The proposed trigger — two candidate designs
Multi-source consensus
Each window votes on (station, model) pairs drawn from the five SWALIM gauges that still report, with all three providers competing freely and a station allowed to carry more than one model. Joint basin calibration: 1-in-3.2 overall, 1-in-4.3 / 5.2 per basin. Includes the SWALIM-threshold check, the impact tilt and a no-Google variant.
/trigger/T. Downing · Aug 2026The balanced trigger
The same open multi-model consensus on the five still-reporting gauges, but with the activation probability balanced across basins and seasons as an explicit constraint rather than left to fall out of the search. Seasons come out exactly level; basins within one activation; 7 of 7 severe years and no activation outside the benchmark.
/balanced/design variant · Sep 2026One model per window
Each river-season window runs on a single forecast source rather than a mixture, monitoring all seven reporting-era points and with no threshold below 1-in-3. Google carries Gu, GloFAS v5 carries Deyr; GEOGloWS is an alternative view. Same 1-in-3.2 envelope. Carries twelve open review notes behind a toggle.
/trigger-single-model/P. Wairimu · Aug 2026Explore the evidence
Mixed models vs one model per window
Both setups on GloFAS and Google only, both on the five reporting gauges, both balanced so each basin-season carries the same activation probability at 1-in-3.2. Two outputs: the year-by-year trigger record, and accuracy (F1, TPR, PPV, lead time) against the gauges and against a target list built from CERF allocations and major EM-DAT years.
/comparison/design delta · Sep 2026Somalia floods analysis: data-source review
Presentation summary of the evidence: SWALIM thresholds, events and crossings, gauge events against flood exposure, model performance on hindsight and forecasts, bias correction, the seven trigger gauges, the grid search, the current best configuration and the envelope decision it implies.
/summary/updated as the analysis movesIndicator explorer
Pick a basin and a year and see how each indicator behaved: the SWALIM gauge record, each source's station-consensus count against its own 1-in-6 thresholds (GEOGloWS included, via its retrospective), and each model's signal at the reference station relative to its own climatology.
/explorer/1999–2023, dailySource material
Analysis notebooksGitHub
The full evidence chain: SWALIM threshold analysis, model-vs-gauge skill, lead-time skill, flood benchmarks, station selection, trigger grid search and the staged-mechanism exploration (notebooks 01–10, P. Wairimu) plus the multi-source mechanism itself (notebook 11).
ds-aa-som-floods/analysisKnowledge-base pageGitHub
The team knowledge-base landing page for this analysis: what was analysed, the headline findings, data sources and how this relates to the (future) Somalia flooding framework.
ds-knowledge-base