Products and models / Kp persistence forecast (baseline)

Kp persistence forecast (baseline)

live experimental forecast

Kp in 3 h2.3

Performance

Verified against “Planetary Kp index” (NOAA SWPC ), near-real-time and as first stored: later revisions by the provider are not seen. Skill = 1 − MSE / persistence MSE; 0 = same as persistence, positive = better. · model version 0.1.0

Lead time	Window	n	MAE	RMSE	Bias	Coverage / nominal
3 h	7 days	96	0.78	0.97	−0.11	88 % / 80 %
3 h	30 days	96	0.78	0.97	−0.11	88 % / 80 %
3 h	90 days	96	0.78	0.97	−0.11	88 % / 80 %
6 h	7 days	93	0.90	1.14	−0.16	90 % / 80 %
6 h	30 days	93	0.90	1.14	−0.16	90 % / 80 %
6 h	90 days	93	0.90	1.14	−0.16	90 % / 80 %
12 h	7 days	87	0.95	1.20	−0.18	89 % / 80 %
12 h	30 days	87	0.95	1.20	−0.18	89 % / 80 %
12 h	90 days	87	0.95	1.20	−0.18	89 % / 80 %
24 h	7 days	75	1.44	1.65	−0.51	64 % / 80 %
24 h	30 days	75	1.44	1.65	−0.51	64 % / 80 %
24 h	90 days	75	1.44	1.65	−0.51	64 % / 80 %

What this shows

Reference forecast: it assumes Kp over the next 3 to 24 hours will equal the last complete Kp interval. It isn't a physical model but the yardstick every model is measured against: a forecast that can't beat persistence adds no information. The band covers the central 80% of persistence errors over the last 30 days.

Model card: docs/products/usach.kp.persistence.md

Product fact sheet

Forecast variable: Planetary Kp index
Source: NOAA SWPC near-real-time Kp, as stored
Original cadence: One issue per hour; 3, 6, 12 and 24 h horizons
Licence: CC BY 4.0
Used by: The skill reference for every Kp model on the portal
Download: Latest issue: JSON · CSV · more