Spatial variability of satellite visible radiances in dust and dust‐cloud mixed conditions: Implications for dust detection

Using data from MODIS for the 2000–2004 time period, we performed a statistical analysis of visible radiances observed in the presence of mineral dust in cloud‐free and cloudy conditions over oceans. Spatial variability of the 555 nm radiances was examined by introducing two different measures: standard deviation (STD) and a local inhomogeneity parameter (LIP). We demonstrate that introducing the probability density function (PDF) of STD offers a new framework for probabilistic dust detection. We show that the probabilistic approach gives more accurate discrimination of dust from clouds compared with the MODIS fixed STD threshold method. Furthermore, the probabilistic approach enables one to determine the confidence level and skill of dust detection. In addition, we examined the capability of the probabilistic detection of dust‐cloud mixed pixels currently not considered by operational algorithms. A low classification skill of dust‐cloud pixels was found. Introducing multivariate PDFs by including multispectral data might help to overcome this problem.