top of page

Dark Floods

How accurately can flood history and terrain data pinpoint electrical substations at risk of shutting down?

Image Credit: Amanda Maciel Perobelli


When a flood reaches an electrical substation, everything it feeds loses power at once, including hospitals and water pumps. In May 2024 that happened across Porto Alegre, Brazil. We will overlay the city's mapped substations on the radar flood maps from that event and on ten years of recorded floods, train a model to predict which substations flood from their height above the river and how often the land around them has flooded before, and assign each hospital and pump to its nearest substation. The output is a map of which parts of the city lose power when which substation floods, with the substation list going to the utility.


Research questions

  1. Which of Porto Alegre's substations were inside the May 2024 flood, and which had been inside earlier recorded floods?

  2. Can a substation's height above the nearest stream, its distance to the river and the flood history of the land around it predict whether it flooded, well enough to apply to substations in a city with no recorded flood?


Datasets

  • Copernicus Global Flood Monitoring radar flood maps, 2015 to now, including May 2024

  • JRC Global Surface Water recurrence, for how often each spot has flooded since 1984

  • FABDEM elevation, for height above the nearest stream

  • OpenStreetMap substations (with voltage tags), power lines, hospitals and water pumping stations

  • WorldPop population; city and utility reports on which areas lost power in May 2024


Methodology in progress.


References

All AR²L labs are guided by our data governance policy to ensure ethical and responsible use of data. 

© 2025 By Applied Risk and Resilience Labs

bottom of page