Generation of new cloud masks from MODIS land surface reflectance products

Abstract

MODIS land surface reflectance product (MOD09) is one of the most popular data sources for characterizing land surface environments. Because those cloudy observations should be excluded from further analysis, the reliable cloud screening is important for downstream applications. In this paper, an approach is proposed to generate cloud masks from time series of MOD09 products. It is found that an inflexion point exists between the clear-sky and cloudy observations if time series of reflectances assembled from the same location are sorted. The maximum surface reflectance can be composited from these inflexions and those observations with reflectance values larger than the inflexion are identified as cloudy. To the best of our knowledge, this is the first method to composite the maximum snow-free surface reflectance. And a new method is proposed to objectively compare cloud detection results derived from different approaches. Comparisons show that this inflexion-based cloud detection algorithm performs generally better than the cloud masks accompanying with the MOD09 products. The new cloud masks are valuable for those applications relying on the MOD09 products as input and for analysis of the uncertainty of the MODIS cloud mask products.