Generating a 30 m Hourly Land Surface Temperatures Based on Spatial Fusion Model and Machine Learning Algorithm.

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Climate Science

Background:

• Land surface temperature (LST) is crucial for climate change studies and hydrological balance.
• Current satellite LST products struggle to balance spatial and temporal resolution, limiting all-weather monitoring.
• High spatiotemporal resolution LST data is needed for detailed urban environmental analysis.

Purpose of the Study:

• To develop and validate a framework for retrieving hourly, high spatial resolution LST data.
• To overcome the limitations of existing satellite LST products.
• To provide valuable data for urban thermal environment monitoring.

Main Methods:

• A spatiotemporal fusion model combined with a machine learning algorithm was developed.
• The framework integrates data from Chinese geostationary weather satellites and polar-orbiting satellites.
• Hourly 30 m resolution LST data was retrieved and validated using in situ measurements and Sentinel-3 SLSTR data.

Main Results:

• The proposed framework achieved root mean squared errors (RMSE) of 0.89 °C to 1.23 °C against in situ LST measurements.
• The predicted LST showed high consistency with Sentinel-3 SLSTR data, with RMSEs ranging from 0.95 °C to 1.25 °C.
• Validation in Xi'an City confirmed the method's accuracy within approximately 1.33 °C.

Conclusions:

• The developed framework successfully generates hourly, high spatial resolution LST data.
• This method addresses the trade-offs in existing satellite LST products.
• The resulting 30 m hourly LST data is valuable for urban thermal environment monitoring and climate studies.