Somalia Riverine Flood Trigger / indicator explorer

Indicator explorer

Pick a basin and a year to see how each streamflow indicator behaved against the SWALIM gauge record. Every source — GEOGloWS included, via its retrospective — is judged against its own 1-in-6-year threshold: absolute discharge is never compared across models.

Basin Year

Observed water level (m). Dashed: official Moderate (amber) and High (red) flood-risk levels. Shaded bands: the Gu and Deyr trigger windows.

Stations over their own 1-in-6 threshold, per source

Colored lines: per source, how many of the season's selected stations exceed their own seasonal 1-in-6-year threshold that day (each source thresholded on its own 1999–2023 climatology). Black line: the trigger's decision variable — the number of pool (station, model) pairs over their adopted thresholds (a station carried by two models counts twice). Dashed grey: the required count. Amber ticks: days the trigger activates.

Each source's discharge at the reference station divided by its own seasonal 1-in-6-year threshold there — 1.0 (dashed) means "at threshold". This is the bias-free way to overlay models whose absolute magnitudes differ by 5–10×.

Reading it. The top panel is the ground truth: the SWALIM gauge the trigger is calibrated against. The middle panel is the trigger's actual decision variable — the station consensus. The bottom panel shows the shape and timing of each model's signal at the reference station. A model can track the river well (bottom) yet peak days late or under-rank an extreme (middle) — GEOGloWS in Deyr 2023 on the Shabelle is the canonical example.

Method. Series are each model's reanalysis/retrospective (this explorer is about indicator behaviour, not forecast lead time — see the trigger report for the reforecast backtests). Consensus counts exist only inside the Gu (Mar–May) and Deyr (Oct–Dec) windows, because the trigger does. Thresholds are Weibull plotting positions on each source's own seasonal maxima, 1999–2023 — we assume bias between reanalysis and forecast for all sources, so operational thresholds must likewise be refit on the operational product's own climatology. Data baked by scripts/export_explorer_data.py.