欧空局冰雪气候变化倡议(Snow_cci)1982-2019年低温气候中的部分积雪数据v1.0

该数据集包含由欧空局气候变化倡议计划雪项目制作的 CryoClim 每日积雪覆盖率(地面积雪)产品。
地面积雪分量(FSC)表示从太空观测到的陆地表面积雪面积，在森林地区，由于树木遮挡了林冠下的地面积雪，因此地面积雪分量得到了补偿。每个网格单元的 FSC 以百分比(%)表示。
除南极洲和格陵兰冰盖外，全球雪_cci CryoClim 积雪覆盖率(FSC)产品的网格大小为 0.05°(约 5 公里)，适用于所有陆地地区。格陵兰沿海地区也包括在内。
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，结合了光学和被动微波辐射计(PMR)数据的观测结果。该产品将高级甚高分辨率辐射计传感器数据的历史记录与来自 SMMR、SSM/I 和 SSMIS 传感器的 PMR 数据相结合。
CryoClim FSC 气候数据记录的总体目标是提供覆盖全球的最长雪盖范围时间序列之一，且不受云层和极夜的影响。为实现这一目标，利用传感器融合算法，利用光学和被动微波辐射计对雪观测的最佳特征，生成了一致的全球FSC产品时间序列(Solberg等人，2014年，2015年；Rudjord等人，2015年)。
snow_cci 项目将最初的 CryoClim 二进制产品提升为 FSC 产品。专题变量表示地面积雪(SCFG)。
NOAA-7、-9、-11、-14、-16、-18、-19 号卫星上的 AVHRR 传感器被用作光学数据源，Nimbus-7、DMSP F8、DMSP F10、DMSP F11、DMSP F13、DMSP F14、DMSP F15、DMSP F16、DMSP F17 和 DMSP F18 号卫星上的 SMMR、SSM/I 和 SSMIS 传感器分别被用作 PMR 数据源。为了尽可能提高输入数据的质量，我们使用了欧洲气象卫星应用组织 CM SAF 为 AVHRR(Karlson 等人，2020 年)和 PMR(Fenning 等人，2017 年)开发的基本气候数据记录(FCDR)。
光学算法组件处理来自 AVHRR GAC 的所有可用扫面。计算以贝叶斯方法为基础，使用一组信号(仪器信道组合)和统计系数。对于扫描路径的每个像素，都会估算出雪、裸地和云等地表类别的概率。统计系数基于对地表类别在电磁波谱不同部分的典型行为的预先了解。
PMR 算法也是基于贝叶斯估计方法。对于 SSM/I 和 SSMIS，定义了四个雪级来模拟雪面状态。对于 SMMR，考虑了两个等级。该算法根据 PMR 测量结果估算出每个雪级的概率。为了提高贝叶斯算法的性能，还加入了土地覆盖数据。这样就有可能为每种土地覆被状况构建一个贝叶斯估算器。
多传感器多时融合算法(Rudjord 等人，2015 年；Solberg 等人，2017 年)基于隐马尔可夫模型(HMM)，根据 PMR 和光学传感器的观测结果模拟雪的状态。其基本思想是通过状态模型模拟雪季中雪面所经历的状态。这些状态无法直接观测到，但遥感观测数据提供了描述雪况的数据，这些数据与雪况相关。HMM 解决方案不仅是一个多传感器模型，也是一个多时态模型。随时间变化的状态序列需要遵循一定的优化标准。
snow_cci 将二进制雪覆盖率提升到分数雪覆盖率主要遵循两条路径： 首先，我们引入了更多的 HMM 状态，以便将积雪覆盖率划分为 10%的分数积雪覆盖率区间。然而，引入 100 个主状态来获得 1%的 FSC 间隔并不能得到一个稳定的模型。为了获得更高的精度，我们使用二级维特比序列在 HMM 状态之间进行插值。这两个概率作为权重用于估算 FSC。
根据 2000 年的土地覆盖 CCI 数据集，对永久冰雪和水域进行了屏蔽。这两个类别分别汇总到 FSC 产品的网格大小。如果网格单元中超过 30% 的区域被归类为水域，则水域区域将被屏蔽；如果在汇总地图中超过 50% 的区域被识别为永久冰雪区域，则永久冰雪区域将被屏蔽。观测到的陆地面积的产品不确定性以辅助变量中每个网格单元的无偏均方根误差 (RMSE) 的形式提供。
FSC 产品旨在满足从事冰冻圈和气候研究与监测活动的用户的需求，包括变异性和趋势评估、气候建模以及水文学、气象学和生物学等方面。
挪威计算中心 (Norsk Regnesentral, NR) 与挪威气象研究所 (MET Norway) 共同负责 FSC 产品的开发和卫星数据的生成。ENVEO IT GmbH 开发并准备了用于产品生成的所有辅助数据集。
在整个时间序列中，有 27 天既没有光学数据，也没有 PMR 数据。这些是单个天数，而不是连续的天数系列。多传感器时间序列算法会根据数据缺失前后的天数，对雪覆盖进行最佳估计，从而解决这一问题。对于没有数据的天数，只要它们只是单个天数，就不会降低雪地图的质量。
估算与 FSC 地图相关的不确定性的算法需要观测与 FSC 产品时间戳相同的协变量。这些协变量部分基于 PMR 传感器的数据。因此，无法对缺乏 PMR 采集的日子进行不确定性估算。缺乏 PMR 的天数大多出现在 1982-1988 年间(53 天)，之后只有两例(2008 年)。

This dataset contains the CryoClim daily fractional snow cover (ground snow) product developed by the European Space Agency (ESA) Climate Change Initiative (CCI) Snow Project.
