A GeoAI Framework for High-Resolution Land Surface Temperature Estimation: Bridging Coarse Thermal Satellites and Sub-Decametric Optical Data—A tentative Application Using Sentinel-3 SLSTR

Land Surface Temperature (LST) is vital for environmental monitoring, yet a trade-off exists between the sparse revisit frequencies of high-resolution thermal sensors and daily, coarse-resolution (1 km) data from Sentinel-3 (SLSTR). This study introduces a multi-scale GeoAI framework to generate high-resolution LST maps by integrating Sentinel-3 data with optical imagery from Sentinel-2 (10 m) and Planet (3 m).

The Gamma Earth super-resolution algorithm is applied to Sentinel-2 imagery to reconstruct NDVI at a 1-meter resolution, while Planet imagery provides a native 3-meter NDVI baseline. Based on the inverse NDVI-LST correlation, statistical regression models translate coarse Sentinel-3 thermal pixels (S7–S9 bands, 3.7–12.0 μm) into high-resolution temperature estimates across three scales (10 m, 3 m, and 1 m). The workflow leverages open-access data, though Gamma Earth is free only for limited-scale areas. High-resolution data from the SatVu (HotSat-1) MWIR sensor (3.7–5.0 μm) are accessed free-of-charge through a European Space Agency research proposal. Due to inherent discrepancies in both thermal wavelengths and acquisition times between sensors, this process is framed as a cross-platform comparison rather than a strict validation, with analyses strictly restricted to proximal overpass time-frames to mitigate diurnal LST fluctuations.

The GeoAI-enhanced downscaling effectively captures fine-scale thermal heterogeneities. The 1-meter super-resolved products reveal localized heat signatures, including urban structures and vegetation stress, invisible in original Sentinel-3 data. Quantitative cross-comparison against HotSat-1 products demonstrates high consistency when overpass time-frames are closely matched.

This research provides a robust, largely open-access workflow for high-resolution thermal mapping. By combining public multi-spectral archives with scalable super-resolution and cross-comparison against high-performance platforms, the proposed framework offers a viable solution for operational thermal monitoring.