Verdant

Cyanobacteria cells under the microscope: the harmful algal blooms Verdant helps rangers find, built for the OneAquaHealth hackathon.
Where to sample first at Lake Mead: the CDC count, a ranked sampling list and shore reports. Prototype demonstration data.
The Verdant loop: report, review, prioritize sampling, share the outcome. Lake Mead pilot, Track 6 Resilience Informatics.
Every score explains itself. Callville Bay is 79, Very High, with factor bars. It shows where to look, not an official advisory.
Reporting takes three steps on a phone: pick the site, compare with the "Is it a bloom?" guide, then check and send.
Rangers triage each report, then the reporter sees the outcome: "You got it right." The ranger review shown is simulated.
Seven languages from one header switch: English, Portuguese, Spanish, French, Italian, Dutch and Norwegian.

In three years, 18 US states alone reported 389 human and 413 animal illnesses from harmful algal blooms. Verdant shows water managers where to sample first, so a sighting at the shore can become an earlier warning for people and pets. A guided citizen report goes to ranger review, and the reporter sees the outcome.

Verdant dashboard: risk map, sampling priority list and score breakdown

Inspiration

Water ecosystem health connects people, animals, and the environment. Across eighteen U.S. states, 421 harmful algal bloom events were reported during 2016–2018, alongside 389 human illnesses and at least 413 animal illnesses. These are national figures, not Lake Mead counts (CDC, MMWR 69(50)). Agencies everywhere face the same gap. They cannot watch every shoreline, and the people near the water rarely hear whether their report helped.

We built Verdant to connect those steps: people report, rangers prioritize follow-up, and reporters see the outcome.

What it does

Verdant is a working demonstration of a report-to-outcome workflow, with six pages and an interface in seven languages (English, Portuguese, Spanish, French, Italian, Dutch, Norwegian).

Step	What happens
1. Find a site	A risk map (Satellite or Map basemap) and a sampling priority list show where follow-up is recommended.
2. Understand the score	Each site breaks its score into factor contributions and names its hazard pathway.
3. Report	A three-step form with a bloom and look-alike guide. Every question and the photo are optional, and the photo can come from the camera or the phone library.
4. Review	A demonstration ranger queue lets users request sampling and record a simulated outcome.
5. See what happened	My reports shows the same report with its status history and outcome.

Review changes how much a report counts. A new report adds 10 citizen-evidence points, a confirmed one adds 20, and one ruled not a bloom adds 0. Here is Callville Bay:

Event	Overall score
Starting score	79
Report submitted	81
Simulated confirmation	82
Ruled not a bloom instead	back to 79

These weights are demonstration rules, not validated. A citizen report never confirms a bloom, and every ranger outcome here is simulated. Verdant flags conditions associated with higher bloom risk. It does not confirm toxins or replace field sampling, and its scores are not an official advisory.

Who it's for

Reporters: visitors, residents and boaters at a lake or stream, who get a guide for what they see and a status for what they send.
Rangers and water managers: the staff who decide where to sample first and need reports they can act on.
Researchers and public-health partners: OneAquaHealth teams who could use consistent, reviewed observations alongside monitoring data.

What is real and what is demonstration data

Item	Status
Lake Mead scores and reports	Demonstration data
Lake Mead elevation (1,037.93 ft on 2026-10-03)	Official U.S. Bureau of Reclamation snapshot, labelled separately
Coimbra satellite chart	Real Resilience Map export (monthly NDVI and NDWI area means)
Coimbra site scores	Synthetic; the chart does not drive them
Review outcomes	Simulated
Translations (pt, es, fr, it, nl, nb)	Machine-drafted, pending native-speaker review
FHIR-shaped JSON export	Demonstration format, not validated against the OneAquaHealth guide

Callville Bay with the ranger queue showing how a review changes the score

How we built it

React, TypeScript and Vite, with Leaflet maps, Recharts charts, PapaParse for CSV parsing and Tailwind with shadcn components. The scoring engine is tested TypeScript with fixed, published weights, and a separate tested module applies the review weights. A tested parser reads Resilience Map CSV exports. Each region is one config object, so the same app runs Lake Mead and Coimbra. The interface strings live in one typed dictionary per language. The project has 151 automated tests in 20 files. There is no backend: reports stay in the browser's local storage.

Challenges

Area-level data. The Coimbra export holds monthly area means, not the site-level measurements scoring needs. We kept the real chart and gave the 20 sites real Resilience Map names with synthetic scores, labelled as such.
Recommendation versus advisory. A sampling priority is not a confirmed hazard or an official advisory. We worded every score as "where to look first" and put the disclaimer next to it.
Seven languages. Each language is a typed copy of the English dictionary, so a missing key fails the type check. The six translations are still machine drafts, and some phrasing needs native review.
Open verification. We could not retrieve the official OneAquaHealth survey fields, so some report questions are placeholders, and the FHIR guide's profile ids are unverified. Duplicate and malicious reports are unsolved.

