ESA Snow Climate Change Initiative (Snow_cci): Fractional Snow Cover in CryoClim, v1.0

This dataset contains the CryoClim Daily Snow Cover Fraction (snow on ground) product, produced by the Snow project of the ESA Climate Change Initiative programme. Fractional snow cover (FSC) on the ground indicates the area of snow observed from space on land surfaces, in forested areas compensated for the effect of trees hiding the ground surface snow cover under the forest canopy. The FSC is given in percentage (%) per grid cell. The global snow_cci CryoClim fractional snow cover (FSC) product is available at 0.05° grid size (about 5 km) for all land areas, excluding Antarctica and Greenland ice sheet. The coastal zones of Greenland are included. The CryoClim FSC time series provides daily products for the period 1982 – 2019. The CryoClim FSC product is based on a multi-sensor time-series fusion algorithm combining observations by optical and passive microwave radiometer (PMR) data. The product combines an historical record of AVHRR sensor data with PMR data from the SMMR, SSM/I and SSMIS sensors. The overall aim of the CryoClim FSC climate data record is to provide one of the longest snow cover extent time series available with global coverage and without hindrance from clouds and polar night. This has been achieved by utilising the best features of optical and passive microwave radiometer observations of snow using a sensor-fusion algorithm generating a consistent time series of global FSC products (Solberg et al. 2014, 2015; Rudjord et al. 2015). The snow_cci project has advanced the original CryoClim binary product to an FSC product. The thematic variable represents snow on the ground (SCFG). AVHRR sensors aboard the satellites NOAA-7, -9, -11, -14, -16, -18, -19 have been used as the optical data source, and SMMR, SSM/I and SSMIS sensors aboard the Nimbus-7, DMSP F8, DMSP F10, DMSP F11, DMSP F13, DMSP F14, DMSP F15, DMSP F16, DMSP F17 and DMSP F18 satellites, respectively, have been used as PMR data source. To have the best possible input data quality, we have used fundamental climate data records (FCDRs) developed by EUMETSAT CM SAF for AVHRR (Karlson et al. 2020) and PMR (Fenning et al. 2017). The optical algorithm component processes all available swaths from AVHRR GAC. The calculations are based on a Bayesian approach using a set of signatures (instrument channel combinations) and statistical coefficients. For each pixel of the swath, the probabilities for the surface classes snow, bare ground and cloud are estimated. The statistical coefficients are based on pre-knowledge of the typical behaviour of the surface classes in the different parts of the electromagnetic spectrum. The algorithm for PMR is also based on a Bayesian estimation approach. For SSM/I and SSMIS four snow classes were defined to model the snow surface state. For SMMR two classes were considered. The algorithm estimates the probability for each snow class given the PMR measurements. Land cover data are included to improve the performance of the Bayesian algorithm. This made it possible to construct a Bayesian estimator for each land cover regime. The multi-sensor multi-temporal fusion algorithm (Rudjord et al. 2015; Solberg et al. 2017) is based on a hidden Markov model (HMM) simulating the snow states based on observations with PMR and optical sensors. The basic idea is to simulate the states the snow surface goes through during the snow season with a state model. The states are not directly observable, but the remote sensing observations give data describing the snow conditions, which are related to the snow states. The HMM solution represents not only a multi-sensor model but also a multi-temporal model. The sequence of states over time is conditioned to follow certain optimisation criteria. The advancement from binary to fractional snow cover carried out by snow_cci has followed two main paths: First, we introduced more HMM states to be able to classify the snow cover into 10% FSC intervals. However, introducing 100 primary states to obtain 1% FSC intervals would not give a stable model. For obtaining higher precision, we have interpolated between HMM states using a secondary Viterbi sequence. The two probabilities are used as weights to estimate the FSC. Permanent snow and ice, and water areas are masked based on the Land Cover CCI data set of the year 2000. Both classes were separately aggregated to the grid size of the FSC product. Water areas are masked if more than 30% of the grid cell is classified as water, permanent snow and ice areas are masked if more than 50% is identified as such areas in the aggregated map. The product uncertainty for observed land areas is provided as unbiased root mean square error (RMSE) per grid cell in the ancillary variable. The FSC product aims to serve the needs of users working with the cryosphere and climate research and monitoring activities, including the assessment of variability and trends, climate modelling and aspects of hydrology, meteorology, and biology. The Norwegian Computing Center (Norsk Regnesentral, NR) is together with the Norwegian Meteorological Institute (MET Norway) responsible for the FSC product development and generation from satellite data. ENVEO IT GmbH developed and prepared all auxiliary data sets used for the product generation. For the whole time series, there are 27 days with neither optical nor PMR retrieval. These are individual days and not series of days in a row. The multi-sensor time-series algorithm handles this by making a best estimate of snow cover, based on days both prior to and following after the lack of data. This will not reduce the quality of the snow maps much for days without data as long as they are just individual days. The algorithm estimating the uncertainty associated with the FSC maps needs observations of covariates from the same day as the time stamp of the FSC product. These covariates are partly based on data from PMR sensors. Hence, estimates of uncertainty could not be produced for days lacking PMR acquisitions. Most days lacking PMR are in the period 1982-1988 (53 days), and there are only two cases after that (in 2008).

