Precip48 — Mid-Atlantic Event QPF Workbench

PRECIP48

SYNTHETIC DEMO

Mid-Atlantic event-aware probabilistic QPF workbench

UTC

Forecast product 48 h

Spatial uncertainty display

Amount/scenario

Map accumulation

Forecast hour: 12

012243647

Scenario mode selects one complete synthetic member by its approximate AOI 48-hour score, so track, timing and footprint move together. It is an AOI-ranked scenario, not a pixel percentile.

Selected synthetic point

40.000, -76.900

–selected product mm

–P50 total mm

–best peak mm/h

–best events

Validated local point feed

Checking local API. The map remains synthetic.

Lead-time blend logic

Mid-Atlantic precipitation field

scenariomember –48-hour total

Precipitation (mm)

0102550100+

Event clock

An event closes only after 6 predicted dry intervals. Five is still the same event. Threshold: 0.254 mm per interval; horizon closure is predicted, not observed confirmation.

Source readiness

Catalog availability and transition dates are documentation claims, not verified live integrations.

The synthetic map is independent of all listed feeds.

A production view must show successful retrieval, freshness, coverage and provenance before it reports a feed as available.

Core evidence

0-2 h

MRMS motion/QPE + HRRR/RRFS

7-18 h

CAM + high-res ensemble + NBM

37-48 h

NBM + ensemble/global + WPC

Hourly precipitation distribution at selected point

P90P75P50P25P10Best guess

Best-guess events

Central-window totals

Programmable forecast-input catalog shown

Browser-direct is reserved for lightweight JSON/WFS. GRIB2/S3 belongs behind the backend cache so the user's phone is not asked to become NCEP.

Rows preserve source-catalog documentation. They do not certify a verified feed, current freshness, or an active integration.

Tier	Dataset	Role	Resolution / cadence	Access	Status

Event definition

A wet hour is ≥ 0.254 mm by default. An event begins at the first wet hour and closes only after six consecutive dry hours. The event ends at the last wet hour before that confirmation gap.

event = { t_start … t_last_wet }
close(event) ⇔ dry(t+1)…dry(t+6)

Member-first event total

Each member is summed over the central-estimate event window before the distribution is summarized. These are central-window totals; displaced or outside-window event rainfall is not captured.

Q_m(x,e) = Σ[t∈e] P_m(x,t)
Q_q(x,e) = Quantile_q {Q_1 … Q_M}

MRMS correction decay

Experimental correction is disabled by default.
Eligible residual taper: max(0, 1-h/6)

This prototype has no live assimilation or calibrated correction.

Zero-inflated rainfall

P(Y<0.254)=1-p_wet
Y | Y≥0.254 ~ synthetic member distribution

This synthetic demo displays raw pseudo-member frequencies; no calibrated probability claim is made.

Processing chain

1 · IngestMRMS, stations, NWP, NBM, WPC

2 · NormalizeUTC, mm, common grid, provenance

3 · CalibrateMRMS bias, PoP, QMD, terrain residuals

4 · Events6-dry-hour segmentation per member

5 · Servetiles, JSON, coherent scenarios

Bundle contract

This HTML is the interactive prototype. The ZIP includes the complete build prompt, 50-source catalog, event code, MRMS downloader, schemas and backend shell.

Production requirement: every displayed pixel must be traceable to source cycle, valid time, ingest time, transformations and weights.

Rules that should be hard to violate

No percentile created by scalar multiplication of one map.
No MRMS bias correction without quality weighting.
No browser-embedded provider secrets.
No source silently substituted when stale/missing.
No claim that parallel RRFS/REFS is operational before actual implementation.
No event split after only five dry hours.