EnvironmentExample project · Fictional nonprofitResearch prototype
How far ahead can you see a flood coming?
An environmental organization works with communities exposed to flooding. There's plenty of public information (river gauges, flood-zone maps, past events), but it's scattered and hard to read. The first goal is to explore a research prototype that brings that data together and visualizes it to support specialists' analysis.
- Nonprofit
- Riverbank conservation association
- Led by
- [TBD]
- Repository
- Opens with the project
With what was known on the 6th, what could have been said about the 9th?
Research prototype. It is not a warning or civil protection system, and it does not replace official alerts.
The problem.
The problem, in the nonprofit's words"When the river rises, the data exists but it's spread across four websites and nobody in town can read it. We want to understand it sooner, without anyone thinking it's an official warning."
2,691 m³/s
peak flow at Castejón, February 6, 2003
corrected official figure; the real-time reading that night was 3,320
Confederación Hidrográfica del Ebro · “Avenida del Ebro 2003”
2,237 m³/s
peak flow at Zaragoza, February 9, 2003
real-time reading: 2,988
Confederación Hidrográfica del Ebro · “Avenida del Ebro 2003”
February. After several days of rain, the upstream gauge rises fast. The association's specialist opens the gauge website, the flood-zone maps and her spreadsheet of past floods, and spends a morning preparing a summary for the towns. By the time she's done, the data has already changed.
Pick a task.
Three tasks, each with what you learn. Open one to see the steps and requirements.
- 01
DefineM
Review data sources
You learn: How to research with AI without believing everything: every source the agent suggests gets opened, checked and cited.
Know which public data exists for a river stretch and which is useful.
What you do
- 1.Find flow, flood-zone and past-event data.
- 2.Note license, format, coverage and gaps.
- 3.Recommend where to start.
Accepted when
- Each source with link, license and date.
- Each source's gaps are stated.
- The specialist confirms nothing is missing.
- You deliver
- A source table and a reasoned recommendation.
- What credit it gives
- Your reviewed contribution record. If it's your first on this project, its badge too.Technique you show: AI research verified source by source.
- 02
DesignL
Design an understandable visualization
You learn: How to communicate uncertainty: a view that doesn't promise more than the data knows.
See on one screen what's spread across four websites today.
What you do
- 1.Understand what the specialist asks when the river rises.
- 2.Design map, flows and past events together.
- 3.Decide how uncertainty shows.
Accepted when
- The specialist finds what she looks for in three websites today.
- Someone non-technical understands it's not an official warning.
- Data gaps are visible.
- You deliver
- A prototype with sample data and notes from two tests.
- What credit it gives
- Your reviewed contribution record. If it's your first on this project, its badge too.Technique you show: Honest visualization of uncertainty.
- 03
TestL
Evaluate against past events
You learn: How to evaluate without cheating with the future, with an agent checking that no later information leaks in.
Know what the prototype would have shown in past floods.
What you do
- 1.Pick two or three events with enough data.
- 2.Replay them without using later information.
- 3.Compare with what actually flooded.
Accepted when
- Runs again with a single command.
- It's shown that no future information is used.
- The report includes the errors.
- You deliver
- A report per event and the code that reproduces it.
- What credit it gives
- Your reviewed contribution record. If it's your first on this project, its badge too.Technique you show: Leak-free historical evaluation.
When it's done.
A specialist understands in minutes, not a morning, what's happening on her stretch: flows, official flood zones and past events in a single view. It would be evaluated against floods that already happened: what it would have shown and how far ahead.
- Review which public sources exist for the stretch, and their quality.
- Bring flows, flood zones and past events into a single view.
- Evaluate against past floods what the prototype would have shown, and how far ahead.
- Clearly document what the prototype can't tell you.
Time the technician needs to prepare a summary for the towns during a flood, before and at 90 days.
Risks and sensitive data
- That it reads as a warning system: it isn't, and the page says so.
- Claiming more certainty than there is: it's evaluated only with what was known at the time.
- Nothing that locates a specific home or person. No mapping finer than the official flood-zone cartography.
Data and license
Apache-2.0. Every source is public (SAIH, SNCZI, IGN, AEMET, Copernicus). The prototype documents what it cannot say.