied to provide products at 0.5°x0.5°.

The traceability chain for DHR FAPAR can be found here