Yesterday our Mill Creek reach jumped from 6.2 to 8.9 ft in 27 minutes; HRRR QPF plus a quick HEC‑RAS 2D run gave us 14 minutes to close low-water crossings and push a CodeRED alert. For those feeding ensemble QPF into stage triggers, what probability/lead-time combo is changing your response timing (say 30% at 45 minutes vs 60% at 20), and how are you verifying after-action to tighten thresholds?
After getting burned by a 25‑minute jump, we switched to two triggers: we close low crossings at about 40% exceedance with ≥30‑min lead if HRRR and the time‑lagged ensemble align, but we require about 60% when lead is <20 minutes for CodeRED — basin‑specific 1‑hr excess depth drives both. when HRRR flips runs, so we verify with MRMS gauge‑corrected QPE (https://mrms.nssl.noaa.gov) against USGS stage residuals, score a quick Brier, and nudge the exceedance depth monthly; that “14 minutes” window is where a warm‑started 2D run still buys us 5–8 minutes — do you treat urban basins differently?
@frank6532 We use 60% at 20 when HRRR+MRMS agree. If cells train, 30% at 40 and verify via gage-cams.
And on our Cedar Creek reach we use 45% at 30 minutes when HRRR plus the 3‑run time‑lagged HRRR agree and API>0.9; otherwise it’s 55% at 25. We verify after‑action by syncing gage‑cam timestamps with USGS 5‑min stage and the HEC‑RAS 2D cell at the crossing, and we track hit/miss on rise rate >0.3 ft/5 min. Small caveat: we drop the threshold about 10% after dark for crew roll time, @frank6532 — have you tried that?
We’ve settled on 50% at about 25 minutes when HRRR QPF lines up with NWM Short‑Range channel_rt and an upstream rise rate >0.3 ft/5 min; otherwise 35% at 45 just closes the low crossings… For AAR, we sync HEC‑RAS 2D hydrographs to gage‑cam timestamps and MRMS QPE, and we pad for the ‘14 minutes’ reality because CodeRED + barricades eats 6–8 minutes — .
We drop to 40% at 35 if crew ETA >12; ‘14 minutes’ is tight; verify via gage-cam timestamps.
On our Pine Run we pull the trigger at 55% with about 20 min lead when HREF neighborhood prob >0.75"/hr sits over the headwaters for two frames and NWM soil saturation >0.85; otherwise it’s 35% at about 45 just for barricades. @christopher_dav16 we skip rise-rate and nudge lead-time using optical-flow cell speed — if vectors point at the outlet >7 m/s we shave 5 minutes. We verify by lining up USGS provisional peaks with CodeRED send-times and AVL pings; snagging 5 minutes there’s like finding a quarter in the couch.
Quick example: if HRRR locks on a cell train, I go 55% at about 20 min once MRMS 5‑min rate (NSSL Projects: Multi-Radar/Multi-Sensor System (MRMS)) holds >0.9 in/hr over the upper third; otherwise 35% at about 45 — 14 minutes is brutal, . HEC‑RAS 2D travel‑time rasters have been gold for turning that into crossing ETAs. After‑action, I match CodeRED delivery timestamps with 1‑min SCADA level data and USGS provisional edits and tweak the trigger if they drift.
I’ve had better luck keying the trigger to rise rate: once the upstream gage hits 0.3 ft per 5 min, I drop the ensemble threshold from 50% to about 35% so crews beat the dips. > crossing ETAs. After‑action, I match CodeRED delivery timestamps with 1‑min SCADA level data and USGS provisional edits and tweak — same here, but I cap that drop if MRMS placement looks off by >5 km to avoid false sprints.