Two years into the Sudan war, how have violence and displacement co-evolved across the country and its neighbors?
# Sudan war displacement mapping
> Two years into the Sudan war, how have violence and displacement co-evolved across the country and its neighbors?
## The question
The war that broke out in Sudan in April 2023 has produced one of the largest displacement crises of the decade — millions internally displaced, and millions more crossing into Chad, South Sudan, Egypt, Ethiopia, and the Central African Republic. This project maps how *violence intensity* and *displacement flows* have co-evolved geographically and over time. We are **not** attempting to causally attribute displacement to specific events or actors — the data does not support that — and we are **not** producing a humanitarian needs assessment, which has its own established methodology. The deliverable is a geographic and temporal description of the conflict's footprint that could serve as input to either of those richer studies.
## Data
| Source | Granularity | Time coverage | Access |
|--------|-------------|---------------|--------|
| ACLED conflict events | Event-level (lat/lon, date, event type, actors) | April 2023 – May 2025 | Registered account; OAuth2 API is **tier-gated** — snapshot built from a manual Data Export Tool download |
| UNHCR Sudan situation portal | Country-of-asylum aggregates | April 2023 onward | Public (Refugee Statistics API + situation-portal snapshot) |
| IOM DTM Sudan | Admin-1 internal-displacement counts | August 2023 onward | Free API subscription key (DTM v3 API) |
| GADM 4.1 administrative boundaries | Admin-1 polygons for Sudan + 5 neighbors | Current | Public download |
**Access date: 2026-05-19.** ACLED revises historical records as new sourcing emerges, so the access date is pinned and a parquet snapshot (`data/processed/acled_snapshot_2026-05-19.parquet`) is the canonical analytic input — the notebook reads the snapshot, never the live API. The ACLED account tier withholds the most recent ~12 months of events, so the analysis window ends at **2025-05-19**; th …