Space Weather Operational
Readiness Development
Track 1
U-Michigan

F10.7 tracks the solar ultraviolet output that heats the upper atmosphere, so it sets how fast low satellites lose altitude to drag. The forecast uses one gradient-boosted tree model per lead day, fed with the last 60 days of F10.7, sunspot number, NOAA SWPC solar region reports, and far-side active-region detections from GONG and SDO/HMI. On 2024–2026 test data its error is 11% lower than the SWPC 27-day outlook.

↔ Drag to pan
Loading…
Black: observed daily F10.7 from the Penticton observatory. Red: the newest forecast, issued once a day from the last full day of data. The shaded bands are the ranges that held the observed value 50% and 90% of the time, refitted for each forecast on the model’s earlier forecast errors at the same level of solar activity. Show forecast from draws the forecast the same model gives from any day since January 2024, a period it was never trained or tuned on, so it can be checked against what happened. These past forecasts use today’s data archive, not the data available on that day. The other lines are the other groups’ newest F10.7 forecasts, each labelled with the day it was issued: from NOAA SWPC, the 45-day forecast, the 27-day outlook (issued on Mondays) and the 3-day forecast (from the daily report of solar and geophysical activity); the 60-day forecast from NJIT; the 7-day forecast from ADAPT-SIFT; SWORD Track 3’s 7-day forecast (converted from 1 AU to Earth’s distance from the Sun); and ESA’s short-term forecast, from ESA’s Space Debris Office. For a past date each is the issue that was out on that date; most have been saved only since early or mid-October 2026. A forecast whose newest issue is more than a few days old is left off, and SWORD Track 3’s only when it has one for the day. Click a name in the legend to hide or show its line. Drag the plot to pan back through the last six months (back to late 2023 for a past date); double-click snaps back. 1 sfu = 10−22 W m−2 Hz−1.

Ap is a daily index of geomagnetic activity from a worldwide network of magnetic observatories. In a geomagnetic storm the upper atmosphere heats and expands, and drag rises. The forecast uses one gradient-boosted tree model per lead day, fed with the last 60 days of 3-hourly ap, sunspot number and F10.7 from GFZ Potsdam. On 2024–2026 test data its error for the next day is 18% lower than always forecasting the long-term average. From the second day on it is only 1–4% lower: nothing in its inputs says when the next storm will arrive, so it does not forecast storms more than a day ahead.

↔ Drag to pan
Loading…
Black: observed daily Ap from GFZ Potsdam (the most recent weeks are GFZ’s preliminary values). Red: the newest forecast, issued once a day from the last full day of data. The red line is the expected Ap; because a few storm days pull that average up, a typical day falls below it. The shaded bands are the ranges that held the observed value 50% and 90% of the time, refitted for each forecast on the forecast errors of the previous 11 years. About 70% of storm days (Ap 30 or more) fall above the 90% band. Show forecast from draws the forecast the same model gives from any day since January 2024, a period it was never trained or tuned on, so it can be checked against what happened. These past forecasts use today’s data archive, not the data available on that day. The other lines are the other groups’ newest Ap forecasts, each labelled with the day it was issued: the NOAA SWPC 27-day outlook (issued on Mondays), SWORD Track 3’s 3-day forecast and ESA’s short-term forecast. For a past date each is the issue that was out on that date; all but SWPC’s have been saved only since mid-October 2026. A forecast whose newest issue is more than a few days old is left off, and SWORD Track 3’s only when it has one for the day. Click a name in the legend to hide or show its line. Drag the plot to pan back through the last six months (back to late 2023 for a past date); double-click snaps back. Ap is in units of 2 nT.