Classification: false_positive Confidence: Model confidence should be lowered for saturated/heavy events due to this significant overprediction. Previous events in dry conditions were well-calibrated, so the model is biased toward overprediction when moisture is high. Further calibration is needed to refine the saturated-state response.
The model significantly overpredicted the peak stage (2.58 ft predicted vs 1.25 ft observed) during a wet, intense rainfall event, likely due to excessive runoff coefficients for saturated conditions.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 2.58 ft | 1.25 ft | +1.33 ft |
| Total rise | — | 0.86 ft | — |
| Band | Zone | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|---|
| 1 | ? | 2.01" | 1.09 ft | HEAVY | VERY DRY |
| 2 | ? | 2.03" | 0.84 ft | HEAVY | VERY DRY |
Headline: No rise indicated by recent rainfall Settled outcome: None (reached None) LLM said headline was correct: False Notes: The empirical model predicted 'No rise indicated' (p_rise 0.0027), which was incorrect as a significant rise (0.86 ft) occurred. The empirical model likely failed to account for the heavy QPE rainfall in the 6-hour window or the antecedent moisture, leading to a false negative in its headline.
The event on 2026-10-02 involved significant rainfall (approx. 2 inches in both bands) with a 'HEAVY' intensity classification. The model predicted a peak of 2.58 ft, but the observed peak was only 1.25 ft, resulting in a 106% overprediction error. This indicates a false positive in magnitude, where the model anticipated a much larger hydrologic response than the watershed actually produced. The hydrograph shape was 'very_broad' with a rise duration of 13 hours, suggesting a slow, dispersed response typical of a large, possibly saturated or high-abstraction basin, but the magnitude was still capped well below the model's estimate.
A key factor in this overprediction is the 'SATURATED' moisture tier classification (7-day avg 2.821") combined with 'HEAVY' intensity. The current calibration uses multipliers that may be too aggressive for this specific wet-state response, or the base response coefficients (0.836 ft/in for Band 1, 0.616 ft/in for Band 2) are overestimating the runoff efficiency when the soil is already saturated. Given that previous events in 'VERY DRY' conditions were predicted accurately with lower coefficients, the jump in multiplier for saturated conditions is likely the primary source of error. The gauge's precipitation sensor recorded 0.0", but QPE showed significant rain, confirming the event was real but the runoff generation was less efficient than modeled.
Predictions issued today (archive): 22 active issuance(s) from 2026-10-02T05:20:02.860426+00:00 to 2026-10-03T02:20:02.555202+00:00; issued peaks ranged 0.35 to 2.58 ft. The prediction graded below is the HIGHEST issued peak of the day (issued 2026-10-02T13:20:02.874378+00:00), not the last card of the day.
| Band | Change | Reason |
|---|---|---|
| 1 | -15% | Reduce response coefficient by 15% to account for overprediction in saturated/heavy conditions. The 2.58 ft prediction vs 1.25 ft actual suggests Band 1's contribution was overestimated when moisture was high. |
| 2 | -10% | Reduce response coefficient by 10% to align with the lower actual peak. Band 2 likely also overestimated runoff during the saturated/heavy event, though slightly less than Band 1 given its tributary status. |
Model confidence should be lowered for saturated/heavy events due to this significant overprediction. Previous events in dry conditions were well-calibrated, so the model is biased toward overprediction when moisture is high. Further calibration is needed to refine the saturated-state response.