A Novel Multitemporal Cloud and Cloud Shadow Detection Method Using the Integrated Cloud Z-Scores Model

Abstract

Accurate cloud and cloud shadow detection in multispectral remote sensing imager due to the high spectral variations of clouds and the complexities of underlying landscapes, especially for images without thermal bands. In this paper, we proposed a multitemporal integrated cloud z-score (MTICZ) method for cloud and cloud shadow detection for multitemporal optical images. First, an integrated cloud z-score (ICZ) index was designed to identify clouds and cloud shadows, and measure the likelihood of a pixel being either a cloudy or shadowed pixel. Clouds and cloud shadows were then detected by differencing the ICZ values between the reference image and the cloudy image, and they were refined using a cloud and shadow matching algorithm. Finally, the MTICZ method was evaluated through cloud fraction estimation and 3,000 random validation samples from the cloudy Landsat scenes. The results indicate that MTICZ method achieved a significant agreement with reference cloud fractions $({{R^2} = {\text{0.97}},{\text{RMSE}} = {\text{4.31}\%}})$ and performed well under complicated land surface conditions, with an average overall accuracy of 91.65%. In addition, the MTICZ method was compared with two popular cloud detection methods, the Fmask method and the multitemporal cloud detection (MTCD) algorithm. This comparison reveals some improvements for cloud identification in complex landscapes, such as impervious surfaces and mixed areas of snow and cloud. Therefore, we argue that MTICZ provides an effective multitemporal method to detect cloud and cloud shadow for optical imagery in complicated landscapes.