Daily Land Surface Temperature Reconstruction in Landsat Cross-Track Areas Using Deep Ensemble Learning With...

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Climate Science

Background:

• High spatiotemporal resolution land surface temperature (LST) is crucial for urban environmental studies.
• Landsat data offers high spatial resolution but suffers from infrequent revisits and cloud interference, limiting its utility in complex urban landscapes.

Purpose of the Study:

• To develop and validate DELAG, a deep ensemble learning method for reconstructing Landsat LST in complex urban areas.
• To enhance LST data availability using Landsat's cross-track capabilities and dual-satellite operation.
• To assess the reliability and uncertainty of reconstructed LST and its application in estimating near-surface air temperature.

Main Methods:

• Implemented DELAG, a deep ensemble learning approach integrating annual temperature cycles and Gaussian processes.
• Utilized Landsat data, enhanced by cross-track characteristics and dual-satellite operation, to increase data frequency to 4 scenes every 16 days.
• Validated DELAG in New York City, London, and Hong Kong under varying cloud cover conditions.

Main Results:

• DELAG successfully reconstructed Landsat LST in complex urban areas with low RMSE values (0.73-0.96 K clear-sky, 0.84-1.62 K cloudy).
• The method provides uncertainty quantification, enhancing the reliability of LST reconstruction.
• Reconstructed LST yielded accurate near-surface air temperature estimates (RMSE = 1.48-2.11 K), comparable to clear-sky LST results.

Conclusions:

• DELAG offers a novel and practical solution for high-resolution Landsat LST reconstruction in complex urban environments.
• The enhanced data availability and accuracy address limitations of traditional LST data acquisition.
• This method advances the potential for studying complex climate events and improving air temperature estimations at high spatiotemporal resolutions.