AVHRR Pixel Level Clear-Sky Classification Using Dynamic Thresholds (CLAVR-3)

Abstract

Clear-sky classifications from the Clouds from the Advanced Very High Resolution Radiometer (AVHRR)-Phase 1 (CLAVR-1) program are used to create an 8-day rotating clear-sky radiation dataset. This dataset is used to create satellite zenith angle dependent, dynamic, cloud/no-cloud albedo and temperature thresholds for ocean and six different vegetation index groups in 10° latitude intervals over the globe. Individual pixels from ambiguously classified 2 ? 2 pixel arrays from CLAVR-1 (MIXED and RESTORED-CLEAR) are reexamined at the individual pixel level, using these dynamic thresholds, and reclassified as CLEAR, MIXED, or CLOUDY. This methodology is referred to as the CLAVR-3 algorithm. It is found that many of the MIXED (partially cloudy or mixed overcast) pixels from CLAVR-1 are cloud free after using the dynamic thresholds. A smaller number of RESTORED-CLEAR pixels are found to be CLEAR, while many are classified MIXED. Poleward of about 80°N and 60°S, the CLAVR-1 algorithm does not provide sufficient detection of unambiguous CLEAR pixels for the creation of the necessary albedo and temperature angular distribution models. Other techniques or datasets will have to be employed at these latitudes to provide the necessary dynamic thresholds for reclassification of CLAVR-1 ambiguous pixels. The CLAVR-3 reclassification leads to a 75% increase in CLEAR pixel population, globally, during the ascending segment (mostly daytime portion) of orbits and a 95% increase during the descending segment (mostly nighttime). Maps of differences in albedo and brightness temperature for grid cells containing clear pixels before and after application of CLAVR-3 are used, as well as histogram analyses, to demonstrate the quality of clear pixels derived from the CLAVR-3 algorithm. These analyses show that the quality of CLEAR pixels from CLAVR-3 may be slightly lower compared with CLAVR-1, particularly over oceans, which is most likely the result of cloud contamination. However, this may be an acceptable consequence in exchange for the resulting dramatic increase in CLEAR pixel population. Each CLEAR pixel application (e.g., sea surface temperature) will have to evaluate how CLAVR-3 impacts their products. It is quite possible that by making minor adjustments and modifications to the threshold tests currently used in CLAVR-3, an optimum increase in CLEAR pixels can be achieved with acceptable levels of cloud contamination. Based on the results presented, the authors conclude that the CLAVR-3 algorithm concept has merit and can be used to enhance the spatial coverage of daily operational land, ocean, and atmospheric parameters that depend on clear-sky radiance observations from AVHRR.