Land surface temperature (LST) is a key variable within Earth's climate system and a necessary input parameter required by numerous land–atmosphere models. It can be directly retrieved from satellite thermal infrared (TIR) observations, which contain many invalid pixels mainly caused by cloud contamination. To investigate the spatial and temporal variations in LST in China, long-term, high-quality, and spatiotemporally continuous LST datasets (i.e., all-weather LST) are urgently needed. Fusing satellite TIR LST and reanalysis datasets is a viable route to obtain long time-series all-weather LSTs. Among satellite TIR LSTs, the MODIS LST is the most commonly used, and a few corresponding all-weather LST products have been reported recently. However, the publicly reported all-weather LSTs were not available during the temporal gaps of MODIS between 2000 and 2002. In this study, we generated a daily (four observations per day) 1 km all-weather LST dataset for China's landmass and surrounding areas, the Thermal and Reanalysis Integrating Moderate-resolution Spatial-seamless (TRIMS) LST, which begins on the first day of the new millennium (1 January 2000). We used the enhanced reanalysis and thermal infrared remote sensing merging (E-RTM) method to generate the TRIMS LST dataset with the temporal gaps being filled, which had not been achieved by the original RTM method. Specifically, we developed two novel approaches, i.e., the random-forest-based spatiotemporal merging (RFSTM) approach and the time-sequential LST-based reconstruction (TSETR) approach, respectively, to produce Terra/MODIS-based and Aqua/MODIS-based TRIMS LSTs during the temporal gaps. We also conducted a thorough evaluation of the TRIMS LST. A comparison with the Global Land Data Assimilation System (GLDAS) and ERA5-Land LST demonstrates that the TRIMS LST has similar spatial patterns but a higher image quality, more spatial details, and no evident spatial discontinuities. The results outside the temporal gap show consistent comparisons of the TRIMS LST with the MODIS LST and the Advanced Along-Track Scanning Radiometer (AATSR) LST, with a mean bias deviation (MBD) of 0.09/0.37 K and a standard deviation of bias (SD) of 1.45/1.55 K. Validation based on the in situ LST at 19 ground sites indicates that the TRIMS LST has a mean bias error (MBE) ranging from −2.26 to 1.73 K and a root mean square error (RMSE) ranging from 0.80 to 3.68 K. There is no significant difference between the clear-sky and cloudy conditions. For the temporal gap, it is observed that RFSTM and TSETR perform similarly to the original RTM method. Additionally, the differences between Aqua and Terra remain stable throughout the temporal gap. The TRIMS LST has already been used by scientific communities in various applications such as soil moisture downscaling, evapotranspiration estimation, and urban heat island modeling. The TRIMS LST is freely and conveniently available at https://doi.org/10.11888/Meteoro.tpdc.271252 (Zhou et al., 2021).

Land surface temperature (LST) plays an important role in the research of climate change and various land surface processes. Before 2000, global LST products with relatively high temporal and spatial resolutions are scarce, despite a variety of operational satellite LST products. In this study, a global $\mathrm{0.05}{}^{\circ }\times \mathrm{0.05}{}^{\circ }$ historical LST product is generated from NOAA advanced very-high-resolution radiometer (AVHRR) data (1981–2000), which includes three data layers: (1) instantaneous LST, a product generated by integrating several split-window algorithms with a random forest (RF-SWA); (2) orbital-drift-corrected (ODC) LST, a drift-corrected version of RF-SWA LST; and (3) monthly averages of ODC LST. For an assumed maximum uncertainty in emissivity and column water vapor content of 0.04 and 1.0 g cm⁻², respectively, evaluated against the simulation dataset, the RF-SWA method has a mean bias error (MBE) of less than 0.10 K and a standard deviation (SD) of 1.10 K. To compensate for the influence of orbital drift on LST, the retrieved RF-SWA LST was normalized with an improved ODC method. The RF-SWA LST were validated with in situ LST from Surface Radiation Budget (SURFRAD) sites and water temperatures obtained from the National Data Buoy Center (NDBC). Against the in situ LST, the RF-SWA LST has a MBE of 0.03 K with a range of −1.59–2.71 K, and SD is 1.18 K with a range of 0.84–2.76 K. Since water temperature only changes slowly, the validation of ODC LST was limited to SURFRAD sites, for which the MBE is 0.54 K with a range of −1.05 to 3.01 K and SD is 3.57 K with a range of 2.34 to 3.69 K, indicating good product accuracy. As global historical datasets, the new AVHRR LST products are useful for filling the gaps in long-term LST data. Furthermore, the new LST products can be used as input to related land surface models and environmental applications. Furthermore, in support of the scientific research community, the datasets are freely available at https://doi.org/10.5281/zenodo.3934354 for RF-SWA LST (Ma et al., 2020a), https://doi.org/10.5281/zenodo.3936627 for ODC LST (Ma et al., 2020c), and https://doi.org/10.5281/zenodo.3936641 for monthly averaged LST (Ma et al., 2020b).