Accomplishments

A full loop from guided report to simulated ranger review to the reporter's status history.
Review-weighted evidence with a tested example: 79, 81, 82, and back to 79.
Scores that explain themselves, factor by factor.
One engine for Lake Mead and Coimbra, switched by config.
A tested seven-language interface, a Satellite or Map basemap that remembers its setting, and a layout that works on a phone.

What we learned

A score is more useful when people can see why it is high. A report is worth sending when the reporter sees what it led to. Weighting evidence by review status also made us decide what a citizen report is worth, and that choice needs field data before anyone should rely on it. A working demonstration is the first step, and testing with citizens and rangers comes before operational use.

What's next

Test with citizens and rangers: reporting completion, clarity, and usefulness for review.
Validate scores against field measurements and revise the weights.
Add shared storage, ranger accounts, and privacy controls.
Handle duplicates, location uncertainty, and malicious reports.
Verify OneAquaHealth fields, add live satellite and weather feeds, and validate exports with partners. None of these exist today.
Have native speakers review the translations.

Impact & Alignment with the OneAquaHealth mission

Verdant supports OneAquaHealth and Track 6: Resilience Informatics. It turns scattered citizen and environmental signals into explainable monitoring priorities for freshwater sites, and it supports monitoring, protection and awareness by keeping water, human, animal and environmental health on one screen.

Expected impact on ecosystems and human health. Blooms threaten the water that people, pets and wildlife depend on. Verdant aims to shorten the time between someone seeing a bloom and someone qualified deciding whether to sample: clearer reports, a ranked list of sites, and visible follow-up. Earlier sampling can mean earlier warnings for people and animals, and a record that links water conditions to human and animal health, the One Health idea behind the program. We have not measured any of this, and field validation comes first.

Lake Mead is the US pilot and Coimbra, Portugal, is a OneAquaHealth case-study city. Benevento, Ghent, Oslo, and Toulouse are configured and need site data. The header already shows "More cities soon".

Innovation & Creativity

Citizen science often ends when the report is sent. In Verdant the report keeps going. An observation enters review, its review status changes its weight in the score, and the reporter sees the outcome. The bloom guide, score explanations, and status timeline make that loop readable to both reporters and rangers.

The same app in three languages

Architecture

The engine combines pathway factors with fixed weights, handles partial data, and explains each contribution. Separate modules cover the report lifecycle, the CSV parser, the basemap choice, and a FHIR-shaped JSON export. Everything is typed, and the logic is covered by the 151 tests. It draws on CDC counts (cited), a U.S. Bureau of Reclamation elevation snapshot, a Resilience Map CSV export for Coimbra, and Esri map tiles. The next milestone is a shared service for citizens and rangers in place of one browser.

UX

The path is map, explanation, report, outcome. The interface offers a header language switch with flags and a Satellite or Map toggle. The report form has three steps and optional answers. Controls carry screen-reader labels, the switches work from the keyboard, and layouts adapt to phone screens. We have not tested accessibility or usability with intended users yet.

Scale

The region config is the unit of reuse. A new city needs site coordinates, suitable data (such as a Resilience Map CSV and a column map), a validated scoring model, and a local review process. Existing systems connect through the CSV import and the FHIR-shaped export, and both need validation with partners. There is no server to run today. We would start with a small agency-supported pilot, measure report quality and ranger workload, and expand from there.

With appreciation to the OneAquaHealth team and the hackathon organizers, whose challenge shaped this project, and to the CDC, the U.S. Bureau of Reclamation, Esri and the Resilience Map, whose data it runs on:

Thank you!

Made by Albert Dang, Michu Zaki, & Nathan Diep.

Live prototype · Methodology and sources · GitHub repository

Built With

cloudflare-workers
fhir
leaflet.js
papaparse
react
react-router
recharts
shadcn-ui
tailwindcss
typescript
vite
vitest

Submitted to

OneAquaHealth IEEE Global Hackathon

Created by

I wrote code for the scoring process, designed the thumbnails and the edited the report. I also recorded and edited the video with a teammate. We covered the dashboard, the report form, the ranger queue and the reporter's outcome page.

Albert Dang
heliaval
I worked on the scoring, the map and the report form. I also did the demos in the video and narrated them. In them I submitted a report at Callville Bay, showed the ranger review, and showed the reporter's outcome page.

Michu Zaki
I wrote code across the app, including the scoring and the report form. With a teammate, I recorded and edited the video. We switched between the slides and the live site as the video moved from the problem to the demo and the next steps.

Nathan Diep
how u been