Abstract The landscape freeze/thaw (FT) state plays an important role in local, regional and global weather and climate, but is difficult to monitor. The Soil Moisture Active Passive (SMAP) satellite mission provides hemispheric estimates of landscape FT state at a spatial resolution of approximately 36 2 km 2 . Previous validation studies of SMAP and other satellite FT products have compared satellite retrievals with point estimates obtained from in-situ measurements of air and/or soil temperature. Differences between the two are attributed to errors in the satellite retrieval. However, significant differences can occur between satellite and in-situ estimates solely due to differences in scale between the measurements; these differences can be viewed as ‘representativeness errors’ in the in-situ product, caused by using a point estimate to represent a large-scale spatial average. Most previous validation studies of landscape FT state have neglected representativeness errors entirely, resulting in conservative estimates of satellite retrieval skill. In this study, we use a variant of triple collocation called ‘categorical triple collocation’ – a technique that uses model, satellite and in-situ estimates to obtain relative performance rankings of all three products, without neglecting representativeness errors – to validate the SMAP landscape FT product. Performance rankings are obtained for nine sites at northern latitudes. We also investigate differences between using air or soil temperatures to estimate FT state, and between using morning (6 AM) or evening (6 PM) estimates. Overall, at most sites, the SMAP product or in-situ FT measurement is ranked first, and the model FT product is ranked last (although rankings vary across sites). These results suggest SMAP is adding value to model simulations, providing higher-accuracy estimates of landscape FT states compared to models and, in some cases, even in-situ estimates, when representativeness errors are properly accounted for in the validation analysis.

中文翻译：

---

使用分类三重搭配验证 SMAP 冷冻/解冻产品

摘要 景观冻融（FT）状态在局地、区域和全球天气和气候中起着重要作用，但难以监测。土壤湿度主动被动 (SMAP) 卫星任务以大约 36 2 km 2 的空间分辨率提供景观 FT 状态的半球估计。先前对 SMAP 和其他卫星 FT 产品的验证研究将卫星反演与从空气和/或土壤温度的原位测量获得的点估计进行了比较。两者之间的差异归因于卫星检索中的错误。然而，仅由于测量之间的规模差异，卫星和原位估计之间可能会出现显着差异；这些差异可以被视为原位产品中的“代表性错误”，由使用点估计来表示大规模空间平均值引起的。大多数先前对景观 FT 状态的验证研究完全忽略了代表性错误，导致对卫星检索技能的保守估计。在本研究中，我们使用一种称为“分类三重搭配”的三重搭配变体——一种使用模型、卫星和原位估计来获得所有三种产品的相对性能排名的技术，而不会忽略代表性错误——来验证 SMAP 格局FT产品。获得了北纬九个站点的性能排名。我们还研究了使用空气或土壤温度来估计 FT 状态与使用早上（早上 6 点）或晚上（下午 6 点）估计之间的差异。总的来说，在大多数网站上，SMAP 产品或原位 FT 测量排名第一，模型 FT 产品排名最后（尽管排名因站点而异）。这些结果表明 SMAP 正在为模型模拟增加价值，与模型相比，提供更准确的景观 FT 状态估计，在某些情况下，甚至在验证分析中正确考虑代表性错误时，甚至可以提供原位估计。