Thermal sharpening of Sentinel-3 images: From 1 Km to 10m using Python in Google Colab
- 🌅 Introduction
- 💾 Downloading Sentinel-3 (1000 m) and Sentinel-2 images (10 m)
- ⚙️ Sentinel-3 Image Processing
- 🌡️ Temperature-NDVI Space
- 📐 Sharpening the Thermal Image (1000 m to 10 m)
- 🗺️ Visualization of the Sharpened Thermal Image
- 📄 Conclusion
- 📚 References
🌅 Introduction
Downscaling the thermal imagery captured by satellites has been extensively studied as a consequence of the tradeoff between the spatial and temporal resolution of satellites that provide thermal images. For instance, the revisit cycle of Landsat-8 is 16 days, with an original thermal resolution of 100 meters. In contrast, Sentinel-3 can provide every day thermal images, but at a spatial resolution of 1000 meters.
One approach to deal with the coarse resolution of thermal images might be launching more satellites equipped with thermal sensors, similar to NASA’s Landsat-9, launched in September 2021. On this case, the temporal resolution for each Landsat-8 and Landsat-9 is 8 days (as a substitute of 16 days with one satellite), assuming clear skies.
Nevertheless, as you may guess, this approach requires a multimillion-dollar investment and several other years of effort. As a substitute, researchers have focused on statistical methods, correlating the visible/near-infrared (VNIR) bands from satellites with higher spatial resolution (but lower temporal resolution) to thermal images from satellites with lower spatial resolution (but higher temporal resolution). For instance, studies have shown that the Normalized Difference Vegetation Index (NDVI) calculated from VNIR bands of Sentinel-2 (10m, every 5 days) might be inversely correlated with thermal images from Sentinel-3 (1000m, every day).
But how can we use this…