Reservoir storage Valle de Bravo · Cutzamala System
Sentinel-2 L2A · CONAGUA SINA Code on GitHub

Valle de Bravo storage

Pick a month to estimate how much water the lake held.

Estimated storage —  
Fill — of NAMO, hm³
CONAGUA reported —  
Clear passes — 95% of the lake visible or more

Storage by month

How full

0NAMO
Estimate
—
CONAGUA
—
Same month, a year before
—

Satellite passes

Date Lake clear Water area Estimate Fill CONAGUA Difference

Observations, January 2025 to November 2026

From the satellite estimates
  1. +100hm³

    The 2025 rainy season refilled the lake. It went from about 250 hm³ (63% of NAMO) in May 2025 to about 349 hm³ (88%) in November, after a low of 98 hm³ (25%) in April 2024.

  2. 2–6hm³

    While CONAGUA still reported (January to April 2025), clear passes from March on came within 2–6 hm³ of its figures, always slightly under. Some January passes were 20–47 hm³ under.

  3. 354 → 303hm³

    The 2026 dry season drew the lake down from January to June. The rains then brought it back up, to about 363 hm³ (92%) on the one pass so far in November 2026, the highest in this showcase.

  4. 3months

    June 2025, September 2025 and September 2026 have no estimate: every pass was over 20% cloud. August 2025 and October 2026 rest on a single pass, so they're less certain.

About this project

Read the full project overview Browse the code

Objective

The objective of this project was to learn some fundamental concepts of Geospatial Science and Earth Observation: Python libraries and satellite imagery, including bands, STAC and the Copernicus API. It also provides a way to estimate the reservoir storage, as the SINA API stopped providing data after April 23, 2025.

Due to time constraints I could not implement the rest of the Cutzamala reservoirs, but I implemented the estimation for the most important one: Valle de Bravo.

Conclusion

Imagery is more robust than I thought. It's amazing to me that using reflectance and other algorithms you can calculate and estimate sizes. No need for ML models for everything: in the end, you can use mathematical models to make the calculations you need.

This was extremely useful for me as an introduction to understand how a geospatial pipeline works, and the standards and systems behind things I did not know before.

Progress, step by step

Low-resolution preview of a whole Sentinel-2 tile with scattered clouds
1First Copernicus response: the quicklook preview of a whole tile, 31 Jan 2024.
True colour image of Valle de Bravo with a low lake and exposed shores
2First bands loaded from S3, read only inside the lake's bounding box. Same day, low water.
True colour image of Valle de Bravo, fuller than a year before
3Bands converted to reflectance, shown as true colour (B04, B03, B02). 1 Mar 2025.
False colour image: vegetation bright red, the lake dark navy
4False colour, NIR shown as red. Water absorbs NIR, so the lake goes dark.
NDWI map: the lake in blue, land in red
5NDWI from green and NIR: water blue, land red. Otsu picks the cutoff; the lake measures 16.31 km².
Hazy true colour image of the lake with clouds over the eastern shore
6Too cloudy: only 94.8% of the lake is clear, under the 95% gate, so 15 Apr 2025 is dropped.