Classification: false_positive Confidence: Model confidence remains medium. This event highlights a critical reliance on ground-truth gauge precipitation sensors (00045) to override QPE inputs when they diverge significantly. Future prediction logic should heavily weight gauge precip over QPE when gauge data is available.
The model predicted a significant rise to 3.19 ft based on QPE data, but the gauge recorded 0.0 inches of precipitation and remained flat at ~0.88 ft, indicating a complete failure to predict no event.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 3.19 ft | 0.88 ft | +2.31 ft |
| Total rise | — | 0.05 ft | — |
| Band | Zone | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|---|
| 1 | upper_richland | 1.25" | 2.45 ft | INTENSE | WET |
| 2 | falling_water | 0.76" | 0.94 ft | HEAVY | WET |
Headline: ~20% chance of rising to LOW in the next 6–12 h. Experimental. Settled outcome: None (reached None) LLM said headline was correct: True Notes: The empirical headline suggested a low probability (~20%) of reaching the LOW threshold. The gauge did not rise, consistent with the low probability assessment.
The predictor issued a forecast of a 3.19 ft peak based on substantial QPE inputs (1.25" in Band 1, 0.76" in Band 2). However, ground truth measurements from the gauge precipitation sensor (USGS 00045) show 0.0 inches of rainfall in 24 hours, and the gauge stage remained effectively flat with a maximum of 0.88 ft. This represents a massive overprediction (262.5% error) and a classic false positive where QPE data failed to match ground conditions.
The discrepancy suggests that the QPE grid data significantly overestimated precipitation for this specific event, or that the rainfall was highly localized in a way not captured by the broader grid cells assigned to these bands. Given the zero gauge precipitation, the calculated runoff contributions were entirely fictitious. The empirical forecast headline, which indicated a ~20% chance of rising to LOW, was prudent and correct in suggesting low probability, as the gauge did not rise.
Since the actual precipitation was zero, the true coefficient values could not be validated against actual runoff. Therefore, adjusting coefficients based on this event would be dangerous as it would penalize the model for data input errors (QPE vs Gauge precip mismatch) rather than hydrological response errors. The most conservative and accurate action is to make no changes, treating this as a data-quality issue rather than a calibration need.
No changes made.
Model confidence remains medium. This event highlights a critical reliance on ground-truth gauge precipitation sensors (00045) to override QPE inputs when they diverge significantly. Future prediction logic should heavily weight gauge precip over QPE when gauge data is available.