Following on from the article “How to increase the resolution of Sentinel 2 images from 10 to 1m,” let’s look at a concrete example of how to use it.
Work plan
Step 1: Prepare the environment
- Create a folder on your Google Drive:
MyDrive/Sentinel2_Coraux_S2DR3 - Open Google Colab
https://colab.research.google.com - Mount your Drive to store the results
Step 2: Install S2DR3 and define the study area
The script will contain:
- The installation of the s2dr3 package
- The coordinates of the area of interest
- The Sentinel-2 acquisition date (ideally cloud-free)
Step 3: Start processing
The script will automatically download the Sentinel-2 scene closest to the date, then apply super-resolution at 1 m/pixel.
The products will be saved to your Drive, in an output/ subfolder.
Step 4: View and compare
We will see:
- how to display a 10 m vs. 1 m RGB composition directly in Colab;
- how to open and compare GeoTIFFs in QGIS + SCP (for classification, NDVI, coral indices, etc.);
- which bands to use to observe corals (mainly B2-B3-B4, B8, B11, B12).
Step 5: Scientific validation
Finally, we will see:
- how to check whether the details generated are plausible (avoiding confusion between artifacts and real structures);
- how to quantitatively compare indices (e.g., NDWI, BSI) between standard Sentinel-2 and S2DR3.
The Blue Bay/Île aux Aigrettes area is perfect for observing the coral reefs of Mauritius, and the date of October 14, 2025 is recent, so it is likely to be covered by a Sentinel-2 acquisition without too many clouds.
Below is your complete and annotated Colab notebook, ready to copy and paste into https://colab.research.google.com.
It will allow you to generate Sentinel-2 images at 1 m resolution with S2DR3, save them to your Drive, then open and compare them in QGIS + SCP.
In Collab, link your notebook to your Google Drive.
Now, to ensure that processing does not take too long, you need to enable a GPU in the runtime environment. Look in the lower right corner of your notebook, you will see something like “Python 3” with a chip icon. Click on it and select Change Runtime Environment Type. A window will open where you will need to search for Hardware Accelerator and select GPU T4. Save the changes and the notebook will automatically restart with the new configuration.
Google Colab: S2DR3 – Sentinel-2 Super Resolution (Blue Bay / Île aux Aigrettes)
Copy this code into a new cell in your notebook (you can do this in one or more code cells) and run it:
At the end of processing, you will see:
# =========================================================
# S2DR3 - Super-résolution Sentinel-2 à 1 m
# Étude des coraux à Blue Bay / Île aux Aigrettes
# =========================================================
# --- 1. Monter Google Drive ---
from google.colab import drive
drive.mount('/content/drive')
# Crée un dossier pour les résultats
!mkdir -p /content/drive/MyDrive/Sentinel2_Coraux_S2DR3/output
# Lien symbolique pour un accès rapide
!ln -s /content/drive/MyDrive/Sentinel2_Coraux_S2DR3/output /content/output
# --- 2. Installer le paquet S2DR3 ---
!pip -q install https://storage.googleapis.com/0x7ff601307fa5/s2dr3-20250905.1-cp312-cp312-linux_x86_64.whl
# --- 3. Importer le module principal ---
import s2dr3.inferutils
# --- 4. Définir la zone d'intérêt (Blue Bay / Île aux Aigrettes) ---
# Coordonnées approximatives du centre de la zone (longitude, latitude)
lonlat = (57.73, -20.44)
# --- 5. Définir la date Sentinel-2 ---
date = '2025-10-14'
# --- 6. Lancer le traitement ---
# Cela télécharge la scène Sentinel-2 la plus proche de cette date
# et génère une version super-résolue à 1 m/pixel.
s2dr3.inferutils.test(lonlat, date)
# Les résultats sont enregistrés dans /content/output/
# (lié à ton Google Drive : MyDrive/Sentinel2_Coraux_S2DR3/output)
Results in Google Drive
Once processing is complete, you will find:
- 1 m GeoTIFF files (per band or composites),
- sometimes an RGB preview,
- and the original Sentinel-2 metadata.
You can then:
- Open these files in QGIS;
- Install or activate the Semi-Automatic Classification Plugin (SCP);
- Load the S2DR3 bands into SCP (as multispectral rasters) to calculate NDVI, NDWI, BSI, or coral indices (B3/B2, B11/B8, etc.);
- Compare with the original Sentinel-2 version (10 m) that you can download via Copernicus Hub or SCP Download.
Visual comparison
At the bottom of the code window, you will find an initial comparison between the 10 m and 1 m resolutions.
Click on the blue link to access a more detailed comparison.
- Use the RGB composition (B4 = red, B3 = green, B2 = blue).
- Overlay the 10 m version and the 1 m version: you will see more detail on the reefs, sandy bottoms, and grassy areas.
- For a rigorous comparison:
- calculate NDWI or NDVI on both images,
- then evaluate the differences in SCP (Band calc or Raster Calculator tool).
Best practices
- Check cloud cover: the S2DR3 model does not correct for clouds.
- Fine details are inferred by the model (not measured), so use them as a visual cue rather than an absolute measurement.
- For publication or scientific projects, always cite the use of S2DR3 (Sentinel-2 Deep Resolution 3.0) and mention that it is an AI reconstruction.
In the next article, we will see how to continue this work in QGis with the SCP (Semi-Automatic Classification Plugin).
