Successive re-calibrations were implemented in Landsat-8 TIRS data since launch. This paper evaluates the performances of both: (1) these re-calibrations, up to the last calibration update announced for TIRS data in the next Landsat Collection 2; and (2) single-channel (SC) corrections and split-window (SW) algorithms to retrieve land surface temperature (LST) from TIRS data. A robust and accurate multi-year (2014–2019) set of reference ground data were used, which included thermal infrared (TIR) radiance measurements taken along transects in a uniform and thermally homogeneous rice paddy area, but also emissivity measurements for the different ground covers at the site through the year. The calibration results showed significant biases at the site for data after the 2014 reprocessing, but negligible biases and root-mean-square differences (RMSDs) <1.5 K were obtained when using the current TIRS data in Collection 1 (i.e., data after the 2017 reprocessing). The last announced calibration update mainly introduced differences in biases, improving slightly the results for band 10 and presenting a calibration response difference between bands. The SC corrections showed negligible LST biases for both bands and the lowest RMSDs (<1.6 K) when using the band 10 data in the current Collection 1, and the bias disappeared for this band after applying the calibration update. Three of the seventeen different SW equations evaluated in the paper showed negligible biases and LST RMSDs lower than or equal to 0.8 K. These three SW algorithms are mainly recommended for users of the current TIRS data in Collection 1; one of them being that proposed to generate a SW LST product in the future Collection 3. Finally, bias differences around 1.4 K were shown in the results of the SW algorithms after applying the calibration update announced for Collection 2.

自发射以来，已在Landsat-8 TIRS数据中进行了连续的重新校准。本文评估了这两种方法的性能：（1）这些重新校准，直到下一个Landsat Collection 2中宣布的TIRS数据的最新校准更新为止；（2）单通道（SC）校正和分割窗口（SW）算法，以从TIRS数据中检索地表温度（LST）。使用了一组可靠且准确的多年（2014-2019年）参考地面数据，其中包括在均匀且热均质的稻田中沿样条进行的热红外（TIR）辐射测量，以及不同地面的发射率测量全年覆盖该站点。校正结果显示，2014年重新处理后，现场数据存在明显偏差，但是当使用集合1中的当前TIRS数据（即2017年再处理后的数据）时，获得的偏差和均方根差（RMSD）<1.5 K可以忽略不计。上一次公布的校准更新主要引入了偏差的差异，从而稍微改善了频段10的结果，并给出了频段之间的校准响应差异。当使用当前Collection 1中的波段10数据时，SC校正显示两个波段的LST偏置都可以忽略不计，最低RMSD（<1.6 K），并且在应用校准更新后该波段的偏置消失了。本文评估的17个不同软件方程中的3个显示出可忽略的偏差，并且LST RMSD小于或等于0.8K。这3个软件算法主要推荐给Collection 1中当前TIRS数据的用户使用。