Why 90 Days
Pre-positioning food aid, mobilising logistics and securing emergency funding through multilateral mechanisms takes 60 to 90 days. A forecast is only useful to that work if it arrives at least that early.
CERES produces falsifiable, probabilistic 90-day forecasts of acute food insecurity, expressed as P(IPC Phase 3+), the probability that a monitored region will reach crisis-level hunger within 90 days, with explicit 90% sensitivity intervals.
CERES predictions are research outputs intended to augment, not replace, field-based IPC assessments and expert humanitarian judgement. All forecasts carry explicit uncertainty quantification and should be interpreted alongside ground-truth verification.
The system ingests public data streams covering rainfall (NASA POWER), vegetation (MODIS NDVI), conflict (UCDP GED), acute food insecurity phase (IPC and Cadre Harmonisé, via HDX HAPI), and market prices (WFP, via HDX HAPI). FEWS NET is not a model input: it supplies the proxy used to grade the forecasts. These are synthesised by the Hypothesis Generation Engine (HGE) into ranked driver hypotheses, which feed a logistic model producing probabilistic risk scores at national level across 43 high-risk countries.
Six-Stage Processing Pipeline
Each pipeline run proceeds through six sequential stages. Runs are scheduled weekly. Each published run carries a run identifier, such as CERES-20260803-074444.
Data adapters ingest rainfall, vegetation, conflict, food security, and market price streams. Rainfall and vegetation are read at each country's centroid point, and each adapter aligns its source to an ISO-week temporal cadence. Raw data is cached with source provenance and retrieval timestamp.
Composite stress scores are computed per country as a weighted sum across six sub-scores: drought stress, vegetation anomaly, conflict intensity, food access, IPC phase, and market price deviation. Weights are set as expert-informed priors, specified with reference to 87 IPC transition records (2011-2023), and will be updated via MLE as the prospective verification archive accumulates.
Elevated signals are clustered into ranked driver hypotheses. Each hypothesis identifies a primary causal mechanism (e.g. conflict-driven market failure), supporting evidence records, and a confidence weight. Up to three hypotheses are generated per region per run.
A logistic regression model converts composite stress scores into P(IPC Phase 3+) at a 90-day horizon. Input-perturbation sensitivity intervals are generated by resampling input signals, producing 90% sensitivity intervals. Both the point estimate and the full interval are reported for every prediction.
Predictions are assigned to one of three alert tiers based on probability thresholds. Tier assignment triggers downstream alerting and determines the urgency framing in intelligence reports.
After T+90 days, each prediction is graded against the FEWS NET proxy described under Validation & Calibration. Brier scores, sensitivity-interval (SI) coverage, and precision/recall metrics are updated continuously. Calibration failures trigger model review.
The HGE: From Signals to Hypotheses
The Hypothesis Generation Engine (HGE) is the core intelligence layer that distinguishes CERES from threshold-based early warning systems. Rather than flagging when a single indicator crosses a threshold, HGE synthesises multi-source signal convergence into causal hypotheses: ranked, evidenced explanations of why risk is elevated.
Signal Convergence Detection
HGE monitors for simultaneous stress elevation across independent data streams. When two or more signals from different domains (e.g. NASA POWER rainfall deficit + UCDP GED conflict events + rising WFP market prices) converge on the same country in the same time window, this constitutes a convergence event: a materially stronger signal than any single indicator in isolation.
Hypothesis Taxonomy
Each convergence event is classified into one of four primary causal archetypes:
| Archetype | Primary Signals | Typical Regions |
|---|---|---|
| Conflict-driven | UCDP GED, IPC | Sudan, Somalia, Yemen, South Sudan |
| Climate-driven | NASA POWER, MODIS NDVI | Sahel, Horn (off-conflict seasons) |
| Economic/market | WFP market prices | Urban centres, import-dependent regions |
| Multi-causal | All core inputs | Active conflict zones with drought overlay |
Evidence Records
Every hypothesis is grounded in structured evidence records: individual signal observations that either support or contradict the hypothesis. Each record specifies source, variable name, observed value, baseline threshold, deviation direction, and a binary support/contradict verdict.
HGE never produces a prediction without an auditable evidence chain. Every probability estimate has a traceable hypothesis. Every hypothesis has traceable evidence records. This keeps every CERES prediction auditable.
Logistic Model & Sensitivity Intervals
CERES uses a logistic regression model to convert composite stress scores into IPC Phase 3+ exceedance probabilities at a 90-day horizon. The choice of logistic regression is deliberate: it is well-understood, natively probabilistic, and produces outputs that are straightforwardly interpretable by non-technical reviewers.
