Data Source

Observations come from the ASOS (Automated Surface Observing System) network via Iowa Environmental Mesonet (IEM). Roughly 2,500 stations across the US report temperature, dew point, pressure, and precipitation.

Models

• GFS — NOAA Global Forecast System, 0.25° resolution, out to 15 days (360h)
• AIFS — ECMWF AI Integrated Forecasting System, out to 15 days (360h)
• Kenny — Station-specific bias-corrected AIFS for temperature and dew point. For each ASOS station and valid hour (00Z/06Z/12Z/18Z), Kenny applies the recent 6-hour AIFS bias as a correction using forecast − bias (so a −6° bias adds +6° to AIFS, and a +6° bias subtracts 6°). If station-hour history is sparse, it falls back to a global valid-hour bias. Other variables are the same as AIFS.
• IFS — ECMWF Integrated Forecasting System (open data), out to 15 days (360h)
• NWS — National Weather Service point forecasts via the NWS API, limited to 7 days (168h)

Variables

• Temperature — 2-meter air temperature (°F). Matched to hourly METAR observations.
• Dew Point — 2-meter dew point temperature (°F). Matched to hourly METAR observations.
• Precipitation (6hr) — 6-hour accumulated precipitation (in). Requires at least 4 of 6 hourly reports to be present; computed from the final per-hour METAR accumulation total.
• Precipitation (24hr) — 24-hour (12Z–12Z) accumulated precipitation (in). Sum of four consecutive 6-hour periods ending at 12Z. Requires at least 20 of 24 hourly reports; shown only at lead times where the valid time falls on 12Z.
• Pressure — Altimeter setting (mb). Matched to automated 5-minute ASOS pressure observations.

Observation Matching

Each model grid point is bilinearly interpolated to the station location. For each forecast valid time, the nearest qualifying observation within a 30-minute window is used. Because ASOS stations report temperature and dew point at :56 past the hour (in METARs) and pressure every 5 minutes, a composite matching approach is used: the closest observation containing each variable is found independently, so no variable is penalized by the timing of another.

Metrics

• MAE (Mean Absolute Error) — average of |forecast − observed| across all matched cases. Lower is better; always ≥ 0.
• WMAE (Weighted MAE, precipitation only) — same absolute error, but each case is weighted by observed precipitation amount (minimum weight 0.01"). Heavier rain/snow events therefore contribute more than marginal events.
• Bias — average of (forecast − observed). Positive = model runs warm/wet/high; negative = model runs cold/dry/low.

Lead Times

Verification is computed at every 6-hour forecast lead from 6h to 360h (15 days). The map and station charts show a subset of these leads.

Averaging Periods

• All-time — cumulative stats since the system started collecting data. Forecast runs are kept for 21 days, but their error sums are folded into a running accumulator before deletion, so the all-time numbers grow indefinitely.
• Last 20 days — rolling window limited to the most recent 20 days of runs.