A multi-year composite of GVI data as a clear reflectance database

A method is developed to generate a top of the atmosphere clear reflectance from the Global Vegetation Index (GVI) product. Our goal is to use this dataset as a threshold to be applied to the forthcoming POLDER observations, for operational cloud detection. The method is based on the hypothesis that clouds add a high frequency signal to the slow variations of the surface reflectance in clear conditions. The validity of our algorithm is verified through an analysis of the temporal profile of the reflectance that it generates. We show that these profiles are better than those resulting from the simpler Maximum Value Composite (MVC) method. The method is applied to five years of GVI products and the results are used to derive a reference database which accounts for the interannual variability of the surface reflectance.