Uncertainty quantification in land surface temperature retrieved from Himawari-8/AHI data by operational algorithms
Uncertainty quantification in land surface temperature retrieved from Himawari-8/AHI data by operational algorithms
复制标题
DOI：
10.1016/j.isprsjprs.2022.07.008
复制
发表时间：
2022-09
影响因子：
12.7
通讯作者：
Y. Yamamoto;K. Ichii;Y. Ryu;Minseok Kang;S. Murayama
中科院分区：
文献类型：
--
作者：
Y. Yamamoto;K. Ichii;Y. Ryu;Minseok Kang;S. Murayama
Himawari-8是新一代地球静止卫星，能够以中等空间分辨率反演亚小时陆地表面温度，为监测亚洲和大洋洲的热环境提供了新的尺度。本研究评估的不确定性LST检索的三个业务算法从先进的Himawari成像仪（AHI）的数据。比较了两种非线性分裂窗算法（WAM算法和WAN算法）和一种非线性三带算法（YAM算法）。首先，利用大气辐射传输模型模拟了不同陆气条件下输入参数误差对反演地表温度的影响。此后，检索LST从实际的AHI数据进行了评估，使用现场观测从AsiaFlux和OzFlux网络和ECOSYSTEM空间站（ECOSTRESS）LST的星载热辐射计实验。仿真结果表明，YAM算法保持了最高的精度，而WAN算法具有最高的鲁棒性输入错误。YAM算法具有最小的总误差，包括在广泛的检索条件下的输入错误。通过来自12个站点的原位LST对三种算法进行的验证显示，所有站点的夜间平均RMSE为1.7 °C，半干旱和潮湿站点的白天平均RMSE分别约为3.0 °C和2.0 °C。这些精度与具有较高空间分辨率的LST产品（如中等分辨率成像光谱仪和Landsat）的精度相当。在Himawari-8盘中，在具有极高视角、温度和湿度的区域（例如，中国北方、澳大利亚和东南亚）。此外，AHI LSTs表现出更密切的协议与ECOSTRESS相比，在原地LSTs，这表明ECOSTRESS的实用性评估的昼夜LSTs来自地球静止卫星。由此产生的LST产品及其误差特性的知识，有可能提高地球能量和水循环的基础上提高准确性和鲁棒性的集体理解。
Himawari-8, a new-generation geostationary satellite, can retrieve sub-hourly land surface temperatures (LSTs) with moderate spatial resolution, providing a new scale for monitoring the thermal environment in Asia and Oceania. This study evaluated uncertainties of LSTs retrieved by three operational algorithms from Advanced Himawari Imager (AHI) data. We compared two nonlinear split-window algorithms (SOB and WAN algorithms) and one nonlinear three-band algorithm (YAM algorithm). First, the error characteristics of the retrieved LSTs caused by the input parameter errors were simulated under various land-atmospheric conditions using an atmospheric radiative transfer model. Thereafter, retrieved LSTs from actual AHI data were evaluated using in-situ observations from AsiaFlux and OzFlux networks and the ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) LSTs. The simulated results showed that the YAM algorithm maintained the highest accuracy, whereas the WAN algorithm had the highest robustness to input errors. The YAM algorithm had the smallest total error including input errors over a wide range of retrieval conditions. Validation of the three algorithms via in-situ LSTs from 12 sites revealed nighttime mean RMSEs for all sites of ∼1.7 °C, and daytime mean RMSEs for semi-arid and humid sites of approximately 3.0 °C and 2.0 °C, respectively. These are comparable to the accuracies reported for LST products with higher spatial resolutions, such as the Moderate Resolution Imaging Spectroradiometer and Landsat. Within the Himawari-8 disk, the estimation error of the YAM algorithm was ∼1.0 °C lower than those of the SOB and WAN algorithms in regions with extremely high viewing angle, temperature, and humidity (e.g., northern China, Australia, and Southeast Asia). Furthermore, AHI LSTs showed closer agreement with ECOSTRESS compared to in-situ LSTs, suggesting the usefulness of ECOSTRESS for assessing the diurnal LSTs derived from geostationary satellites. The resulting LST products and the knowledge of their error characteristics have the potential to improve the collective understanding of terrestrial energy and water cycles based on improved accuracy and robustness.