Constraining & Calibrating the HWT Return Period

Goals:

• To validate the data driven return period flow used to visualize FIM.
• Scaffold out post-processing improvements.

Outcomes and Takeaways:

• A deeper appreciation for the forecast-to-operations process.
• A better grasp of the tangled dependencies of this wicked system.

(placeholder)

Slide layout: default     Items linked/bordered in green are cited in the tooltip on hover.

: narrative-oriented     Items linked/bordered in blue are hyperlinked to relevant resources.

Datasets

Links to: https://water.noaa.gov/gauges/SACT2

Directed clamps

Regional breakdown: Pacific NorthWest

Loose responses:

1. As Brent calls out, a blanket value is unlikely to result in the desired changes, since we are modeling a coupled system and streams respond differently depending on their particular geo-physiographic characteristics. A stream order 1 does not respond in the same way that a stream order 6 does.
2. A more complex workflow could include direct threshold based on hydrofabric or FIM derived attributes but that breaks the more approachable “knob” that a single basin value exposes.
3. I do think implementing that logical threshold check (Q>2) could be useful but again, that involves more extensive workflow alterations than the drop-in replacement this was meant to replace.
4. I would avoid using a “precipitation present” check, since the NWM includes hydraulic routing and we wish to capture large flood pulses and “sunny day flooding”.
5. We have created a pre - post map of the called out domains. The changes to the HWT are inline with the called out values and much improved over both that initial 2.9 as well as the 2.2 Brent calls out.
6. Basin boundary issues called out in #2 are still present but should be lessened now that the threshold distributions are tighter.
7. We have started exploring direct paramiterizations using categorical discharge values to clamp maxima. It is critical to note however, that if the water model distributions are not in line with recorded observations, there is still a large potential that large categorical flooding values fall below the calculated HWT discharge.

Describing changes

Statistical Variations & Visual Trends

Site ID	Flowpath ID	Hours (N)	NSE	PBIAS (%)	r
12131500	23963211	51,876	0.68	−5.0	0.87
12134500	23963021	303,979	0.73	−9.9	0.86
12138160	23963701	271,012	−0.17	−18.7	0.55

Gage Spesific Behavior

Regional breakdown: (over the shared domain)

Regional breakdown: (in the only state that exists :) )

Regional breakdown: (the thrid hottest state in the union)

A “Gut check” assessment

What’s Next?

Outcomes and Takeaways:

• It seems that in the Pacific northwest at least, the distributions of discharge predicted by the National Water Model (v3) do not match the desired benchlines.
• This is a wicked system; CATFIM_Q is a poor separation of concerns relative to the RFC/WFO perspective - use case - FIM access pattern?

Next Steps:

• Opportunities in FIM cross-pollination
  • More CATFIM_WSE
• Cost estimates?
• Solve Hydrology.

This presentation

Sources:

Explaining the Math: How We Picked the AEP

From Flow to AEP (Annual Exceedance Probability):

Every river behaves differently, so a “one-size-fits-all” equation doesn’t work. Here is how we let the data decide:

1. Rank the History: We sort the historical peak flows for the gauge from smallest to largest to see the actual behavior.
2. Test 5 Standard ‘Shapes’: We apply 5 standard hydrological distribution curves (like GEV, LN3, or Gumbel) to the data. Think of it like trying 5 different tailored suits on the river.
3. Measure the Error: We calculate the mathematical distance (RMSE) between the actual data points and each theoretical curve.
4. Pick the Winner: The curve with the smallest error “wins” and becomes our official profile for that gauge.
5. Assign the Threshold: We use that winning curve to look up exactly what Return Period / AEP matches our High Water Threshold (HWT) discharge.

To get a single HUC8 value: We look at the gauges inside that basin and select the most conservative (minimum) AEP to ensure we don’t miss early flooding signals.

The Bottom Line for the Field:

We didn’t guess or hard-code the threshold. We let the local historical data choose the most accurate mathematical model, and used that model to define the risk.