Precip48
One rainfall number hides most of what matters. Precip48 forecasts rain as events: when each one starts and ends, how much falls, how much the models disagree, and what a low, middle and high outcome would actually look like on the map.
A wet hour is 0.01″ (0.254 mm) or more. Breaks shorter than six dry hours stay inside the same event.
Case viewer: Sep 2, 2026 12Z
Five RRFS ensemble members, 48 hourly steps each, checked against MRMS. Member maps, percentiles, exceedance odds, coherent scenarios and Camp Hill-area event timing.
Design prototype · demo dataWorkbench prototype
How the finished tool will feel: map, point timeline, scenario switcher and source inventory. Every value here is synthetic.
Live now · GEFSEnsemble Meteogram
Today's 31-member GEFS precipitation and snow odds for any point in PA, the live sibling to Precip48 at longer range.
Live now · MRMSRainfall Point Map
What actually fell: radar and gauge totals by day or range, the truth Precip48 is verified against.
Three products that are easy to mix up
Percentile maps
P10 to P90 computed cell by cell across the members. Useful for "how much could fall here", but a P90 map is not a storm that can happen: each cell's high value may come from a different member.
Exceedance odds
The share of members at or above 0.01″, 0.25″, 0.5″, 1″ and 2″. With five members the only possible values are 0, 20, 40, 60, 80 and 100 percent, and they are not yet calibrated against history.
Coherent scenarios
Lower, central and upper outcomes are whole model runs chosen by rank, never stitched together. In the validated case they are members 4, 3 and 5, and the page always says which member it is showing.
The validated case
MRMS Pass 2 is used only after the fact, as truth to verify against, and never feeds the forecast. Area totals and edges are computed with conservative area-overlap remapping, because the simpler centre-point method shifted hourly values by about 1 mm at the 95th percentile, mostly near sharp gradients and edges.
What it takes to run it live
- Pull each new RRFS ensemble cycle (5 members × 48 hours) on the lab server, the same way the GEFS and MRMS pipelines already run.
- Replay the Sep 2 case and require identical numbers before anything goes public.
- Publish versioned products per cycle (members, percentiles, odds, scenarios, events) with clear fresh, stale or incomplete status.
- Verify every past cycle against MRMS and build the record needed to calibrate the probabilities.
- Turn the prototype into the live workbench on this site.