Space weather nowcast: from L1 to the aurora

Why L1 gives lead time

The L1 Lagrange point sits about 1.5 million km from Earth toward the Sun. A monitor there samples the solar wind before it reaches the magnetosphere; at 300 to 800 km/s that head start is roughly 30 to 85 minutes. The RTSW files list several spacecraft (the feed labels them ACE, IMAP and SOLAR1) and flag one as active. Only active rows are used, and the labels are shown exactly as the feed writes them.

Joining the feeds

Plasma and magnetic field arrive in separate files. Rows are matched on the UTC minute (time tags are floored to the minute, because some sources stamp seconds). The join keeps a minute that has only one of the two, so each chart uses every sample it has, while every derived quantity needs both. A gap longer than 5 minutes breaks the plotted line, and the solar wind card reports coverage, the longest gap and any change of active spacecraft.

Ballistic propagation

x_GSE is the active spacecraft's distance sunward of Earth from the hourly ephemeris file, taken at the hour nearest each sample, and Vx is the measured GSE x velocity (the bulk speed if Vx is missing). This assumes the plasma travels straight down the Sun Earth line at its measured speed and that structures in it are flat sheets perpendicular to that line. Real phase fronts are tilted, and the spacecraft sits well off the line (its y and z offsets are in the ephemeris), so actual arrival can differ by several minutes either way. The delay is to Earth's centre; the bow shock is met a few minutes earlier. When fast wind overtakes slow wind the mapping stops being one to one, and the line marked "reaching Earth now" picks the sample whose arrival time is nearest to the present.

The drawing at the top

Every dot is a real 1 minute plasma sample from the last 2 hours, placed by its own progress under the same ballistic assumption. Its colour is that minute's Bz on a diverging scale (southward and northward, grey near zero, full colour at 6 nT along a (|Bz|/6)^0.7 curve), its size grows with the square root of density (a minute with plasma but no field sample is an empty ring), and its trail is the distance it covers in 4 minutes. Distances are not to scale. The Sun to L1 leg (computed from the Almanac's Sun distance minus x_GSE) is shortened, the L1 to Earth leg is stretched, and Earth's magnetosphere is drawn larger again. A dot moves linearly in distance from L1 until it meets the Shue magnetopause at its height, then slides along the boundary and fades by its ballistic arrival time. The magnetopause is drawn for the sample reaching Earth now. Dots are spread vertically by a fixed golden ratio sequence of their minute so they do not overlap; their height carries no data.

Dynamic pressure

This is the proton ram pressure, m_p n v^2 with the unit conversions folded into the constant. Alpha particles carry a few percent of the mass flux but their columns are empty in these files, so they are left out.

Newell coupling

Newell et al. (2007, J. Geophys. Res. 112, A01206) fitted this form against ten magnetospheric state variables. It is proportional to the rate at which magnetic flux is opened at the dayside magnetopause; with v in km/s and B in nT the value is in (km/s)^4/3 nT^2/3. It is zero for a due north field and largest for due south. Bt here is the field transverse to the Sun Earth line; the solar wind card plots the total field. The function is a predictor evaluated at L1, and it acts on Earth one propagation delay later.

Shue magnetopause

Shue et al. (1998, J. Geophys. Res. 103(A8), 17691) fitted this shape to magnetopause crossings including extreme solar wind conditions. r0 is the subsolar standoff distance and alpha sets how much the flanks flare. The model is symmetric about the Sun Earth line and ignores the tilt of Earth's dipole, so the drawing is a cross section through any plane containing that line. The tiles and the to scale card use the latest L1 sample, which reaches Earth one delay later. If r0 falls below 6.6 Re, geosynchronous satellites near local noon are outside the model magnetopause.

Kp and the G scale

Kp is a planetary index of geomagnetic disturbance over 3 hour windows, built from a network of ground magnetometers, on a quasi logarithmic scale from 0 to 9 in thirds (4.67 is written 5−). The forecast product labels each window observed, estimated or predicted, and the chart keeps those labels. NOAA's 1 min estimated Kp is a provisional nowcast, and NOAA rewrites its newest minutes after first publishing them (while this page was being built, minutes first published as 2.67 were rewritten to 1.33 a few minutes later). The page keeps the file's latest version of each minute and counts the revisions it sees. NOAA's G scale starts at Kp 5 (G1) and ends at Kp 9 (G5). Fractional values can be read two ways; this page rounds to the nearest whole Kp before mapping, because that reproduces the noaa_scale labels in NOAA's own forecast file (4.67 is labelled G1 there and 5.67 is labelled G2). A strict threshold, under which 4.67 would be below G1, is implemented and tested alongside it.

The comparison in that card is narrow on purpose. It sets NOAA's value for the current 3 hour window against the maximum and mean of the 1 min estimate so far inside it. These feeds carry no archive of past forecasts, so the page does not claim to score them.

GOES flare classes

The class comes from the 0.1 to 0.8 nm channel. The letter is the decade of the flux in W/m² (A from 1e-8, B from 1e-7, C from 1e-6, M from 1e-5, X from 1e-4) and the number is the flux divided by the start of that decade, truncated to one decimal, so 1.94e-6 is C1.9 and X continues past 10. The words small (A to C), medium (M) and large (X) are the usual plain descriptions of the classes. The 0.05 to 0.4 nm channel is plotted for context. The plotted flux is NOAA's corrected value; the file also flags minutes it considers affected by electron contamination.

OVATION aurora

SWPC's OVATION model, driven by solar wind measured at L1, gives the probability of visible aurora on a 1 degree grid. Each file states an observation time and a forecast time, and the gap between them is the lead time the model claims; the aurora card prints both as written. Overhead probability is the value of the nearest grid cell. The southernmost 10% latitude is found by walking north along the location's longitude from the equator. Aurora low on the horizon can be seen from farther south than either number suggests, and cloud, moonlight and city light are not in the model.

Day and night on the map

The subsolar point comes from the low precision solar coordinates in the Astronomical Almanac: mean longitude and mean anomaly give the Sun's ecliptic longitude, which with the obliquity gives declination and right ascension. The equation of time is mean longitude minus right ascension, and the subsolar longitude is where apparent solar time is noon. The Almanac quotes these formulas to about 0.01 degree for 1950 to 2050, and the tests check the equinox, both solstices and the equation of time extremes. The shaded hemisphere is everything more than 90 degrees from the subsolar point, the geometric terminator with no refraction and no twilight; aurora is usually seen only once the Sun is well below the horizon.

Freshness

Each feed has two clocks. The poll clock (header) says whether the last fetch succeeded. The sample clock (freshness card) is the age of the newest time stamp inside the file, judged against a per feed threshold: 1 minute products are fresh within 10 minutes and late within 30, the hourly ephemeris within 4 and 12 hours, the final 3 hour Kp within 4 and 7 hours of its window's end, OVATION within 20 and 60 minutes of its observation time, the forecast file within 3 and 12 hours of its Last-Modified header, and the scales file within 3 and 12 hours of the time stamp on its current entry.

Link to the Kp forecasting project

My multi horizon Kp forecasting project trained an LSTM on hourly NASA OMNI2 history. OMNI2 is a cleaned archive, already shifted from the spacecraft to Earth, merged across missions, with gaps marked. This page is the live input such a model would need in production, with everything the archive smooths over: plasma minutes that go missing, a spacecraft label that can change mid stream, ephemerides that lag the data, and propagation that has to be done before the inputs match the training distribution. That model's strongest predictors were ap and the auroral electrojet indices, which are not in these feeds, so a production version would need a source for them or a retrain on solar wind inputs alone.