Classification: no_prediction Confidence: Model confidence remains low for Bands 2-4 due to recent instability and overprediction issues. The failure to generate a prediction today for a clear rainfall event indicates a systemic issue with the prediction trigger or data integration rather than coefficient accuracy. Until the physics engine reliably generates output for clear events, calibration of coefficients is halted to prevent compounding errors.
The prediction engine failed to generate a forecast for a significant 1.11 ft rise driven by widespread, multi-band rainfall on wet soils.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | N/A | 3.71 ft | N/A |
| Total rise | — | 1.11 ft | — |
| Band | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|
Headline: Watch — Recent rainfall is in the lower-bound range preceding rises to MEDIUM Settled outcome: verified (reached medium) LLM said headline was correct: True Notes: The empirical model predicted 'MEDIUM' tier and the actual outcome was verified as 'MEDIUM'. The empirical model performed better than the physics engine today, which failed to run.
On June 22, 2026, the watershed experienced a coordinated rainfall event with significant precipitation across all bands (Band 1: 0.47", Bands 2-5: 0.42-0.50") starting at 06:12 UTC. Despite zero recorded precipitation at the gauge itself, the wet antecedent conditions (WET tier, 1.93" 7-day avg) facilitated a substantial 1.11 ft rise, peaking at 3.71 ft. The hydrograph shape was very broad, consistent with the multi-band nature of the event and the wet soil conditions.
Critically, the physics-based prediction engine returned no output (Predicted peak: None), classifying this as a 'no_prediction' event. This is a significant operational gap, as the QPE data clearly showed rain that should have triggered a forecast. Unlike previous 'no_prediction' events which were unexplained anomalies, this event had clear input data that the model ignored or failed to process. The empirical model, however, successfully predicted the MEDIUM tier outcome.
Since no prediction was generated, no coefficient adjustments can be calculated via error minimization. The current coefficients, which have recently undergone reductions for Bands 2-4 due to overprediction, appear appropriately conservative. Increasing them now without a baseline prediction to compare against risks reintroducing the overprediction bias we are currently correcting. The failure lies in the trigger mechanism or data pipeline rather than the response coefficients.
No changes made.
Model confidence remains low for Bands 2-4 due to recent instability and overprediction issues. The failure to generate a prediction today for a clear rainfall event indicates a systemic issue with the prediction trigger or data integration rather than coefficient accuracy. Until the physics engine reliably generates output for clear events, calibration of coefficients is halted to prevent compounding errors.