CERES ingests six formal model inputs (CHIRPS, MODIS NDVI, UCDP GED, IPC, WFP VAM, FAO GIEWS). FEWS NET projections are used as a supplementary cross-check. UNHCR displacement data is not currently an active model input. All sources are publicly available. No proprietary data is used.
Each data stream feeds into a specific stage of the CERES pipeline. Sub-scores are combined into a composite stress score which drives the probabilistic forecast. See the Methodology page for full mathematical specification.
| Signal | Role in Pipeline | Stage |
|---|---|---|
| CHIRPS SPI-3 | Drought stress sub-score | Stress scoring |
| MODIS NDVI anomaly | Vegetation stress sub-score | Stress scoring |
| UCDP GED conflict events | Conflict intensity sub-score | Stress scoring |
| IPC phase estimate | Food security phase sub-score | Stress scoring |
| WFP VAM food access | Food access sub-score | Stress scoring |
| FAO GIEWS price index | Market deviation sub-score | Stress scoring |
| IPC cadre outcome | Ground-truth for grading | Calibration (T+90d) |
| FEWS NET outlook | Signal corroboration (supplementary) | Hypothesis generation |
| UNHCR displacement | Signal corroboration (supplementary) | Hypothesis generation |
All CERES predictions are freely available as HXL-tagged CSV — no authentication required. Compatible with HDX pipelines, Excel, R, Python, and Tableau. Licensed under CC BY 4.0.
Best for dashboards, situational awareness, and map data integrations.
Download ↓Best for time-series analysis, trend research, and calibration studies.
Download ↓All predictions that have reached T+90, with actual IPC phase and Brier scores.
Download ↓HXL format: Row 1 = column headers · Row 2 = HXL hashtags · Row 3+ = data. Full API documentation →
Every source used by CERES is publicly available and free to access. No proprietary or licensed data is used in the pipeline.
Every prediction stores the data retrieval timestamps for all ingested sources. Full audit trail from raw signal to probability estimate.
Data latency for each source is published here. Known latency constraints are explicitly modelled as limitations in forecast uncertainty.