Improvement of MODIS cloud mask over severe polluted eastern China

Previous studies have proved that in the regions with severe air pollution, MODIS cloud mask product (MYD35) tends to overestimate the cloud cover largely. An important reason is that the dense aerosols could be misclassified as clouds. Identification of the misdetected "clouds" of passive remote sensing satellites remains challenging. In this study, we built an algorithm combining screening method and adjusted Fisher Discriminant Analysis (AFDA) to rectify the cloud free pixels misclassified as cloudy in the MYD35 product over the eastern China (EC), where heavy haze pollution occurs frequently in fall and winter. The CALIPSO vertical feature mask (VFM) product was used as an accurate reference. The results showed that our algorithm performs well in the discrimination of the true clouds and misdetected clouds, including the ones caused by the misjudgment of near surface aerosols in heavy haze. The average accuracy reached 96.72%. In EC, fogs ought to be classified as clouds often mixed with haze, resulting difficulty to distinguish fogs and haze. Compared with surface observed fogs, our algorithm also has a good effect on identification of the surface fog in EC with an accuracy of 81.53%. Mean values of a series of cloud properties showed great changes after filtering the misclassified MYD35 cloudy pixels. Thereinto, cloud cover decreased by 0.13, other parameters, including cloud top height, cloud optical thickness, cloud effective radius and cloud water path, also changed significantly.