A CO2-free cloud mask from IASI radiances for climate applications (from IASI/Metop-A)

The IASI Cloud Detection product (IASI CLD) is a cloud mask available at the IASI pixel level (L2) that was developed for climate applications purposes. The algorithm is detailed in Whitburn et al. (2022, Atmospheric Measurement Techniques). It combines (1) a high sensitivity to cloud detection, (2) a very good consistency over time and between the three IASI instruments and (3) simplicity in its parametrization. The method is based on the use of a supervised neural network using the version 6.5 of the operational IASI L2 cloud product as a reference dataset. As input parameters, it relies on IASI radiance information only. The consistency over time is ensured by careful selection of the IASI input channels, avoiding especially the spectral regions affected by long-lived absorbers and water vapor. The version number is 1.0. IASI/Metop CLD data are available daily from 2008 to 2020 for Metop-A, Metop-B and Metop-C in hdf format. For Metop-A before 2016, the algorithm was applied to reprocessed IASI L1C data for more homogeneity in the data set. Each file includes the latitude, the longitude, the neural network’s retrieval, and the IASI overpass time. Note that the retrieval of the neural network provides a value comprised between 0 and 1. This value is converted in a cloud mask using an appropriate threshold. We recommend using the thresholds provided in Whitburn et al. (2022) (i.e. 0.175 for land and 0.275 for oceans) but these can be adjusted depending on the application.

IASI云检测产品（IASI CLD）是一款面向气候应用开发的IASI像元级（L2）云掩膜产品。其算法细节详见Whitburn等人2022年发表于《Atmospheric Measurement Techniques》的研究论文。该算法具备三大核心特性：一是对云检测拥有高灵敏度，二是在时间维度及三台IASI仪器间具有优异的一致性，三是参数化设置简洁易用。
该方法基于监督式神经网络（supervised neural network）构建，以运行型IASI L2云产品的6.5版本作为参考数据集，输入参数仅依赖IASI辐射亮度信息。为保障时间一致性，研究团队通过精心筛选IASI输入通道，刻意规避了受长效吸收体及水蒸气影响的光谱区域。本产品版本号为1.0。
IASI/Metop CLD数据以HDF格式存储，自2008年至2020年每日均可获取，覆盖Metop-A、Metop-B及Metop-C三颗卫星。针对2016年之前的Metop-A数据，算法采用了经过再处理的IASI L1C数据，以提升数据集的整体均一性。每个数据文件均包含纬度、经度、神经网络反演结果及IASI过境时间。
需特别说明的是，神经网络的反演输出值介于0至1之间，该值需通过合适的阈值转换为云掩膜。我们建议采用Whitburn等人（2022）中提出的阈值（陆地为0.175，海洋为0.275），但亦可根据具体应用场景自行调整阈值。