Model Performance — Northwest Air Quality Forecast

1 · How accurate is the forecast?

Mean bias

–

Typical error (RMSE)

–

Correlation (pooled / median site)

–

Paired values

–

2 · Does it track observations?

Forecast vs observed

every hourdaily averages by site

3 · How fast does skill fade?

Bias by forecast lead hour

Daily domain mean — ● observed — forecast

Day-by-day tracking of the whole filtered domain: dots are what monitors measured, the line is what was forecast.

4 · Where is it weakest?

Sites with largest RMSE

The hardest sites — usually smoke-affected or complex terrain. r in coral marks poor hour-to-hour timing at that site. What do these numbers mean? →

Method: hourly AirNow observations are paired with the archived forecast value at the nearest grid cell for every cycle covering that hour (lead day 1 = same-day cycle, day 2–3 = earlier cycles). The scatter offers two views: "every hour" plots true hour × site day-1 pairs (archived for the trailing two weeks); "daily averages by site" plots site-day means for any window and lead (only sufficient statistics are archived long-term). Bias = forecast − observed. OTH = non-US or unrecognized site codes (e.g. BC).