Clear-sky detection (CSD) is a crucial process in numerous solar energy applications. Many CSD models have been proposed over the years, though model performance is generally found unsatisfactory for worldwide use. We demonstrate this qualitatively on 22 CSD models at five climatologically-diverse radiometric stations; all exhibit one or more limitations: (1) unreliability at high zenith; (2) unrealistic “clear” periods immediately before or after cloudy conditions; (3) relaxed (short-term false positives); (4) over-relaxed during clear conditions (longer-term false positives); (5) conservative (short-term false negatives); and (6) over-conservative during clear conditions (longer-term false negatives).

A new globally applicable CSD methodology is proposed for a posteriori detection of apparent “cloudless sky” conditions on 1-min irradiance time series, named the Bright-Sun model. This new tool requires measured global horizontal irradiance (GHI) and diffuse horizontal irradiance (DIF), and consists of three stages: (1) clear-sky irradiance optimisation, (2) tri-component CSD analysis with the newly derived Modified-Reno method, and (3) a cascading durational filters to determine periods of apparent cloudless sky. Through qualitative evaluation and exploring sensitivity to clear-sky model selection, the Bright-Sun model does not suffer any of the aforementioned limitations at any of the five stations, despite their distinctive climates. Due to the significant influence of bright or dark clouds on DIF, which have much lower impact on GHI, the new model also exhibits extra discretionary power by including analysis on DIF and can thus identify apparently clear periods with zero or near-zero cloudiness.

The Bright-Sun CSD model is coded in Matlab® and freely available (future releases in R and Python are anticipated). A script is attached as supplementary material in the original form. For a supported and version controlled release of the Bright-Sun model, as well as other CSD models mentioned within this document, the reader can refer to the CSD Library at https://jamiembright.github.io/csd-library/.

晴空检测（CSD）是众多太阳能应用中的关键过程。多年来，已经提出了许多CSD模型，尽管通常在全球范围内都无法令人满意地获得模型性能。我们在五个气候多样的辐射站的22个CSD模型上定性地证明了这一点；所有这些都表现出一个或多个限制：（1）高天顶下的不可靠性；（2）在阴天之前或之后不切实际的“晴朗”时期；（3）放松（短期误报）；（4）在明确条件下过度放松（长期误报）；（5）保守（短期假阴性）；（6）在明确的条件下（长期的假阴性）过于保守。

提出了一种新的全球适用的CSD方法，该方法用于在1分钟辐照时间序列上后验检测明显的“无云的天空”条件，称为Bright-Sun模型。此新工具需要测量的整体水平辐照度（GHI）和弥散水平辐照度（DIF），并且包括三个阶段：（1）晴空辐照度优化；（2）使用最新衍生的Modified-Reno方法进行三组分CSD分析，以及（3）级联的持续时间滤波器来确定无云天空的周期。通过定性评估并探索对晴空模型选择的敏感性，Bright-Sun尽管它们的气候非常独特，但是在五个站点中的任何一个站点上，该模型都不会受到任何上述限制。由于亮云或暗云对DIF的显着影响，而DIF对GHI的影响要小得多，因此新模型还包括对DIF的分析，从而显示了更大的自由裁量权，因此可以识别出零或接近零浊度的明显晴朗时段。

在明亮的孙CSD模型在MATLAB®编码，并免费提供（未来R和Python的版本预计）。脚本以原始形式作为补充材料随附。有关Bright-Sun模型以及本文档中提到的其他CSD模型的受支持且受版本控制的发行版，读者可以参考https://jamiembright.github.io/csd-library/上的CSD库。