30-meter land surface temperature from Landsat via progressive self-training downscaling

Land surface temperature (LST) is a critical parameter for characterizing surface energy balance and hydrothermal processes. While Landsat has provided invaluable LST observations at medium spatial resolution for over 40 years, its native spatial resolution of thermal bands (e.g., 100 m) remains insufficient compared to its 30 m optical bands, failing to meet the demands of fine-scale studies. To address this issue, this study proposes a progressive self-training framework for downscaling Landsat LST to 30 m without relying on fine-scale ground truth, while requiring minimal external data dependence. The framework progressively optimizes a cross-modal fusion network to refine thermal details in a coarse-to-fine manner, characterized by one pre-training stage and two fine-tuning stages. This network leverages a cross-modal dictionary attention mechanism combined with a hybrid CNN-Transformer architecture to effectively extract complementary information between LST and auxiliary variables across multiple spatial resolutions. Quantitative ablation analyses demonstrate that the fusion design and auxiliary variable combination effectively reduce downscaling errors. In parallel, a frequency-domain guidance strategy, implemented at both the network structure and loss-function levels, is introduced to enhance structural fidelity and suppress artifact accumulation during progressive refinement. Ablation of the high-frequency structural loss further demonstrates its effectiveness in reconstructing crisp fine-scale thermal textures. Spatial validation against SDGSAT-1 30 m LST and temporal validation using in situ measurements confirm its reliability and accuracy, yielding a station-averaged RMSE of 3.465 K and an R 2 of 0.909, compared with 4.195 K and 0.861 for the official cubic interpolation product. Comparisons with existing methods further demonstrate that the proposed framework effectively reconstructs fine-scale thermal patterns while preserving spatial heterogeneity. Overall, the framework provides a stable pathway for enhancing the spatial resolution of Landsat LST, offering fine-resolution data support for fine-scale surface process studies and localized environmental monitoring.