A Multi-source Dataset of All-sky Images and Cloud-mask Annotations (2024–2025)

This dataset comprises three components based on data sources:(1) Open-source annotated subset: pixel-level cloud-mask annotations were produced for the Total-Sky Cloud Image Set (TCIS) using a unified binary cloud/non-cloud definition, including 195 images and their corresponding annotation files.(2) Open-source sequence annotated subset: for all-sky image sequences acquired by the Total Sky Imager (TSI) within the Atmospheric Radiation Measurement (ARM) program, initial cloud masks were generated using the Segment Anything Model (SAM) and subsequently corrected and validated by annotators.

This semi-automatic workflow was applied to extend pixel-level labels to continuous time-series imagery, resulting in 2560 images and their corresponding annotation files.(3) Site-observation sequence subset: continuous all-sky image sequences at a 5-minute interval were collected at the Daihai Aquatic Ecosystem Integrated Management Command Center and the Lhasa Division of the Institute of Tibetan Plateau Research, Chinese Academy of Sciences, using a self-developed Ground-Based Intelligent Multi-Parameter Environmental Health Remote Sensing Diagnostic Instrument (GB-EHRSDI), containing 1356 and 218 images, respectively.