Researchers have developed A$^2$E$^2$, a novel end-to-end U-Net model designed to reconstruct land surface temperature (LST) data from Landsat imagery. This model integrates an annual temperature cycle, coarse LST from Earth system modeling, and AlphaEarth embeddings to overcome cloud contamination issues that limit continuous observation. Experiments over New York City demonstrated A$^2$E$^2$'s effectiveness, achieving a mean absolute error of 0.78 K and reducing error by over 60% compared to baseline methods. The study also found that Earth foundation model embeddings performed better for LST reconstruction than traditional spectral bands. AI
IMPACT This research could improve the accuracy and availability of land surface temperature data for environmental monitoring and climate studies.
RANK_REASON The cluster contains an academic paper detailing a new AI model and its performance on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →