Identification of contaminated pixels in AVHRR composite images for studies of land biosphere

The objective of this study was to develop a robust method for Identifying contaminated pixels from AVHRR composite data sets intended for biospheric studies. Of particular interest were land pixels partially contaminated by clouds, smoke, or other atmospheric aerosols, as well as pixels with partial snow or ice cover. The method developed for this purpose, dubbed CECANT (cloud elimination from composites using Albedo and NDVI trend), uses Channel 1 surface reflectance to identify fully cloud-, snow-, and ice-covered pixels; and two parameters (R, Z) to distinguish bright (in Channel 1), clear-sky pixels from intrinsically darker but partly contaminated pixels. R and Z are based on the seasonal NDVI trajectory. Three thresholds are required to apply this approach; they are obtained from the histograms of Channel 1, R and Z. The thresholds are adjusted separately for each compositing period. Tests with multidate histograms and single-date images indicated the consistency and robustness of the method in an area where most seasonal NDVI trajectories have a single peak. An application of the CECANT procedure to AVHRR composites over Canada for 1993 showed that an average 51% of the land pixels were fully or partly contaminated (37% +/- 5% in June to August period), with a range from 31% (midsummer) to 91% (October). The method should be applicable to other geographic regions where seasonal AVHRR composite data sets (minimum is Channel 1 and NDVI for each composite period) are available.