Abstract: Significance　Land surface temperature (LST) serves as a critical physical parameter characterizing the surface energy balance of the earth, and is essential for understanding global change and guiding human activities. It directly drives thermal processes in the near-surface layer and atmospheric circulation, acting as a fundamental input variable for both global and regional climate modeling and prediction. In the fields of environment and ecology, LST is a fundamental indicator for monitoring urban heat island effects, assessing vegetation stress, and studying drought and hydrological cycles. In agricultural applications, it is closely related to crop yield estimation, frost warning, and irrigation management. In addtion, the changing trend of LST stands as one of the most direct evidence reflecting global warming, playing a vital role in the formulation and evaluation of energy conservation and emission reduction policies. Consequently, the precise monitoring and in-depth study of LST constitute a crucial scientific foundation for addressing climate change and achieving sustainable development. Progress　The precise estimation of LST utilizing satellite remote sensing technology has become a significant frontier in the field of global environmental and climate change research. Among the diverse methodological approaches, the retrieval technique based on the thermal infrared (TIR) band stands out as the most mature and widely adopted. This method operates by detecting the thermal infrared radiation naturally emitted from the earth's surface. To derive accurate LST measurements from this radiance, it is imperative to account for and remove the confounding effects of the atmosphere, which can only be achieved through sophisticated atmospheric correction models. Notably, algorithms such as split-window technique and single-channel algorithm have been effectively employed to mitigate the impacts of atmospheric absorption and emission. Therefore, this process yields LST products characterized by high spatial resolution. Under optimal clear-sky conditions, this methodology attains a high degree of accuracy, and its retrieval results typically demonstrate strong consistency and correlation with in-situ temperature data obtained from ground-based meteorological stations. Due to these advantages, TIRbased LST data has become an indispensable tool in numerous applications. It is extensively utilized for monitoring urban heat island effects, quantifying the surface energy balance for hydrological and climate models, and conducting detailed analysis of long-term climate change patterns. However, a fundamental and persistent limitation of TIR remote sensing is its pronounced sensitivity to cloud cover. Since clouds are largely opaque in the TIR spectra, they effectively prevent the surface-emitted radiation from reaching satellite sensors, resulting in data gaps. This inherent characteristic makes it extremely difficult to achieve continuous, all-weather, and diurnal LST monitoring using only TIR sensors. To overcome the inherent limitation of TIR remote sensing, passive microwave (PMW) remote sensing, employing microwave radiometers as sensors, provides a vital and complementary data source. Microwaves, for their longer wavelengths, possess a significant ability to penetrate non-precipitating clouds, atmospheric aerosols, and even light vegetation canopies. This unique property enables the acquisition of surface radiation data under virtually all weather conditions and throughout day and night. The primary trade-off, however, lies in spatial resolution. Due to diffraction limits associated with longer wavelengths, PMW radiometers typically offer much coarser spatial resolution, typically on the order of several kilometers to tens of kilometers, compared to the sub-kilometer resolution of TIR sensors. Despite this limitation, microwave radiometers deployed on polarorbiting satellites or geostationary platforms can generate large-scale, synoptic-scale LST products. These datasets are particularly valuable for applications in high-latitude regions frequently obscured by clouds, filling observational gaps in continuously cloudy areas, and constructing consistent, long-term climate data records essential for trend analysis. In order to overcome the inherent constraints of single-source data, whether the cloud-obstruction of TIR or the coarse resolution of PMW, the synergistic retrieval and fusion of LST from multi-source remote sensing data has emerged as a pivotal and dynamic research direction. This advanced paradigm is fundamentally based on the theories of spatiotemporal fusion and data assimilation. Its core objective is to intelligently integrate the complementary advantages of different datasets, such as the "high spatial resolution" advantage of TIR images and the "all-weather, temporally continuous" observation capability of PMW measurements. And achieving this integration involves leveraging a suite of advanced computational techniques. Machine learning approaches, including deep learning architectures and ensemble methods like random forest, are increasingly used to learn complex, non-linear relationships between different sensor data and target LST. Alternative or joint methods based on physical model coupling or sophisticated statistical interpolation and gap-filling algorithms are employed. The ultimate goal is to reconstruct seamless, spatially comprehensive, and temporally dense LST datasets with rich detail. A prominent example of this approach involves the synergistic fusion of data from the moderate resolution imaging spectroradiometer (MODIS) TIR sensor and PMW observations from AMSR-E. Such kind of fusion strategies can generate high-quality, gap-free LST products that are continuous in both space and time. The availability of these enhanced datasets significantly enhances their practical application potential, offering improved capabilities for modeling and monitoring hydrological cycle, detecting and assessing agricultural drought conditions with greater reliability, and conducting in-depth analyses of the frequency, intensity, and impacts of extreme heatwave events on ecosystems and human society. Conclusions and Prospects　LST estimation is crucial for disaster monitoring, agricultural assessment, and urban environmental studies. The current mainstream methods have their own advantages and disadvantages. The TIR method is mature and relatively accurate, but its retrieval precision is affected by clouds and surface heterogeneity in complex terrains. PMW sensing enables all-weather observation by penetrating clouds, yet its low spatial resolution limits fine-scale analysis. Although multi-source data fusion yields spatiotemporally continuous and detailed temperature information, its accuracy is constrained by land surface changes and mixed-pixel issues. Future research should focus on: (1) developing high-precision retrieval algorithms for complex terrains using topographic data; (2) advancing neural network-based spatiotemporal fusion techniques with optimized inputs and parameters; (3) integrating PMW and TIR data to achieve high spatiotemporal resolution. Additionally, developing algorithms suitable for China's Gaofen satellites and constructing long-term, highresolution regional or global LST products will provide reliable data support for agricultural management, urban planning, and climate change response in China.

Key words: estimation of LST, thermal infrared remote sensing, passive microwave remote sensing, multivariate data collaboration, spatio-temporal fusion