Neural networks, especially the latest deep learning, have exhibited good ability in estimating surface parameters from satellite remote sensing. However, thorough examinations of neural networks in the estimation of land surface temperature (LST) from satellite passive microwave (MW) observations are still lacking. Here, we examined the performances of the traditional neural network (NN), deep belief network (DBN), and convolutional neural network (CNN) in estimating LST from the AMSR-E and AMSR2 data over the Chinese landmass. The examinations were based on the same training set, validation set, and test set extracted from 2003, 2004, and 2009, respectively, for AMSR-E with a spatial resolution of 0.25°. For AMSR2, the three sets were extracted from 2013, 2014, and 2016 with a spatial resolution of 0.1°, respectively. MODIS LST played the role of “ground truth” in the training, validation, and testing. The examination results show that CNN is better than NN and DBN by 0.1–0.4 K. Different combinations of input parameters were examined to get the best combinations for the daytime and nighttime conditions. The best combinations are the brightness temperatures (BTs), NDVI, air temperature, and day of the year (DOY) for the daytime and BTs and air temperature for the nighttime. By adding three and one easily obtained parameters on the basis of BTs, the accuracies of LST estimates can be improved by 0.8 K and 0.3 K for the daytime and nighttime conditions, respectively. Compared with the MODIS LST, the CNN LST estimates yielded root-mean-square differences (RMSDs) of 2.19–3.58 K for the daytime and 1.43–2.14 K for the nighttime for diverse land cover types for AMSR-E. Validation against the in-situ LSTs showed that the CNN LSTs yielded root-mean-square errors of 2.10–4.72 K for forest and cropland sites. Further intercomparison indicated that ~50% of the CNN LSTs were closer to the MODIS LSTs than ESA’s GlobTemperature AMSR-E LSTs, and the average RMSDs of the CNN LSTs were less than 3 K over dense vegetation compared to NASA’s global land parameter data record air temperatures. This study helps better the understanding of the use of neural networks for estimating LST from satellite MW observations.

Thermal infrared (TIR) land surface temperature (LST) products derived from geostationary satellites have a high temporal resolution in a diurnal cycle, but they have many missing values under cloudy-sky conditions. Therefore, it is pressing to obtain all-weather LST (AW LST) with a high temporal resolution by filling the gap of TIR LST. In this study, a method integrating reanalysis data and TIR data from geostationary satellites (RTG) was proposed for reconstructing hourly AW LST. Then, taking the Tibetan Plateau, which is a focus of climate change as a case, RTG was applied to the Chinese Fengyun-4A (FY-4A) TIR LST and China Land Surface Data Assimilation System (CLDAS) data. Validation based on the in-situ LST shows that the accuracy of the AW LST is better than the FY-4A LST and CLDAS LST under clear-sky, cloudy-sky, and all-weather conditions. The mean RMSEs are 3.02 K for clear-sky conditions, 3.94 K for cloudy-sky conditions, and 3.57 K for all-weather conditions. Uncertainty and coarse resolution of the original FY-4A and CLDAS data affect the accuracy of the obtained AW LST. The results of the LST time series comparison also show that the reconstructed AW LST is consistent with in-situ LST. The reconstructed AW LST also has good image quality and provides reliable spatial patterns. RTG is practical in obtaining high temporal resolution AW LST from the Chinese FY-4A to satisfy related applications. It can also be extended to other geostationary satellites and reanalysis datasets.