Land surface temperature (LST) is one of the key parameters in the process of energy exchange between the land surface and atmosphere, and thermal infrared (TIR) remote sensing is an important approach to efficiently obtain LST over a large area. Algorithms for retrieval of LST from TIR remote sensing data have been studied for decades, and the split-window (SW) algorithm can directly eliminate atmospheric effects by using the brightness temperature (BT) at the top of the atmosphere in two adjacent TIR channels and, thus, is widely applied. Landsat-9, the latest launch in the Landsat series of satellites, provides two-channel TIR images with the same 100-m spatial resolution as Landsat-8, and it is meaningful to develop the SW algorithm for LST retrieval using Landsat-9 data. In this letter, four SW algorithms were developed, and the accuracy and noise sensitivity of the results under different observation conditions were compared based on the simulation dataset to select the algorithm with the best performance. The ground measurement data under different land cover types and the global Landsat-9 LST products, produced by the single-channel algorithm, were selected to verify the accuracy of the proposed algorithm. The results show that the ground validation accuracy is about 1.574 K, better than the Landsat-9 existing LST product. Moreover, the retrieved LST images have similar spatial distribution to the Landsat-9 LST products, with RMSEs from 0.31 to 2.87 K in various regions.

Thermal infrared remote sensing can acquire large-scale land surface thermal radiance effectively. However, the observed data are affected by surface and atmospheric conditions. Traditional methods require some prior knowledge, such as emissivity in split-window algorithm and atmospheric correction in temperature–emissivity separation algorithm. This information is difficult to obtain directly and accurately. Hyperspectral thermal infrared data provide the possibility for simultaneous retrieval of atmospheric parameters, land surface temperature (LST), and emissivity because of their abundant band information. This study proposed a feature-band linear-format hybrid (FebLihy) algorithm by combining a deep neural network (DNN) model and a physical model with thermal airborne hyperspectral imager (TASI) data. The proposed algorithm was divided into three steps. First, the radiative transfer equation was converted into a linear form, and seven feature bands were chosen to reduce the unknowns. Second, the initial values of atmospheric and land surface parameters were estimated with the DNN model. Finally, least-squares optimization was used in the physical model to retrieve the final results. Results of the simulation data showed that the root-mean-square error (RMSE) of LST was 0.86 K, the RMSE of emissivity was less than 0.015, and the accuracy of atmospheric parameters was improved effectively by the physical model. The FebLihy algorithm was applied in a real TASI image in Fuyun County and verified with CE312 ground measurement data. Accurate results were achieved. The FebLihy algorithm will be optimized in terms of model and data in the future study.

The second-generation Chinese polar-orbit meteorological satellite Fengyun-3D carries the MEdium-Resolution Spectral Imager (MERSI-II). MERSI-II has 19 solar channels and 6 middle and thermal infrared channels and can provide daily global coverage observation at daytime and nighttime. Land surface temperature (LST) products have a wide range of application requirements. Unfortunately, there is no official LST product of MERSI-II yet. This article proposed a temperature-emissivity separation algorithm to estimate LST and emissivity from two mid-infrared (MIR) and two thermal infrared (TIR) nighttime MERSI-II images. The sensitive analysis and ground validation show the following. First, the algorithm can theoretically retrieve LST with an error of less than 0.7 K, emissivity with errors of about 0.025 for MIR channel and 0.01 for TIR channel, respectively. Second, the nighttime LST retrieval error was approximately 2.19 K, and the bias was approximately 0.60 K among 6 SURFRAD sites and 2 PKULSTNet sites. The retrieved emissivity was cross-validated using MODIS emissivity product over a desert area and its difference was obtained as 0.01 and 0.016 for two TIR channels. Finally, the TES algorithm was applied to obtain LST and emissivity images over North China Plain as an example, and the nighttime LST and land surface emissivity (LSE) were obtained well. As a result, this study shows that the proposed TES algorithm provides an effective way to get LST and LSE image from FY-3D/ MERSI-II nighttime observation, which will improve their applications in different fields.

Accurately estimating land surface temperature (LST) is a critical concern in thermal infrared (TIR) remote sensing. According to the thermal radiance transfer equation, the observed data in each channel are coupled with both emissivity and atmospheric parameters in addition to the LST. To solve this ill-posed problem, classical algorithms often require the input of external parameters such as land surface emissivity and atmospheric profiles, which are often difficult to obtain accurately and timely, and this may introduce additional errors and limit the applicability of the LST retrieval algorithms. To reduce the dependence on external parameters, this paper proposes a new algorithm to directly estimate the LST from the top-of-atmosphere brightness temperature in Landsat-9 two-channel TIR data (channels 10 and 11) without external parameters. The proposed algorithm takes full advantage of the adeptness of the ensemble learning method to solve nonlinear problems. It considers the physical radiance transfer process and adds the leaving-ground bright temperature and atmospheric water vapor index to the input feature set. The experimental results show that the new algorithm achieves accurate LST estimation results compared with the ground-measured LST and is consistent with the Landsat-9 LST product. In subsequent work, further studies will be undertaken on developing end-to-end deep learning models, mining more in-depth features between TIR channels, and reducing the effect of spatial heterogeneity on accuracy validation.