Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images


Lake Maggiore - Italy, 3-6 April 2000

Improved atmospheric corrections and data compositing methods for surface reflectance retrieval

Ph. Maisongrande (1), B. Duchemin (1), M. Leroy (1) (P.I.), G. Dedieu (1), J.L. Roujean (2), B. Berthelot (1), Ch. Dubegny (1), R. Lacaze (2)
CESBIO, 18, avenue E. Belin, 31401 Toulouse Cedex 4, France
CNRM/Météo-France, 42 av. Gustave Coriolis, 31057 Toulouse Cedex, France

Paper on Atmospheric correction (pdf file, 509 k)

Paper on Bidirectionnal Effects (pdf file, 1.58 Mo)

This investigation aims at an improvement of the accuracy of surface reflectance products delivered by the VEGETATION system. The foreseen improvement of products concerns the algorithms of atmospheric corrections and removal of anisotropy effects.

Atmospheric corrections: Inaccuracies of atmospheric correction procedures lead to significant errors on surface reflectances retrieved from the sensor measurements. The nominal VEGETATION products are corrected for atmospheric effects using the SMAC code (Rahman and Dedieu, 1994), with the water vapor, ozone and aerosol contents given as inputs. It is clear that any error on these contents translates into a corresponding error on retrieved surface reflectances. In the prelaunch phase, we have seeked to improve the accuracy of these atmospheric contents. One of the issues concerned the choice of sources for the water vapor and ozone concentrations. The conclusion was that while a climatology is sufficient for ozone, daily water vapor contents derived from meteorological models have to be used. This modification has been implemented operationally in the VEGETATION products.

So far, a crude correction for aerosol effects is made with VEGETATION data, using fixed aerosol amounts per band of latitude. We have proposed an original method of retrieval of aerosol content based mainly on the use of the B0 VEGETATION blue channel. A version of this method was partially validated during the prelaunch phase with radiative transfer simulations of TOA reflectances and with time series of POLDER/ADEOS data on sites equipped with AERONET sunphotometer measurements. In the postlaunch phase, VEGETATION data were acquired during one year on 15 areas worldwide equipped with AERONET sunphotometers. These data were used to investigate the potential of self consistent estimates of the ratio of surface reflectances SWIR/B0 to improve the retrieval of the atmospheric part of the top of atmosphere B0 signal. It has been shown that the use of this ratio generally improves the atmospheric correction performances, even on areas without dense vegetation cover. The algorithms of retrieval of this ratio and implementation in an atmospheric correction scheme, and their overall accuracy assessment with actual VEGETATION data are presented elsewhere (see the other paper by Maisongrande et al. in this conference).

Anisotropy removal : The nominal VEGETATION products use the Maximum Value Composite (MVC) technique as a compositing method of temporal series of data. This compositing attenuates but does not suppress undesirable effects due to directional effects (and also cloud contamination of pixels, and atmospheric effects). The basic idea developed here is to remove the directional effects with a BRDF estimated self consistently with a time series of VEGETATION data. In the prelaunch phase, the problem was analyzed with time series of AVHRR data on 3 selected sites : semi-arid (HAPEX 92), boreal forest (BOREAS 94), and agricultural temperate site (Alpilles 96). Three BRDF models adjusted against the data were intercompared on the basis of their ability to provide smooth temporal profiles of normalized reflectances, and to reconstruct BRDFs as close as possible from reference BRDFs measured with the airborne POLDER instrument on each of the 3 sites. It was concluded that Walthall’s model satisfies better the first criterion, and Roujean’s model the second. It was recommended that the ouput variables produced by the operational processing center be spectral hemispherical reflectances. A period of composition of 30 days was also recommended.

The problem was revisited in the post-launch phase 1) by means of numerical simulations ; the simulations were conducted so as to test various cloud cover situations (say, 20%, 40% … cover) and various ways of obtaining the BRDF shape and of applying this knowledge for normalizing the daily data ; 2) by designing and testing an algorithm of anisotropy removal on actual VEGETATION data. This presentation focuses on the numerical simulation aspect, while the presentation by Duchemin et al. describes the results obtained with the VEGETATION data and the algorithmic synthesis. The main outcome of the simulations are 1) that the concept of a fixed 30-day compositing period can not be operative in many cloudy regions of the world (due to lack of data for the estimation of the BRDF shape) and should be replaced by a more flexible compositing period concept, 2) to recommend that the compositing should be first to correct every daily data for the shape of the BRDF, and second produce a 10-day composite by simple average of the directionally corrected daily data.