Satellite Image Enhancement

Author: Sheng(Jerry) Wang

Subject: Project Summary

Overview

This project uses a deep learning method to enhance satellite resolution for land surface temperature, achieving up to 100× higher resolution with a correlation coefficient of R = 0.97 and an RMSE of 1.77 °C.

In the picture above, the 5×5 low-resolution data has been converted into 50×50 high-resolution data. But how accurate is it? Let’s check the picture below.

As you can see, it's hard to tell the difference—the changes are minimal, and the overall pattern remains the same. Roads and residential areas can still be clearly identified.

Details regarding data cleaning, processing procedures, and model parameters are not provided here, as the results are pending publication.