Clouds over the Antarctic Plateau exert a strong influence on the regional radiation budget, yet observations and modelling of their properties remain scarce. Here, nine years (2012–2020) of ground-based high-resolution spectral radiance measurements from the REFIR-PAD spectroradiometer at Concordia Station (Dome C, Antarctica) are analyzed in synergy with co-located lidar observations. A machine-learning Cloud Identification and Classification (CIC) algorithm is applied to discriminate clear sky, ice cloud, and mixed-phase cloud conditions, enabling the construction of a long-term cloud climatology. Cloud optical and microphysical properties are subsequently retrieved using a simultaneous atmospheric and cloud retrieval framework, for cases with reliable cloud boundaries and cloud base heights above 500 m. Results confirm that cloud occurrence over Dome C is dominated by optically thin ice clouds, with approximately 95 % of cases exhibiting optical depths below 1. Median optical depth ranges from 0.11 in summer to 0.32 in winter. The median temperature of the ice layers is approximately 237 K. Mixed-phase clouds are rare and mainly confined to the austral summer, but exhibit larger optical depths (median 1.7) and warmer temperatures (approximately 246 K). Based on the retrieved dataset, a new parameterization of ice crystal effective dimension is derived. Compared with commonly used parameterizations developed for tropical and midlatitude conditions, the proposed scheme predicts systematically smaller particle sizes, highlighting the inadequacy of existing formulations for the Antarctic environments. These results provide new observational constraints on Antarctic cloud microphysics and support improved cloud representation in climate and numerical weather prediction models.