Abstract
<title>Abstract</title> <p> Intraseasonal rainfall breaks spells of consecutive dry days within the monsoon season are a principal driver of crop failure and food insecurity in the Sahel, yet they remain difficult to anticipate. This study evaluates the Sahel Integrated Radiative Index (IRIS), an experimental scalar built by projecting a 22-variable, multi-frequency observation vector (passive-microwave brightness temperatures from 19 to 92 GHz combined with ERA5 thermodynamic and dynamic fields at the surface, 700 hPa and 200 hPa) onto its first principal component, as a candidate diagnostic and early-warning input over the domain 10–20°N, 18°W–45°E for 1992–2025. Rather than asserting the superiority of IRIS, we adopt a deliberately conservative, comparative and reproducible framework organised in three declared epistemic layers: transcribed source statistics, deterministic algebra on the published construct, and a synthetic observing-system simulation experiment (OSSE) used only for illustration and for pre-registering falsifiable predictions. The dominant mode reported in the source dataset explains <bold>72.1%</bold> of the standardised variance; we retain this value as reported and, in a transparent reconciliation ledger, document its tension with the heterogeneity of the published loading vector, for which an independent standardised recomputation is required. In the synthetic experiment, the ROC-AUC for anticipating dry-spell onset at a 7-day lead ( <bold>≈ 0.66</bold> ) is close to a perfect-information ceiling (≈ 0.68), suggesting that the limiting factor is the stochasticity of rainfall occurrence rather than the index; consistently, a single 22-GHz channel reproduces most of the score, a direct corollary of a strongly low-rank system. The frequently cited “4–6 week anticipation” is more parsimoniously interpreted as a biosphere-response lag (IRIS leading NDVI by roughly one month) than as onset predictability. A synthetic scenario projection to 2050 yields a small forced shift of <bold>+ 0.05 to + 0.08 σ</bold> , a fraction of interannual variability and thus not separable from zero at seasonal scale. We conclude that IRIS shows potential as a compact, physically-grounded state descriptor in the studied region, while its predictive claims require independent, out-of-sample and multi-climate validation before operational use. </p>