Statistics on the occurrence of clear skies, ice clouds, and mixed-phase clouds over Concordia Station, in the Antarctic Plateau, are provided for multiple timescales and analyzed in relation to simultaneous meteorological parameters measured at the surface. Results are obtained by applying a machine learning cloud identification and classification (CIC) code to 4 years of measurements between 2012–2015 of downwelling high-spectral-resolution radiances, measured by the Radiation Explorer in the Far Infrared – Prototype for Applications and Development (REFIR-PAD) spectroradiometer. The CIC algorithm is optimized for Antarctic sky conditions and results in a total hit rate of almost 0.98, where 1.0 is a perfect score, for the identification of the clear-sky, ice cloud, and mixed-phase cloud classes. Scene truth is provided by lidar measurements that are concurrent with REFIR-PAD. The CIC approach demonstrates the key role of far-infrared spectral measurements for clear–cloud discrimination and for cloud phase classification. Mean annual occurrences are 72.3 %, 24.9 %, and 2.7 % for clear sky, ice clouds, and mixed-phase clouds, respectively, with an inter-annual variability of a few percent. The seasonal occurrence of clear sky shows a minimum in winter (66.8 %) and maxima (75 %–76 %) during intermediate seasons. In winter the mean surface temperature is about 9 ^∘C colder in clear conditions than when ice clouds are present. Mixed-phase clouds are observed only in the warm season; in summer they amount to more than one-third of total observed clouds. Their occurrence is correlated with warmer surface temperatures. In the austral summer, the mean surface air temperature is about 5 ^∘C warmer when clouds are present than in clear-sky conditions. This difference is larger during the night than in daylight hours, likely due to increased solar warming. Monthly mean results are compared to cloud occurrence and fraction derived from gridded (Level 3) satellite products from both passive and active sensors. The differences observed among the considered products and the CIC results are analyzed in terms of footprint sizes and sensors' sensitivities to cloud optical and geometrical features. The comparison highlights the ability of the CIC–REFIR-PAD synergy to identify multiple cloud conditions and study their variability at different timescales.

中文翻译：

---

南极高原的冰和混合相云统计

南极高原康科迪亚站上空晴空、冰云和混合相云的统计数据提供了多个时间尺度，并结合在地表测量的同步气象参数进行了分析。结果是通过将机器学习云识别和分类 (CIC) 代码应用到 4 年获得的2012 年至 2015 年间下涌高光谱分辨率辐射的测量结果，由远红外辐射探测器测量 - 应用和开发原型 (REFIR-PAD) 光谱辐射计。CIC 算法针对南极天空条件进行了优化，总命中率接近 0.98，其中 1.0 是一个完美的分数，用于识别晴天、冰云和混合相云类。场景真实值由与 REFIR-PAD 并行的激光雷达测量提供。CIC 方法证明了远红外光谱测量对于晴云鉴别和云相分类的关键作用。年平均发生率为72.3  %、 24.9  %和 2.7  %分别为晴空、冰云和混合相云，年际变化为百分之几。晴空的季节性出现在冬季最小（66.8  %），在中间季节最大（75  % –76  %）。在冬季，晴朗条件下的平均地表温度比存在冰云时要低约 9  ^∘ C。混合相云仅在暖季观察到；在夏季，它们占观测到的云总数的三分之一以上。它们的发生与表面温度升高有关。在南方夏季，平均地表气温约为 5  ^∘ C有云时比晴天时更暖和。这种差异在夜间比白天更大，可能是由于太阳变暖加剧。将月平均结果与来自无源和有源传感器的网格（3 级）卫星产品的云发生率和分数进行比较。在所考虑的产品和 CIC 结果之间观察到的差异在足迹大小和传感器对云光学和几何特征的敏感性方面进行了分析。该比较突出了 CIC-REFIR-PAD 协同识别多个云条件并研究其在不同时间尺度上的可变性的能力。