The model coefficients are expert-informed priors, specified at initialisation with reference to 87 IPC transition records across 31 countries (2011-2023), adjusted on 11 March 2026, and not yet fit to graded outcomes. They are deployed across 43 high-risk countries; forward performance on the full deployment set is established prospectively.
A CRITICAL convergence multiplies the logit by 1.18. No probability is published above 0.99.
where σ is the logistic function, CSS is the composite stress score,
and coefficients β are expert-informed priors. The live forecasts use these priors. Coefficient values are documented in Appendix C of the preprint.
Sensitivity Intervals
Point estimates alone are insufficient for humanitarian decision-making. CERES generates 90% sensitivity intervals via input-perturbation resampling: input signals are resampled within their stated uncertainty bounds across n=2,000 replications, and the resulting distribution of predictions forms the interval. This captures data variability and model sensitivity to input uncertainty.
n_replications = 2,000 · Empirical SI coverage target: ≥88% (populating as graded outcomes accumulate)
The coefficients from arXiv v1 were adjusted so that probabilities spread across the stress range: intercept −4.50, composite_stress β 3.20, convergence_score β 1.40 and n_independent β 0.20.
Alert Tier Definitions
Predictions are assigned to one of three alert tiers based on the point estimate of P(IPC Phase 3+). Thresholds were specified at initialisation and are reviewed as graded outcomes accumulate. In the ledger, a Tier I alert counts as confirmed when the observed outcome at T+90 is IPC Phase 3 or above.
IPC Phase 3+ (Crisis or above) probable within 90 days. Pre-positioning recommended.
IPC Phase 3+ (Crisis or above) possible within 90 days. Enhanced monitoring and contingency planning.
IPC Phase 3+ less likely than not within 90 days. Situational monitoring.
Tier I classification does not constitute a famine declaration. Only the IPC Global Platform, through its established cadre process and field verification, has the mandate to declare famine (IPC Phase 5). CERES Tier I indicates a probability of reaching Phase 3 or above.
Model Performance
CERES priors were specified with reference to 87 IPC transition records spanning 31 countries between 2011 and 2023. Forecasts are measured against the pre-registered targets on the Validation page. A prediction is graded after its T+90 horizon passes, once the data for the grading window is published. Grades currently use a FEWS NET proxy, which is the highest Current Situation phase FEWS NET reports for any area of the country within 30 days either side of the grading date. It is a national worst-area reading on the IPC phase scale, and it stands in for an IPC classification. The published rule gives an IPC or Cadre Harmonisé classification precedence where one is available to the grader. The figures below are read live from the verification ledger.
- TIER-I precision counts Phase 3+. Of 62 TIER-I alerts graded, 62 reached Phase 3+ on the FEWS NET proxy at T+90.
- TIER-I recall counts Phase 4+. Of 35 graded outcomes at Phase 4+ on the FEWS NET proxy, 32 had a TIER-I alert. Phase 3 outcomes are not counted. These outcomes cover 7 countries and 5 run dates.
- Brier decomposition. Reliability 0.065, resolution 0.061, uncertainty 0.185. The uncertainty term is fixed by how the outcomes fell.
- 143 of 1,376 predictions are graded, all against the FEWS NET proxy described in the Methodology. Its source data publishes on a lag. Metrics below their pre-registered minimum sample are marked not yet assessed.
Published ledger figures, refreshed hourly. Full verification ledger and pre-registered protocol
Known Limitations & Constraints
Data Latency
Several input streams (IPC and Cadre Harmonisé classifications) are updated bi-annually. Between updates, predictions rely on interpolated or lagged data, which may not capture rapidly deteriorating situations driven by sudden shocks (conflict escalation, flash flooding).
Spatial Resolution
Forecasts are issued at national level, with rainfall and vegetation read at each country's centroid point. Variation inside a country may be significant, and field assessments would capture it.
Model Transferability
The logistic model uses expert-informed priors and is deployed across 43 high-risk countries. Performance in geographically or structurally distinct contexts (South Asia, Central America) is established prospectively and should not be assumed.
Conflict Dynamics
Conflict data is sourced from UCDP GED (Uppsala Conflict Data Program, Uppsala University). UCDP GED captures georeferenced conflict events with approximately 3 days reporting lag for candidate data. In active conflict zones, the most acute areas may be the least reported. CERES may systematically under-estimate risk in media-dark conflict environments.
Reviewers, funders, and operational partners are invited to raise further questions with the Northflow research team at hello@northflow.no.
How to Cite CERES
If you reference CERES predictions or methodology in published work, please use the following citation format. The paper is published on arXiv and freely available under CC BY 4.0.