Probabilistic dust-cloud mask as a discrimination tool and a data confidence level indicator

Reliable discrimination of aerosols from clouds is critical for retrieving both aerosol and cloud properties as well as other atmospheric characteristics. Using MODIS data for the period 2000-2004, this study examines several techniques that were proposed for discriminating mineral dust from clouds. A number of representative cases of dust plumes mixed with clouds over oceans were analyzed. Selected cases represent the main dust sources located in East and South Asia, Middle East, Northern Africa, and Australia. The commonly used 3x3 1km pixel standard deviation approach is extended beyond the use of a threshold value as a discrimination criterion, allowing for determining not only cloudy and dusty pixels but mixed dust-cloud pixels as well. The new approach allows for calculating the probability of a pixel to belong to any of these types, and to calculate the confidence levels associated with the use of such aerosol-cloud mask. As an alternative to the 3x3 standard deviation, we adopt and test a local inhomogeneity parameter, which theoretically offers advantages in measuring the variability of the reflectances. We also tested the techniques based on brightness temperature differences. Our study demonstrates various limitations of the existing methods and stresses the need for improved techniques. The results will be presented and implications for the discrimination of dust from clouds with passive sensors will be discussed.