From Track to Toll

Does stronger wind really explain who suffers most? An interactive D3 data story about tropical cyclones and reported human impact across Pacific island states, 2005 to 2023.

Open the live visualisation · also on Hugging Face Spaces · dataset and analysis notebook on Kaggle

What it shows

The story links physical storm tracks to country-level impact records. It tests how far the wind a cyclone actually brought to a country lines up with the share of people reported affected. The finding: wind explains little. Country size explains several times more. Of the records from countries under 60,000 people, 12 of 14 sit above the overall median share. Population size explains 13 % of the differences (p = 0.003).

The page has three parts:

• Guided story: a scrollytelling narrative from storm tracks to impact records.
• Evidence Lab: freely filterable views of the same data.
• Data & methods: a public provenance section with every source, transformation and limitation.

Storm trends 2001 to 2025	Opening

Highlights

• Reproducible pipeline: a Python pipeline turns raw source data into JSON artefacts. meta.json records the source catalogue, transformations, story evidence, Git state and SHA-256 checksums.
• Publication gate: npm run build:public only builds when every source is open or permissioned and fully verified. A leak guard blocks restricted fields permanently.
• Tested end to end: unit tests, pipeline tests and a Playwright browser audit.
• Honest limits: the page states what the data cannot prove (see below).

Tech stack

D3 v7 · Vanilla JS · Vite · Python (pandas, NumPy, SciPy) · Playwright · GitHub Actions and GitHub Pages

Data sources

Source	Used for	License
SPC / Pacific Data Hub, SDG 11.5.1 (VC_DSR_AFFCT)	Annual people affected per country	Open (PDH)
SPC / Pacific Data Hub, Climate Change Indicators (SST_ANOM)	Sea surface temperature anomalies	Open (PDH)
IBTrACS v04r01, NOAA/NCEI	Tracks, intensity, categories, R34 radii, seasonal trends	US government work
UN World Population Prospects 2024	Population normalisation	CC BY 3.0 IGO
Natural Earth via world-atlas	110m base map	Public domain

Storm linkage: for each country and year, the strongest IBTrACS storm whose track passed within 500 km of the country centroid. Details per file: data/SOURCES.md.

Run locally

Requirements: Node.js 20+, Python 3 with pandas, NumPy and SciPy. The generated artefacts are checked in, so the frontend runs without a pipeline run.

cd project/app
npm ci
npm run dev

Reproduce and test:

cd project
python3 scripts/build_track_to_toll.py   # reads ../data, writes app/public/data
cd app
npm run check                            # unit + pipeline tests + build
npm run test:browser                     # Playwright audit
npm run build:public                     # gated public build

Project structure

.
├── project/
│   ├── app/       # Vanilla JS / Vite / D3 frontend (deployed to GitHub Pages)
│   ├── scripts/   # Python data pipeline
│   └── tests/     # unit, pipeline and Playwright checks
├── data/          # raw source data + SOURCES.md (provenance and licenses)
└── short paper/   # short paper, screen recording, problem statement

What the visualisation does not claim

It does not prove causal vulnerability. The affected counts are annual totals across all disasters, not storm-specific. Reported impacts are incomplete. Peak wind differs from local wind at landfall, and a missing value does not mean zero people affected.

Context

Built for the Pacific Dataviz Challenge 2026. The submission meets the challenge rules: official PDH datasets, only open supplementary data, every dataset cited, English interface, and AI tools used in a supporting role only (disclosed in the short paper). Problem statement: short paper/problem-statement-EN.md.

Author

Yannik Huber · Website · LinkedIn

License

Code: MIT. The data in data/ keeps its original licenses, listed above and in data/SOURCES.md.