The Fractional Snow Cover (FSC) refers to the snow-covered land surface area observed from space; in forested regions, ground snow under the forest canopy is obscured by trees, so the FSC value is adjusted to account for this occlusion. The FSC for each grid cell is expressed as a percentage (%).
Excluding Antarctica and the Greenland Ice Sheet, the global snow_cci CryoClim FSC product has a grid size of 0.05° (approximately 5 km), and is applicable to all terrestrial areas, including coastal Greenland.
The CryoClim FSC time series provides daily products spanning the period from 1982 to 2019.
The CryoClim FSC product is built on a multi-sensor time-series fusion algorithm that combines observations from optical and Passive Microwave Radiometer (PMR) data. It integrates historical records of Advanced Very-High-Resolution Radiometer (AVHRR) sensor data with PMR data from SMMR, SSM/I, and SSMIS sensors.
The overarching goal of the CryoClim FSC Climate Data Record is to deliver one of the longest global snow cover extent time series, free from the impacts of cloud cover and polar night. To achieve this, a sensor fusion algorithm leveraging the optimal characteristics of optical and PMR snow observations was used to generate a consistent global FSC product time series (Solberg et al., 2014, 2015; Rudjord et al., 2015).
The snow_cci project upgraded the original CryoClim binary snow product to an FSC product, with the thematic variable representing ground snow (SCFG).
AVHRR sensors onboard NOAA-7, -9, -11, -14, -16, -18, and -19 satellites were used as the optical data source, while SMMR, SSM/I, and SSMIS sensors onboard Nimbus-7, DMSP F8, DMSP F10, DMSP F11, DMSP F13, DMSP F14, DMSP F15, DMSP F16, DMSP F17, and DMSP F18 satellites were used as the PMR data source. To maximize the quality of input data, Fundamental Climate Data Records (FCDR) developed by the EUMETSAT Satellite Application Facility on Climate Monitoring (CM SAF) for AVHRR (Karlson et al., 2020) and PMR (Fenning et al., 2017) were utilized.
The optical algorithm component processes all available swaths from AVHRR GAC. Calculations are based on a Bayesian method, using a set of signals (instrument channel combinations) and statistical coefficients. For each pixel in a swath, the probabilities of surface classes including snow, bare ground, and clouds are estimated. The statistical coefficients are derived from prior knowledge of the typical spectral behavior of each surface class across the electromagnetic spectrum.
The PMR algorithm is also based on Bayesian estimation. For SSM/I and SSMIS, four snow classes are defined to simulate snow surface conditions, while two classes are considered for SMMR. The algorithm estimates the probability of each snow class based on PMR measurements. To improve the performance of the Bayesian algorithm, land cover data was incorporated, enabling the construction of a Bayesian estimator for each land cover condition.
The multi-sensor, multi-temporal fusion algorithm (Rudjord et al., 2015; Solberg et al., 2017) is based on the Hidden Markov Model (HMM), which simulates snow conditions using observations from PMR and optical sensors. The core idea is to model the states that snow surfaces undergo during the snow season via a state model. These states are not directly observable, but remote sensing observations provide data describing snow conditions that correlate with the actual snow state. The HMM solution is both a multi-sensor and multi-temporal model: the sequence of states over time must follow specific optimization criteria.
The snow_cci project upgraded binary snow cover to fractional snow cover through two primary pathways: firstly, additional HMM states were introduced to partition snow cover into 10% fractional snow cover intervals. However, introducing 100 primary states to achieve 1% FSC bins would result in an unstable model. To achieve higher precision, a two-level Viterbi sequence was used to interpolate between HMM states. These two probabilities are used as weights to estimate the FSC value.
Permanent ice and water bodies were masked based on the 2000 Land Cover CCI dataset, with both classes aggregated to the grid size of the FSC product. A grid cell will have its water body area masked if more than 30% of the cell is classified as water, and its permanent ice area masked if more than 50% of the cell is identified as permanent ice in the aggregated map. The product uncertainty for observed terrestrial areas is provided as the unbiased Root Mean Square Error (RMSE) for each grid cell in the auxiliary variables.
The FSC product is designed to meet the needs of users engaged in cryosphere and climate research and monitoring, including variability and trend assessment, climate modeling, and applications in hydrology, meteorology, and biology.
The Norwegian Computing Center (Norsk Regnesentral, NR) and the Norwegian Meteorological Institute (MET Norway) are jointly responsible for the development of the FSC product and the generation of satellite data. ENVEO IT GmbH developed and prepared all auxiliary datasets used for product generation.
Over the entire time series, there are 27 days with neither optical nor PMR data. These are single, non-consecutive days. The multi-sensor time-series algorithm addresses this issue by generating optimal snow cover estimates based on the days immediately before and after the missing data period. For single missing days, this does not degrade the quality of the snow maps.
The algorithm used to estimate uncertainty associated with FSC maps requires covariates observed at the same timestamp as the FSC product. These covariates are partially based on PMR sensor data, so uncertainty estimates cannot be generated for days without PMR acquisitions. Days without PMR data mostly occurred between 1982 and 1988 (53 days), with only two additional cases in 2008.