A Multi-Annotator Ground-Based Sky Camera Dataset for Cloud Observation in Montenegro

2,522 ground-based sky images captured by a low cost outdoor camera in Montenegro between 3 October and 17 December 2025, together with 11,191 independent annotations produced by nine professional observers employed at national meteorological stations. Each annotation records seven standard WMO synoptic cloud observation variables: total cloud cover (N), amount of low cloud (Nh), height of the base of the lowest cloud (h), and the CL, CM and CH cloud type codes, plus a coarse dominant cloud altitude class. Most images were annotated independently by four to five experts (median 5, maximum 9), and every individual judgement is retained rather than collapsed into a single consensus label. The dataset therefore supports both supervised cloud classification from low-cost imagery and the study of inter-observer variability in visual cloud observation. Expert agreement is high for total cloud cover (Krippendorff's alpha = 0.91 with the interval difference function) and substantially lower for the discrete cloud type codes, providing an empirical estimate of the irreducible label noise against which automatic classifiers can be benchmarked. All textual content has been translated from Serbian into English. See README.md for the full data dictionary, code books, distributions, agreement statistics, known issues and recommended splitting strategy.