Artificial Intelligence-Based Joint Retrieval Algorithm for Land Surface Temperature, Emissivity, and Atmospheric Water Vapor

The thermal infrared remote sensing parameters exhibit interdependent and mutually constrained relationships, which conventional methods fail to fully exploit for improving the overall retrieval accuracy across different parameters. To address this challenge, this study proposes an artificial intelligence-based method for jointly retrieving land surface temperature (LST), emissivity (LSE), and atmospheric water vapor content (WVC) from thermal infrared remote sensing data, achieving an organic integration of physical mechanism-based and statistical approaches. The model is initially applied to retrieve LST and LSE, which are then utilized as prior knowledge for cross-iterative WVC retrieval. The simulation validation results demonstrate that the four-band combination scheme for LST retrieval achieved optimal theoretical accuracy, with a mean absolute error (MAE) of 0.51 K and root mean square error (RMSE) of 0.69 K. The retrieval RMSE values for both LSE31 and LSE32 remained below 0.01. Incorporating LST and LSE information with the four thermal infrared bands further enhanced WVC retrieval stability, yielding an MAE of 0.05 g/cm² and RMSE of 0.08 g/cm². Finally, cross-validation and ground-based verification using the optimal band combination confirmed the overall reliability of the retrieval results. The retrieval errors for all parameters were reduced during nighttime due to decreased solar irradiation interference. Overall, the proposed joint retrieval method improved both accuracy and stability for all parameters, overcoming the limitations of conventional techniques and enhancing overall retrieval performance.