Does ENSO information improve subnational forecasts of acute food insecurity across Africa's East–Southern dipole? A preregistered Bayesian study with a prospective test on the 2026–27 El Niño (Protocol v5 - round-clustered check of the ENSO coeffici
VERSION 5Version 5 (28 September 2026) adds a round-clustered classical check of the ENSO coefficients (maximum-likelihood refit with cluster-robust standard errors by target round), with its code and output. Because ENSO varies only between the 47 assessment rounds, the check shows that the ENSO effect is robust in pastoral areas and for El Niño in Southern Africa, but not for cropping areas (H2) or for the La Niña–El Niño asymmetry in the Horn (H4). It also corrects the H4 statement in the version 4 changelog. The two committed 2026–27 forecast files remain sealed (hashes in versions 2 and 3) and will be published unchanged with the 2027 scorecards.VERSION 4Version 4 (28 September 2026) adds the rolling-origin validation predictions and scores for all models and leads, the H4 analysis, the registered robustness checks (leave one ENSO event out, leave one country out, highest-phase mapping rule) and the brms check of the committed forecast model, with the code that produced them. These outputs underlie the manuscript submitted to Earth's Future. The two committed 2026–27 forecast files remain sealed (hashes in versions 2 and 3) and will be published unchanged with the 2027 scorecards.VERSION 3Version 3 (27 September 2026) commits a second, exploratory set of 2026–27 projections by hash, before the October 2026 FEWS NET release and before any target assessment exists. SHA-256 of projections_2026_27_M41_exploratory.csv: 02874adc50673523722b653208604977b62152e1a29f6eee6ae96511081e033e (created 2026-09-27 17:48 UTC). The model (M41: M3 plus a BYM2 spatial effect) was identified after the rolling-origin validation and is therefore exploratory. The confirmatory forecast for H3 remains the version 2 commitment (SHA-256 c180ca86ebdbe04469c881909227e0e6b001c630fe644d727d67b54925c0aa4c), which is unchanged. Both files will be published unchanged with the 2027 scorecards.VERSION 2Version 2 (27 September 2026) commits the prospective 2026–27 projections by hash, before the October 2026 FEWS NET release and before any target assessment exists. SHA-256 of projections_2026_27.csv:c180ca86ebdbe04469c881909227e0e6b001c630fe644d727d67b54925c0aa4c (created 2026-09-27 13:49 UTC). The file will be published unchanged after the February and June 2027 assessments, with the scorecards. This version also adds the analysis code, model inputs, the digitised official outlooks, and a change log declaring two deviations from Protocol v1 (R-INLA instead of brms for the prospective model; priors kept after the prior predictive check) and the implementation choices made before fitting.VERSION 1Preregistered protocol, frozen analysis data and processing code for an area-level Bayesian study of whether El Niño–Southern Oscillation (ENSO) information improves probabilistic forecasts of FEWS NET acute food-insecurity phases in 2,978 subnational areas of eight countries (Ethiopia, Kenya, Somalia, Uganda; Malawi, Mozambique, Zimbabwe, Madagascar). The model is a cumulative-logit ordinal model with a measurement layer for classification noise, BYM2 spatial and RW1 temporal effects. Confirmatory hypotheses: H1, ENSO/IOD information improves out-of-sample ranked probability skill over a history-only baseline; H2, ENSO effects follow the local rainfall teleconnection; H3, prospective projections for the Feb 2027 (both regions) and Jun 2027 (Southern Africa) assessments, issued from information available on 18 and 28 Aug 2026, are at least as skilful as FEWS NET's own projections. The package holds the protocol (PDF), a SHA-256 manifest, 16 frozen data files, the Python scripts that rebuild them from the raw downloads, and the Earth Engine and Copernicus download notebooks. No relationship between ENSO and any food-security outcome was estimated before this deposit. The projections will be committed by hash in version 2 before the October 2026 FEWS NET release.
Authors
- Johan G.L. Verheyden (ORCID: https://orcid.org/0000-0002-6129-2794)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-09-28
- DOI
- https://doi.org/10.5281/zenodo.22980950
- Primary Topic
- Food Security and Health in Diverse Populations
- Type
- article
- Field-Weighted Citation Impact
- 0.00