Classification: false_positive Confidence: Model confidence in dry/light rain events remains moderate. Recent 'correct' classifications on near-baseline peaks (Sept 6, 7, 10) suggest the model is good at saying 'quiet day', but struggles when there is some rainfall in the headwaters (Bands 3-5) under dry conditions, leading to over-prediction. The adjustments aim to tighten the dry-condition response of intermediate bands.
The model over-predicted the peak height by 28% and mis-timed the peak significantly during a very dry, light-rain event.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 2.69 ft | 2.1 ft | +0.59 ft |
| Total rise | — | 0.26 ft | — |
| Band | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|
| 1 | 0.24" | 0.00 ft | MODERATE | VERY DRY |
| 2 | 0.41" | 0.14 ft | MODERATE | VERY DRY |
| 3 | 0.54" | 0.27 ft | MODERATE | VERY DRY |
| 4 | 0.57" | 0.12 ft | MODERATE | VERY DRY |
| 5 | 0.64" | 0.17 ft | MODERATE | VERY DRY |
Headline: No rise indicated by recent rainfall Settled outcome: None (reached None) LLM said headline was correct: True Notes: The empirical model correctly identified a low probability of significant rise ('No rise indicated'). The actual rise was minimal (0.26 ft), consistent with the 'low' or 'below_low' tier expectations, whereas the physics model over-predicted the magnitude.
The event occurred under 'VERY DRY' antecedent moisture conditions (7-day avg 1.97") with LIGHT intensity rainfall (<0.15 in/hr). The gauge recorded a modest rise of 0.26 ft to a peak of 2.1 ft, with a very broad hydrograph shape indicating diffuse headwaters contribution. The predicted peak of 2.69 ft represents a 28.1% overestimation, exceeding the 20% threshold for 'correct' classification and fitting the 'false_positive' category (predicted meaningful rise, actual smaller) or 'partial'. Given the magnitude of error relative to the small total rise, it is a clear over-prediction.
The model likely attributed too much runoff efficiency to the bands, particularly Band 3 and 5, despite the dry soil conditions. The 'VERY DRY' tier multiplier is 0.05, which should heavily dampen the response, but the base response coefficients for Bands 2-5 seem to still generate a 0.59 ft excess rise when convolved with the ~0.4-0.6" of QPE in those bands. The timing error of -4.7 hours suggests the predicted lag was too long or the initial rise was predicted too late, but the primary failure is magnitude.
The empirical headline 'No rise indicated by recent rainfall' was partially correct in direction (low probability of significant rise) but the physics model issued active rise predictions. The discrepancy suggests the base response coefficients for intermediate bands (2, 3, 5) are still too high for dry conditions, or the moisture damping is not being applied effectively to the QPE signal in these bands.
Predictions issued today (archive): 24 active issuance(s) from 2026-09-11T05:20:02.778729+00:00 to 2026-09-12T04:20:02.023385+00:00; issued peaks ranged 1.92 to 2.69 ft. The prediction graded below is the HIGHEST issued peak of the day (issued 2026-09-11T15:20:03.184106+00:00), not the last card of the day.
| Band | Change | Reason |
|---|---|---|
| 2 | -5% | Over-prediction likely contributed by Band 2 (0.41" precip). Reducing response slightly to better match dry-condition runoff. |
| 3 | -10% | Band 3 had 0.54" precip and a high lag. In dry conditions, this band likely contributed less than modeled. A 10% reduction is a conservative correction for the magnitude over-prediction. |
| 5 | -10% | Band 5 (0.64" precip) is a major contributor to broad headwaters rises. Over-prediction suggests this band is too responsive in dry conditions. Reducing by 10% helps align with the observed 0.26 ft rise. |
Model confidence in dry/light rain events remains moderate. Recent 'correct' classifications on near-baseline peaks (Sept 6, 7, 10) suggest the model is good at saying 'quiet day', but struggles when there is some rainfall in the headwaters (Bands 3-5) under dry conditions, leading to over-prediction. The adjustments aim to tighten the dry-condition response of intermediate bands.