Multitemporal, multichannel AVHRR data sets for land biosphere studies—Artifacts and corrections

Temporal compositing of daily optical satellite data has become an accepted methodology for obtaining frequent images of large areas for studies of the land surface. However, such composite data sets sometimes contain large “artifacts” (i.e., errors due to sources unrelated to the land surface itself). The goal of this study was to develop a series of operations and algorithms that would identify and remove as many of these errors as possible. The specific objective was to obtain, every ten days, surface reflectance in advanced very high resolution radiometer (AVHRR) channels 1 and 2 and normalized difference vegetation index (NDVI)—all referenced to a constant viewing geometry—and surface temperature. The processing steps of the resulting methodology, dubbed “ABC3” for atmospheric, bidirectional, and contamination corrections of the Canada Centre for Remote Sensing (CCRS), includes atmospheric corrections; the identification of pixels “contaminated” by clouds, snow (including subpixel), and shadows from optically thick clouds; bidirectional reflectance and thermal emissivity corrections; and the replacement of the contaminated pixels in the composites through interpolation. The resulting procedures yield surface reflectance and NDVI fully corrected for bidirectional effects; a version of NDVI corrected for solar zenith only; and surface temperature corrected for atmospheric and surface emissivity effects. An evaluation of the resulting data set shows that the procedures provide significantly improved data products compared with the raw composites, but they do not approximate a single-date image sufficiently closely, especially for AVHRR channel 1, whose reflectance values are generally low. This is attributed to the limitations of the input data and knowledge of atmospheric (and partly bidirectional) characteristics applicable to each composite pixel.