Land Surface Temperature End-to-End Retrieval From Fengyun-4B AGRI Thermal Infrared Remote Sensing Data Considering the Emissivity Angular Effect

Thermal infrared remote sensing (TIRS) technology can rapidly obtain long-time series and spatially continuous land surface temperature (LST), a key parameter that has received wide attention in agriculture and meteorology. Existing LST retrieval methods face limitations in addressing directional emissivity analysis for mixed pixels during simulation dataset construction. This study proposes a novel end-to-end LST retrieval algorithm using Fengyun-4B Advanced Geostationary Orbiting Radiation Imager (AGRI) TIR data. The algorithm replaces traditional material emissivity with mixed vegetation canopy emissivity derived from a modified FRA97 model during dataset simulation and trains multiple ensemble learning models based on varying leaf area indices and viewing zenith angles (VZAs). The corresponding optimal model is selected according to pixel-specific parameter ranges in practical applications. The validation results show that the maximum error reduction reaches 0.57 K and the average error reduction is more than 0.27K, which illustrates the effectiveness of the algorithm proposed in this paper. The high temporal resolution LST data can reveal diurnal temperature dynamics in cropland, which in turn can realize the real-time monitoring of farmland moisture, guide irrigation, and disaster warning, and optimize the model to improve yield management. Future work will extend the algorithm to diverse land cover types, analyze thermal band combinations, and integrate diurnal temperature cycle (DTC) models for filling the LST under the cloud to enhance the effectiveness of the application of TIR remote sensing technology in agriculture and other fields.