Abstract. Surface air temperature (SAT) is a key indicator of global warming and plays an important role in glacier melting. On the Tibetan Plateau (TP), there exist a large number of glaciers. However, station SAT observations on these glaciers are extremely scarce, and moreover the available ones are characterized by short time series, which substantively hinder our deep understanding of glacier dynamics due to climate changes on the TP. In this study, an ensemble learning model is constructed and trained to estimate glacial SATs with a spatial resolution of 1 km × 1 km from 2002 to 2020 using monthly MODIS land surface temperature products and many auxiliary variables, such as vegetation index, satellite overpass time and air pressure. The satellite-estimated glacial SATs are validated against SAT observations at glacier validation stations. Then, long-term (1961–2020) glacial SATs on the TP are reconstructed by temporally extending the satellite SAT estimates through Bayesian linear regression. The long-term glacial SAT estimates are validated with root mean squared error, mean bias error, and determination coefficient being 1.61 °C, 0.21 °C, and 0.93, respectively. The comparisons are conducted with other satellite SAT estimates and ERA5-Land reanalysis data over the validation glaciers, showing that the accuracy of our satellite glacial SATs and their temporal extensions are both higher. The preliminary analysis illustrates that the glaciers on the TP as a whole have been undergoing a fast warming but the warming exhibits a great spatial heterogeneity. Our dataset can contribute to the monitoring of glaciers’ warming, analysis of their evolution, etc. on the TP. The dataset is freely available from the National Tibetan Plateau Data Centre at https://doi.org/10.11888/Atmos.tpdc.272550 (Qin, 2022).

中文翻译：

---

青藏高原冰川上空 60 年（1961-2020）近地表气温数据集

摘要。地表气温（SAT）是全球变暖的关键指标，在冰川融化中起着重要作用。在青藏高原（TP）上，存在大量冰川。然而，这些冰川的站台SAT观测资料极为稀少，而且可用的观测资料具有时间序列短的特点，这极大地阻碍了我们对青藏高原气候变化引起的冰川动力学的深入了解。在这项研究中，构建并训练了一个集成学习模型，使用每月 MODIS 地表温度产品和许多辅助变量，如植被指数、卫星通过时间，估计 2002 年至 2020 年空间分辨率为 1 km × 1 km 的冰川 SAT。和气压。卫星估计的冰川 SAT 将根据冰川验证站的 SAT 观测结果进行验证。然后，TP上的长期（1961-2020）冰川SAT是通过贝叶斯线性回归对卫星SAT估计进行时间扩展来重建的。长期冰川 SAT 估计值得到验证，均方根误差、平均偏差误差和决定系数分别为 1.61°C、0.21°C 和 0.93。与验证冰川上的其他卫星 SAT 估计和 ERA5-Land 再分析数据进行了比较，表明我们的卫星冰川 SAT 及其时间扩展的准确性都更高。初步分析表明，青藏高原冰川整体经历了快速升温，但升温表现出很大的空间异质性。我们的数据集有助于监测青藏高原冰川的变暖、分析其演化等。