Start here · 11:25–11:35
The lake and the city
Bellandur is Bengaluru's largest lake. In the last ten years it has foamed over
onto the roads, caught fire, been drained and dug out, and watched the fields around it turn into
offices and apartments. Today we follow it through open data: first by finding and reading that
data the cloud-native way, then by letting a foundation model read the pixels for us.
Flip between the two pictures. Both were taken by the Sentinel-2 satellite in February, eight
years apart. The green mat on the left is the lake in 2017, covered in weeds.Your own catalog
As you go, you save what you find into your own folder,my_catalog/, in the same format the
big archives use. By the end of the day it holds a satellite picture, the rain of
the flood night, the watersheds and buildings around the lake, and a map of what changed: a small
catalog of the lake that you could publish.
How these notebooks work
A marimo notebook reruns itself: change a number in a cell and every cell that depends on it updates. Read the code cells that are shown; the plotting code lives incngeo/charts.py.
The charts are meant to be played with. Hover them, click them, drag the sliders. Nothing you do
there can break anything.
Check your setup
This checks the Python packages and the internet services we use. Everything should sayTrue.from cngeo.setup_check import checks
checks()
| check | ok | detail |
|---|---|---|
| marimo | True | 0.25.1 |
| numpy | True | 2.5.3 |
| pandas | True | 3.0.6 |
| geopandas | True | 1.2.0 |
| rasterio | True | 1.5.2 |
| xarray | True | 2026.9.0 |
| zarr | True | 3.4.1 |
| pystac | True | 1.15.2 |
| pystac_client | True | 0.9.0 |
| odc.stac | True | 0.5.3 |
| duckdb | True | 1.5.6 |
| sklearn | True | 1.9.1 |
| altair | True | 6.3.0 |
| Earth Search (Sentinel-2) | True | HTTP 200 in 0.8s |
| dynamical.org (rain) | True | HTTP 200 in 0.4s |
| CoRE Stack | True | HTTP 200 in 0.2s |
| Overture Maps | True | HTTP 200 in 0.1s |
| Source Cooperative (AlphaEarth) | True | HTTP 200 in 1.1s |
| Hugging Face (Major TOM) | True | HTTP 200 in 0.3s |
| Workshop files (GitHub) | True | HTTP 200 in 0.3s |
The data
Every dataset is open, and each lesson reads it live from the cloud. A few charts use small files we prepared in advance (27 years of rain at the lake, nine years of embeddings); each one downloads the first time a notebook needs it.| What | Source | Stored as |
|---|---|---|
| Satellite pictures | Sentinel-2, via Element 84 Earth Search | COG + STAC |
| Rain | NASA IMERG via dynamical.org | Zarr (Icechunk) |
| Temperature | NASA POWER analysis-ready store | Zarr |
| Watersheds, water bodies | CoRE Stack | STAC + GeoJSON |
| Buildings | Overture Maps | GeoParquet + STAC |
| Embeddings | AlphaEarth Foundations (Google DeepMind) and TESSERA (Cambridge) | COG, Zarr, GeoParquet |
| Land cover labels | ESA WorldCover 2021 | COG |