Answers are grounded in this paper and cite their sources.
Machine learning evaluation demonstrates accurate hourly land surface temperature reconstruction under cloudy skies, indicating improved all-sky thermal monitoring.
Geostationary satellite-based land surface temperature (LST) is vital for monitoring diurnal thermal dynamics owing to its extensive spatial coverage and high temporal resolution. However, frequent cloud cover results in substantial data gaps, limiting its practical utility for reliable thermal monitoring applications. The solar-cloud-satellite geometry (SCSG) effect has recently been introduced to reflect direct cloud radiative effects in LST reconstruction, but its application has been largely limited to polar-orbiting satellites. To address this limitation, we propose HELIOS-Net, a deep learning (DL) framework for hourly daytime LST reconstruction using the GEO-KOMPSAT-2A satellite across South Korea, central Japan, and southeastern Australia. The framework combines SCSG-based regime partitioning with a generative adversarial network (GAN) to reconstruct spatially continuous all-sky LST by leveraging satellite observations and various meteorological and topographic variables. Validation against in-situ observations demonstrated that HELIOS-Net achieved high performance across all regions, with coefficient of determination (R 2 ) values of 0.63–0.94 and root mean square error (RMSE) values of 2.45–7.65 °C under cloudy-sky conditions. Comparative evaluation further indicated an overall performance advantage over ERA5-Land and an existing all-sky LST product derived from satellite observations. Additional analyses confirmed that both the SCSG effect and the DL framework contribute to enhancing the fidelity of hourly daytime LST reconstruction. These results highlight the potential of the proposed framework to enable high-quality, spatially continuous all-sky LST monitoring from geostationary satellites, providing a promising approach mitigating cloud-induced data gaps under diverse viewing geometries and environmental conditions.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: