WorldView-3 imager has 16 different narrowband spectral channels between 425 and 2300nm. WorldView-3 imager lacks a cloud mask product, which greatly impacts data usability especially in polar regions. Unfortunately, only half of those spectral channels are present at all times. The number of observations made in shortwave infrared range (channels 9-16) is very limited. Last year we presented a spectral cloud mask for WorldView-3 using a more traditional approach of spectral thresholding. That approach, while quite successful, is necessarily limited to scenes where all 16 channels are present.

We set out to develop a new approach based on statistics to perform cloud mapping. The calculated statistical envelopes are designed to be suitable for training AI/ML models for cloud signature recognition. In this presentation we show the results of this new statistics-based approach.