Classification: false_negative Confidence: Model confidence remains 'medium'. While we are correcting underprediction, the consistent failure to capture saturated peaks suggests the base coefficients are still too low. Two consecutive events showing underprediction in saturation warrants this second adjustment.
The model significantly underpredicted the peak stage (3.09 ft vs 4.06 ft actual) during a saturated condition event, despite minimal rainfall contributing to a sharp rise likely driven by antecedent moisture release.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 3.09 ft | 4.06 ft | -0.97 ft |
| Total rise | — | 1.01 ft | — |
| Band | Zone | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|---|
The prediction yielded a peak of 3.09 ft, while the actual peak was 4.06 ft, representing a -23.9% error. This is a clear false negative, as the model failed to capture the magnitude of the rise. The watershed was in a SATURATED state (7-day avg 4.97"), meaning the multiplier of 2.0 was active. However, the rainfall inputs were extremely low (0.06" Band 1, 0.05" Band 2), yet produced a 1.01 ft rise. This suggests the current response coefficients (1.01 for Band 1, 0.777 for Band 2) are still insufficient to account for the high runoff efficiency in saturated conditions, even with the moisture multiplier applied.
The timing error of nearly 24 hours is also significant, though the hydrograph shape (sharp) and the 'daily max' timestamp at midnight suggest the rise was immediate or carried over from previous saturation dynamics rather than a typical delayed runoff from today's negligible rain. The previous event (May 29) already identified underprediction in saturated conditions and increased coefficients by 10%. The persistent underprediction indicates the response coefficients need further upward adjustment to reflect the true hydrologic sensitivity of the watershed when fully saturated.
Given the sharp rise and saturated conditions, the issue is primarily magnitude. The model is still too conservative. We must increase the response coefficients again, respecting the 20% maximum change limit per event. Band 1, being the primary headwaters with more cells, should take a slightly larger share of the adjustment to correct the systemic underprediction.
| Band | Change | Reason |
|---|---|---|
| 1 | +15% | Underprediction continues in saturated conditions; increase Band 1 response to better capture high runoff efficiency. |
| 2 | +10% | Band 2 contributed to the sharp rise; modest increase to align with overall watershed sensitivity in saturated states. |
Model confidence remains 'medium'. While we are correcting underprediction, the consistent failure to capture saturated peaks suggests the base coefficients are still too low. Two consecutive events showing underprediction in saturation warrants this second adjustment.