Station
Loading this station’s forecasts and observations…
One day: forecasts and outcome
Map and nearby stations
Forecast office ↗Current forecasts
Reading the issued ensemble cutout…
Shading is the 10–90% range of issued downscaled ensemble members, not a confidence interval for an observation. The dashed native baseline is sampled onto the same cell. These native forecast extrema use the printed UTC windows, which differ from the station’s full climate day. The daily high/low selector above also changes this view; rain retains the temperature outlook.
Summary
Model rankings
Choose the models in the contest
Models need the selected minimum and at least 80% of accepted observation dates in this window. Every ranked model then uses their common intersection. Excluded models remain in the coverage audit. Persistence is a separate baseline. Selecting fewer models can change the dates and results.
Compare two models
Accuracy by lead time
Each cell reports its own sample size. Gain uses identical model/NBM dates within that cell; MAE, RMSE and bias use each model’s available accepted cases. Samples can differ across cells. Click a cell to select its model and forecast lead.
Error charts
The dashed diagonal is a perfect forecast. Points above it are too warm/wet; below it too cold/dry. Each point has its date and exact values.
Colors show Model A’s signed error. Blank gray days are missing or unscored; they are never treated as correct forecasts.
Accuracy by type of day
Regimes use the observed weather, retrospectively. Rankings use the main contest’s common dates, require at least five cases per regime, and can overlap. They describe the past; they do not choose tomorrow’s model or establish the cause of an error.
Only contest models on the same common dates compete. Ties split one day’s credit equally. A tiny advantage and a large advantage both count as one daily result; the MAE board measures magnitude.
These accepted observations remain in every applicable score. Review the daily replay and source windows before interpreting an unusual case.
Rain forecasts around this station
Native station-grid cells and exact 10/25/50-mile circles are verified against the selected rainfall analysis. These are separate from station-gauge observations and the daily temperature controls above. NDFD forecasts must be available before the accumulation begins; this view uses the 0–24-hour pre-start lead. Mean depths and errors use the same valid area. Wet-area percentages use at least 0.01″ and describe deterministic rainfall coverage, not probability-of-precipitation verification. Analyses are provisional and can be revised.
Climate normals, 1991–2020
1991–2020 station referenceMonthly reference values and completeness
Data coverage and sources
A dash means unavailable, not zero error. Forecast availability varies by model, element, lead and archived run length. Every current run is labeled; forecasts older than 36 hours are excluded from the disagreement summaries.
Full daily forecast table
Held observations and quality-control reasons
Forecast windows, extraction and decision-time provenance
The decision cutoff is 12Z on the issue day, using publication availability rather than initialization alone. Some availability lags are conservative estimates. NDFD/NBM daytime and overnight extrema differ from official full local-standard calendar days; global interval-midpoint windows can differ at boundaries. Identical dates do not make those meteorological windows identical. The distribution replay separately identifies eligible scoring windows. No future forecast is included in a verified score.
Station observations verify the station point. URMA and Stage IV rainfall analyses verify grid cells and neighborhoods, separately in Rain Bullseye. Downscale Duel shows forecast changes until matching gridded truth is available. Full methodology ↗