Reducing bias in satellite derived cloud fraction estimates caused by finite resolution measurements

Clouds are one of the biggest uncertainties in climate modeling, and cloud fraction is the property showing the largest disagreement with observations. Satellite observations provide the only reasonable way to compile a global picture of cloud coverage on a reasonable time scale. One of the tradeoffs for global coverage over short time scales is the spatial resolution of the measurements, which for modern sensors tend to be ~ 1 km in scale. Many clouds, especially boundary layer clouds, such as those observed during the Rain in shallow Cumulus over the Ocean (RICO) field campaign, exist at scales less than 1 km.

The Multiangle Imaging SpectroRadiometer (MISR) instrument's clear-conservative Radiometric Camera-by-camera Cloud Mask (RCCM) has a 1.1km resolution. Its clear conservative nature prevents cloud contamination in clear air products, such as aerosol optical depth, by providing excellent detection of sub-pixel clouds. The cloud fraction derived from the RCCM has been shown to be an overestimate that depends on the underlying cloud area distribution and true cloud fraction, on average overestimating by 0.35 for cumulus clouds in the RICO region.

This study aims to develop a pattern recognition based correction technique that can be implemented on clear conservative cloud masks, such as the RCCM, to remove the bias in cloud fraction and provide a more accurate global cloud fraction product. It also aims to determine the optimal resolution at which to design future instruments to balance the tradeoff between coverage and resolution and finds that a 250-400m cloud mask would minimize the effect on the cloud fraction product. We will also discuss the ongoing work to implement the technique operationally as part of the MISR mission.