Model check

Is the activity model honest? Calibration skill, reliability and a live hindcast against the observed indices. Back to the dashboard

Calibration on two years of history

Reliability: how often Hp30 actually reached the threshold, by predicted value
Residuals of the fit on the test period

Forecast verification from the Worker's log

Tier probabilities issued every 5 minutes at +10, +30 and +60 min, scored against what followed.

Reliability of the naked-eye (dark site) tier at +30 min: forecast probability versus how often the truth followed

Live hindcast, last 3 days

Kp modeled from NOAA's propagated solar wind versus what was observed.

Modeled Kp against observed Hp30 (GFZ, via iSWA mirror) and NOAA estimated Kp

CME arrival predictions: historical error

From NASA CCMC's CME Scoreboard (observed arrivals since 2023).