Methodology
How we collect, process, and publish global crisis events.
Data feeds (current status)
- USGS — Earthquakes; working. Auto-published, high confidence.
- GDACS — Natural disasters (cyclones, floods, etc.); working. Auto-published, high confidence.
- ReliefWeb — Humanitarian disasters; migrating to v2 API.
- GDELT — Conflict data (Ukraine, Russia, Iran, Israel, Gaza) from daily export; filtered for EventRootCode 14–20 or GoldsteinScale ≤ -5 or actor keywords. GDELT conflicts now auto-published (beta). Category: 14–15 Political Tension, 16–20 Armed Conflict; medium confidence.
- ACLED — Armed conflict (beta). Ukraine, Israel, Iran; last 7 days. Auto-published with category Armed Conflict. Requires ACLED API token (myACLED).
Conflicts are now live (GDELT + ACLED beta). GDELT conflicts now auto-published (beta).
Pipeline
Ingest → Dedupe → Under Review → Manual or auto publish.
- Ingest — Scripts and cron jobs fetch from each feed (USGS, GDACS, ReliefWeb, GDELT, ACLED).
- Dedupe — Events are matched by source URL and similarity to avoid duplicates.
- Under Review — New events enter as drafts with status Under Review (except trusted feeds that auto-publish).
- Publish — Reviewers approve or reject; USGS, GDACS, GDELT conflict (top-impact), and ACLED auto-publish. Published events appear on the map and public list.
Limitations
- Disasters (earthquakes, cyclones, etc.) are prioritized and generally higher quality.
- Conflict and protest data (e.g. GDELT) are noisier; filtering and tuning are ongoing.
- Confidence is based on source reliability and corroboration count; see footer note on the home and map pages.
Plans
- ACLED integration (beta) is live; expand countries or date range as needed.
- Improve clustering so related events are grouped into incidents more reliably.
- Refine GDELT filters and confidence rules.