本数据集包含CryoClim每日地表积雪覆盖率（snow on ground）产品，由欧洲空间局气候变化倡议计划（ESA Climate Change Initiative programme）的积雪项目生成。地表积雪覆盖率（Fractional Snow Cover, FSC）指从太空观测到的陆地表面积雪区域，其中林区需补偿树木遮蔽林冠下地表积雪的影响。FSC以每个网格单元的百分比（%）表示。

全球snow_cci CryoClim积雪覆盖率（FSC）产品的空间分辨率为0.05°网格（约5公里），覆盖除南极洲及格陵兰冰盖外的所有陆地区域，格陵兰沿海地带则包含在内。CryoClim FSC时间序列提供1982-2019年期间的每日产品。

CryoClim FSC产品基于多传感器时间序列融合算法，整合了光学传感器与被动微波辐射计（Passive Microwave Radiometer, PMR）的观测数据。该产品融合了高级甚高分辨率辐射计（Advanced Very High Resolution Radiometer, AVHRR）传感器数据的历史记录，以及来自扫描多通道微波辐射计（SMMR）、特殊微波成像仪（SSM/I）和特殊微波成像仪/探测器（SSMIS）的PMR数据。

CryoClim FSC气候数据记录的总体目标是提供目前可用的最长积雪范围时间序列之一，该序列覆盖全球且不受云层和极夜的影响。这一目标通过以下方式实现：利用传感器融合算法整合光学与被动微波辐射计积雪观测的优势，生成一致的全球FSC产品时间序列（Solberg等，2014、2015；Rudjord等，2015）。

snow_cci项目已将原始CryoClim二元产品升级为积雪覆盖率产品。该专题变量代表地表积雪（snow on ground, SCFG）。光学数据源采用搭载于NOAA-7、-9、-11、-14、-16、-18、-19卫星的AVHRR传感器；PMR数据源则分别采用搭载于Nimbus-7、DMSP F8、DMSP F10、DMSP F11、DMSP F13、DMSP F14、DMSP F15、DMSP F16、DMSP F17及DMSP F18卫星的SMMR、SSM/I和SSMIS传感器。

为确保输入数据质量最优，我们采用了欧洲气象卫星组织气候监测卫星应用设施（EUMETSAT CM SAF）为AVHRR（Karlson等，2020）和PMR（Fenning等，2017）开发的基础气候数据记录（Fundamental Climate Data Records, FCDRs）。光学算法模块处理AVHRR全球区域覆盖（Global Area Coverage, GAC）的所有可用扫描带。计算基于贝叶斯方法，采用一组特征（仪器通道组合）和统计系数。针对扫描带的每个像素，估算积雪、裸地和云三类地表的概率。统计系数基于不同电磁波谱区域内地表类别典型特征的先验知识。

PMR算法同样基于贝叶斯估计方法。针对SSM/I和SSMIS，定义了四类积雪以模拟积雪表面状态；针对SMMR，则考虑两类积雪。该算法基于PMR观测数据估算每类积雪的概率。引入土地覆盖数据以提升贝叶斯算法的性能，从而能够为每种土地覆盖类型构建贝叶斯估计器。

多传感器多时相融合算法（Rudjord等，2015；Solberg等，2017）基于隐马尔可夫模型（Hidden Markov Model, HMM），该模型利用PMR和光学传感器的观测数据模拟积雪状态。其核心思想是通过状态模型模拟积雪季节中积雪表面经历的状态变化。这些状态无法直接观测，但遥感观测提供了描述积雪状况的数据，而这些数据与积雪状态相关联。HMM解决方案不仅是多传感器模型，亦是多时相模型。随时间变化的状态序列需满足特定优化准则。

snow_cci项目将二元积雪产品升级为积雪覆盖率产品主要遵循两条路径：其一，引入更多HMM状态，以实现10%积雪覆盖率区间的分类；然而，引入100个主状态以获取1%积雪覆盖率区间会导致模型不稳定。为提高精度，我们采用次级维特比序列（Viterbi sequence）在HMM状态间进行插值，并将两种概率作为权重估算积雪覆盖率。

基于2000年土地覆盖CCI数据集，对永久积雪冰区及水域进行掩膜处理。这两类区域均单独聚合至积雪覆盖率产品的网格分辨率。若网格单元中超过30%被归类为水域，则对该单元进行水域掩膜；若聚合地图中超过50%被识别为永久积雪冰区，则对该区域进行永久积雪冰区掩膜。

观测陆地区域的产品不确定性以每个网格单元的无偏均方根误差（unbiased root mean square error, RMSE）形式在辅助变量中提供。积雪覆盖率产品旨在满足冰冻圈与气候研究及监测领域用户的需求，包括变率与趋势评估、气候建模以及水文学、气象学和生物学相关研究。

挪威计算中心（Norsk Regnesentral, NR）与挪威气象研究所（MET Norway）共同负责积雪覆盖率产品的开发及基于卫星数据的生成工作。ENVEO IT GmbH公司开发并制备了产品生成所需的所有辅助数据集。

在整个时间序列中，存在27天既无光学数据也无PMR数据反演结果，且这些均为单日，并非连续多日。多传感器时间序列算法通过基于缺失数据前后日期的信息，对积雪覆盖情况进行最优估算，以处理此类情况。只要缺失数据的日期为单日，就不会显著降低无数据日期积雪地图的质量。

估算积雪覆盖率地图相关不确定性的算法需要与积雪覆盖率产品时间戳同日的协变量观测数据，而这些协变量部分基于PMR传感器数据。因此，对于缺乏PMR数据采集的日期，无法生成不确定性估算结果。大多数缺乏PMR数据的日期集中在1982-1988年期间（共53天），此后仅出现两例（2008年）。