Abstract Satellites produce global monitoring data, while field measurements are made at a local station over the land. Due to difference in scale, it has been a challenge how to define and correct the satellite retrieval biases. Although the relative approach of cumulative distribution functions (CDF) matching compares a long-term climatology of reference data with that of satellite data, it does not mitigate the retrieval biases generated from Instantaneous Field of View (IFOV) measurements over short timescales. As an alternative, we suggest stochastic retrievals (using probabilistic distribution function) to reduce the dry bias in soil moisture retrievals from the satellite SMOS (Soil Moisture and Ocean Salinity) that occurs at the time scale of several days. Rank Probability Skill Score (RPSS) is also proposed as non-local Root Mean Square Errors (RMSEs) of a probabilistic version to optimize stochastic retrievals. With this approach, the time-averaged RMSEs of retrieved SMOS soil moisture is reduced from 0.072 to 0.035 m3/m3. Dry bias also decreases from −0.055 to −0.020 m3/m3. As the proposed approach does not rely on local field measurements, it has a potential as a global operational scheme.

中文翻译：

---

土壤水分海洋盐度 (SMOS) 土壤水分干偏差的自校正

摘要 卫星产生全球监测数据，而现场测量则在陆地上的本地站进行。由于规模差异，如何定义和纠正卫星检索偏差一直是一个挑战。尽管累积分布函数 (CDF) 匹配的相对方法将参考数据的长期气候学与卫星数据的气候学进行了比较，但它并不能减轻短期视场 (IFOV) 测量产生的反演偏差。作为替代方案，我们建议使用随机反演（使用概率分布函数）来减少来自卫星 SMOS（土壤湿度和海洋盐度）的土壤湿度反演在几天的时间尺度上发生的干偏差。Rank Probability Skill Score (RPSS) 也被提议作为概率版本的非局部均方根误差 (RMSE)，以优化随机检索。通过这种方法，检索到的 SMOS 土壤水分的时间平均 RMSE 从 0.072 降低到 0.035 m3/m3。干偏压也从 -0.055 降低到 -0.020 m3/m3。由于所提出的方法不依赖于局部现场测量，因此它具有作为全球操作方案的潜力。