Classification: false_negative Confidence: Model confidence for Band 1 remains medium due to persistent gauge precipitation discrepancies. Saturated event tracking is active.
The model significantly underestimated a sharp 0.81 ft rise driven by widespread light rainfall on saturated soils, predicting only 0.08 ft from Band 5 and missing the near-gauge contribution entirely.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 3.67 ft | 4.46 ft | -0.79 ft |
| Total rise | — | 0.81 ft | — |
| Band | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|
| 5 | 0.21" | 0.08 ft | LIGHT | SATURATED |
The actual hydrograph shows a sharp 0.81 ft rise peaking at 4.46 ft early in the day, despite the gauge recording 0" of precipitation. This discrepancy suggests a 'gauge shadow' effect or sensor failure, where rain fell in the near-gauge area (Band 1) but was not captured by the point sensor. The QPE data shows consistent, albeit light (0.01-0.07"/hr), rainfall across all bands during the rise period (05:00-15:00). The sharp hydrograph shape indicates rapid response from Band 1 (0-2 hr lag), which the prediction engine completely missed, likely relying on the 0" gauge reading to suppress Band 1 contribution.
The prediction of 3.67 ft was effectively a no-rise prediction relative to the observed peak, resulting in a 17.7% error and a false negative classification. The model failed to recognize that on SATURATED soils (multiplier 2.0), even light, widespread rainfall can generate significant runoff. The 23.6-hour timing error further highlights the model's failure to detect the immediate near-gauge response. The predicted Band 5 contribution (0.08 ft) was negligible compared to the observed rise, indicating that headwaters were not the primary driver despite some QPE signal.
Given the saturated conditions and the sharp rise indicative of near-gauge input, Band 1 response coefficient needs adjustment upwards to account for unmeasured precipitation. Band 5 was over-contributed in relative terms (predicted as main driver vs actual minor role), but given its low rainfall and typical behavior, it is better to leave Band 5 stable and focus on correcting the missed Band 1 signal. The intensity tier 'LIGHT' is correct, but the base response coefficients for saturated conditions appear too low for widespread light rain events where gauge measurement fails.
| Band | Change | Reason |
|---|---|---|
| 1 | +20% | Sharp rise with 0" gauge precip suggests unmeasured near-gauge rainfall on saturated soils; maximum upward adjustment recommended to capture this hidden signal. |
Model confidence for Band 1 remains medium due to persistent gauge precipitation discrepancies. Saturated event tracking is active.