Classification: correct Confidence: Model confidence in magnitude is high, but timing confidence is low due to persistent systematic lag. Coefficients are stable.
The model correctly predicted the peak magnitude (2.06 ft vs 2.18 ft actual, -5.5% error) despite a consistent ~10-hour timing lag, with significant QPE inputs but zero recorded gauge precipitation.
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 2.06 ft | 2.18 ft | -0.12 ft |
| Total rise | — | 0.74 ft | — |
| Band | Zone | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|---|
Headline: WATCH — Recent rainfall is within the typical range of historical rises to LOW Settled outcome: no_change (reached below_low) LLM said headline was correct: True Notes: The empirical headline predicted 'WATCH' with target tier 'low' (meaning rise to ~3.0 ft range potentially, or just attention). The actual outcome was 'no_change' in tier (stayed below_low). The headline was cautious ('within typical range of rises to LOW'), so it was not wrong, just not fully realized. The gauge did not reach the 'low' threshold of 3.0 ft.
The prediction was highly accurate regarding magnitude, falling well within the acceptable ±15-20% error margin. The predicted peak of 2.06 ft compared to the actual 2.18 ft represents a mere 5.5% underestimation, which is excellent for hydrological modeling. However, the timing error remains a persistent systemic issue; the model predicted the peak for ~04:39 the next day (UTC), while the actual peak occurred at 13:00 CT (18:00 UTC) on the day of the rain. This ~10-hour discrepancy matches the pattern seen in previous events (May 5 and May 10), suggesting the lag calculation or the way the model aligns precipitation input with response time needs review, rather than the response coefficients themselves.
Despite QPE showing ~0.6-0.7 inches of rain across both bands, the gauge precipitation sensor recorded 0.0 inches. This discrepancy is notable but common in radar-based QPE vs local gauge data. The model generated a prediction based on the QPE, and the resulting rise (0.74 ft) was captured accurately. The broad hydrograph shape suggests the rainfall was distributed or the basin response was attenuated, consistent with the NORMAL moisture tier (1.22" 7-day avg). Since the magnitude prediction was spot-on, no coefficient adjustments are warranted. Adjusting coefficients now could destabilize the good magnitude fit to try to fix a timing issue that is likely structural/lag-related.
No changes made.
Model confidence in magnitude is high, but timing confidence is low due to persistent systematic lag. Coefficients are stable.