Classification: no_prediction Confidence: Model confidence is currently hampered by a systemic failure to generate predictions when Band 1 QPE data conflicts with gauge zeros. Until the ingestion logic is updated to weight QPE in the absence of gauge confirmation, the physics model will continue to miss events. Empirical model needs threshold refinement for wet-soil conditions.
The physics prediction model failed to generate an output for today's significant 1.63 ft rise event, while the empirical forecast correctly indicated no rise despite the eventual peak reaching the medium tier.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | N/A | 3.82 ft | N/A |
| Total rise | — | 1.63 ft | — |
| Band | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|
Headline: No rise indicated by recent rainfall Settled outcome: None (reached None) LLM said headline was correct: None Notes: The empirical headline 'No rise indicated' was generated. The actual peak reached 3.82 ft, which is in the 'medium' tier (threshold 3.4 ft). The empirical model failed to predict the magnitude rise, classifying it as below the percentile threshold for a rise, despite the significant rainfall volume on wet soil.
A substantial hydrological event occurred with a 1.63 ft rise to a peak of 3.82 ft, driven by multi-band rainfall (particularly a 0.74" burst in Band 1 and significant volume in Bands 2-5) on WET antecedent soil conditions. The prediction model failed to generate any output (Predicted peak: None), which is a critical failure for an event of this magnitude. This non-generation is likely due to the known data conflict where QPE shows significant Band 1 rainfall (1.13" total) but the gauge recorded 0", causing the predictor to potentially discard the data as invalid or insufficient to trigger a response calculation. The empirical model, conversely, output a 'No rise indicated' headline, which was technically correct regarding the forecast at the time (based on low 6h QPE sum in its window), but the actual outcome exceeded the 'low' threshold into 'medium' (3.82 ft > 3.4 ft), highlighting a sensitivity gap in the empirical thresholds for wet-soil multi-band events.
Since no prediction was generated, coefficient adjustments cannot be calculated based on error magnitude. However, the root cause appears to be the persistent issue with Band 1 QPE-Gauge mismatch preventing model activation. The system effectively ignored a major headwaters/near-gauge mixed event. The broad hydrograph shape and rise duration (6.5 hours) confirm the contribution of Bands 2-5 on wet soil, consistent with previous wet-event calibrations, but the lack of prediction output prevents fine-tuning these coefficients further today. The failure to predict underscores the need to trust QPE for Band 1 when gauge data is missing/zero during known storm events, rather than discarding the event entirely.
No changes made.
Model confidence is currently hampered by a systemic failure to generate predictions when Band 1 QPE data conflicts with gauge zeros. Until the ingestion logic is updated to weight QPE in the absence of gauge confirmation, the physics model will continue to miss events. Empirical model needs threshold refinement for wet-soil conditions.