Classification: false_positive Confidence: Confidence remains medium. The model is consistently susceptible to QPE artifacts. Until QPE validation against gauge precip is improved or coefficients are dynamically scaled by data confidence, false positives will persist when QPE shows rain but gauge does not.
The model falsely predicted a 2.63 ft peak based on late-day QPE rainfall that was not corroborated by gauge precipitation sensors, resulting in a massive overprediction against a flat hydrograph.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 2.63 ft | 0.46 ft | +2.17 ft |
| Total rise | — | 0.09 ft | — |
| Band | Zone | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|---|
| 1 | upper_richland | 1.35" | 1.76 ft | INTENSE | NORMAL |
| 2 | falling_water | 1.42" | 1.36 ft | INTENSE | NORMAL |
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 predicted a low probability of rising to 'low' tier (~20%). The actual outcome was 'below_low', consistent with the headline's conservative assessment despite the deterministic predictor's failure.
Today's event presents a classic false positive driven by unverified QPE data. The model predicted a significant rise to 2.63 ft, attributing 1.76 ft to Band 1 and 1.36 ft to Band 2 based on intense rainfall signals at 23:13 CT. However, the gauge precipitation sensor (USGS 00045) recorded 0.0 inches for the 24-hour period, and the hydrograph remained flat with a peak of only 0.46 ft. This indicates that the QPE rainfall at the end of the day was likely noise or misclassified precipitation that did not translate into runoff, or that the rain did not reach the gauge's immediate catchment area.
The large error (471.7%) stems from the model's reliance on QPE without sufficient ground-truth validation. In previous events, such as July 6th, a similar false positive occurred where QPE predicted rain but the gauge remained flat. The current coefficients (0.929 for Band 1 and 0.684 for Band 2) are already reduced from earlier calibrations, yet they still produce large responses when fed inaccurate QPE. Since the 'rain' itself was false rather than the runoff response being incorrect, adjusting coefficients further downward risks under-predicting future genuine events. The primary issue is data quality (QPE vs. Gauge), not coefficient calibration.
Therefore, no coefficient adjustments are recommended. The model correctly identified the potential for a rise based on input data, but the input data was flawed. The empirical forecast headline, which suggested only a ~20% chance of rising to 'low' tier, was technically correct as the gauge stayed in 'below_low', but the deterministic predictor failed significantly. Confidence remains medium due to ongoing issues with QPE reliability in this watershed.
No changes made.
Confidence remains medium. The model is consistently susceptible to QPE artifacts. Until QPE validation against gauge precip is improved or coefficients are dynamically scaled by data confidence, false positives will persist when QPE shows rain but gauge does not.