Revisit of the Temperature and Emissivity Separation Algorithm (TES) towards Model Refinement
Authors
Zhang H., Hu T., Tang B.H., Olioso A., Didry Y., Mallick K., Hitzelberger P., Jiang Y., Cheng Y., Szantoi Z.
Reference
IEEE Transactions on Geoscience and Remote Sensing, 2025
Description
A constellation of high-resolution thermal infrared (TIR) missions is expected to be launched in the upcoming years. Land surface temperature (LST), as a key parameter retrieved from TIR observations, constrains the variations of energy and water exchanges in the surface-atmosphere continuum. The widely used temperature and emissivity separation (TES) algorithm stands as a promising candidate for LST retrieval from these future missions due to the availability of ≥3 TIR bands. To explore the possibilities of further refinements of TES, a revisit of the TES algorithm in terms of the error propagation from different sources is necessary. Until now, the respective uncertainties introduced by the three modules in TES (i.e., the normalized emissivity method (NEM), the ratio algorithm (RATIO), and the maximum minimum difference (MMD) module) remain unclear. In addition, the controversy over the performances of TES on gray and non-gray bodies is still unresolved. To address these research gaps, a comprehensive simulation analysis was conducted for the ECOSTRESS sensor to quantify the independent impact of each error source in TES on LST retrieval accuracy, including: (1) sensor measurement noise, (2) atmospheric correction errors, (3) the NEM and RATIO modules, and (4) the MMD module. The respective responses of gray and non-gray bodies to these factors were also compared based on the simulation dataset. Furthermore, the influence of the calibration scale of the minimum emissivity (ϵmin)-MMD relationship (i.e., cavity effect within vegetation canopies) was evaluated using the ECOSTRESS observations at 11 vegetated ground sites. The simulation analyses revealed that the error in atmospheric correction is the dominant impact factor significantly affecting the performances of all the other modules in TES, followed by the deviation from the regressed relationship in the MMD module. The measurement noise has minor impacts when it is well controlled (e.g., NEdT≤0.1 K), and uncertainties caused by the NEM and RATIO modules are negligible. The performance discrepancy of TES over gray and non-gray bodies is insignificant under low measurement noise, a condition anticipated to be met by the majority of current and future sensors. Regarding the calibration scale, the benefit of cavity effect correction is not evident according to the evaluation using the ground measurements. Based on the analyses, it is recommended that more efforts should be put into refining the atmospheric correction module and improving the fitting of emissivity samples to the ϵmin-MMD curve. In contrast, the expected benefits of refining the NEM and RATIO modules appear minimal.
