Abstract There is considerable demand for satellite observations that can support spatiotemporally continuous mapping of land surface temperature (LST) because of its strong relationships with many surface processes. However, the frequent occurrence of cloud cover induces a large blank area in current thermal infrared-based LST products. To effectively fill this blank area, a new method for reconstructing the cloud-covered LSTs of Terra Moderate Resolution Imaging Spectroradiometer (MODIS) daytime observations is described using random forest (RF) regression approach. The high temporal resolution of the Meteosat Second Generation (MSG) LST product assisted in identifying the temporal variations in cloud cover. The cumulative downward shortwave radiation flux (DSSF) was estimated as the solar radiation factor for each MODIS pixel based on the MSG DSSF product to represent the impact from cloud cover on incident solar radiation. The RF approach was used to fit an LST linking model based on the datasets collected from clear-sky pixels that depicted the complicated relationship between LST and the predictor variables, including the surface vegetation index (the normalized difference vegetation index and the enhanced vegetation index), normalized difference water index, solar radiation factor, surface albedo, surface elevation, surface slope, and latitude. The fitted model was then used to reconstruct the LSTs of cloud-covered pixels. The proposed method was applied to the Terra/MODIS daytime LST product for four days in 2015, spanning different seasons in southwestern Europe. A visual inspection indicated that the reconstructed LSTs thoroughly captured the distribution of surface temperature associated with surface vegetation cover, solar radiation, and topography. The reconstructed LSTs showed similar spatial pattern according to the comparison with clear-sky LSTs from temporally adjacent days. In addition, evaluations against Global Land Data Assimilation System (GLDAS) NOAH 0.25° 3-h LST data and reference LST data derived based on in-situ air temperature measurements showed that the reconstructed LSTs presented a stable and reliable performance. The coefficients of determination derived with the GLDAS LST data were all above 0.59 on the four examined days. These results indicate that the proposed method has a strong potential for reconstructing LSTs under cloud-covered conditions and can also accurately depict the spatial patterns of LST.

中文翻译：

---

使用综合 MODIS/Terra 土地产品和 MSG 地球静止卫星数据重建云覆盖条件下的白天地表温度

摘要 由于地表温度与许多地表过程密切相关，因此对能够支持地表温度 (LST) 时空连续制图的卫星观测有相当大的需求。然而，云层的频繁出现导致当前基于热红外的 LST 产品存在很大的空白区域。为了有效填补这一空白区域，本文采用随机森林 (RF) 回归方法描述了一种重建 Terra 中分辨率成像光谱仪 (MODIS) 白天观测的云覆盖 LST 的新方法。第二代气象卫星 (MSG) LST 产品的高时间分辨率有助于识别云量的时间变化。累积向下短波辐射通量 (DSSF) 被估计为基于 MSG DSSF 产品的每个 MODIS 像素的太阳辐射因子，以表示云层对入射太阳辐射的影响。RF方法用于基于从晴空像素收集的数据集拟合LST链接模型，该数据集描述了LST与预测变量之间的复杂关系，包括地表植被指数（归一化差异植被指数和增强植被指数） ，归一化差异水指数、太阳辐射因子、地表反照率、地表高程、地表坡度和纬度。然后使用拟合模型来重建云覆盖像素的 LST。2015年提出的方法应用于Terra/MODIS日间LST产品四天，跨越欧洲西南部的不同季节。目视检查表明，重建的 LST 彻底捕获了与地表植被覆盖、太阳辐射和地形相关的地表温度分布。根据与时间相邻的晴天 LST 的比较，重建的 LST 显示出相似的空间模式。此外，对全球陆地数据同化系统 (GLDAS) NOAH 0.25° 3-h LST 数据和基于原位气温测量的参考 LST 数据的评估表明，重建的 LST 表现出稳定可靠的性能。由 GLDAS LST 数据得出的决定系数在四个检查日均高于 0.59。