Nighttime Cloud Cover Estimation Method at the Saishiteng 3850 m Site

Abstract

Cloud cover is critical for astronomical sites because it can be used to assess the observability of the local sky and further the fractional photometric time. For cloud monitoring in site-testing campaigns with all-sky cameras, previous studies have mainly focused on moonless images, while the automatic processing methods for moonlight images are explored quite few. This paper proposes an automatic estimation method for cloud cover, which takes all cases of nighttime gray-scale all-sky images into account. For moonless images, the efficient Otsu algorithm is directly used to detect clouds. For moonlight images, they are transformed into cloud feature image using a colorization procedure, and then the Otsu algorithm is used to distinguish cloud pixels from sky pixels on the cloud feature image. The reliability of this method was evaluated on manually labeled images. The results show that the cloud cover error of this method is less than 9% in all scenarios. The fractional photometric time derived from this method is basically consistent with the published result of the Lenghu site.