The science & math behind CME Tracker

How the tracker turns NASA's CME measurements into a cone, an arrival time and a storm estimate: the equations, the tests, and the known weak points.

1The pipeline

• Measure: NASA analysts fit each CME's speed, direction and width from coronagraph images (§2).
• Aim: a cone test decides whether it covers Earth, Mercury, Venus or Mars (§3).
• Propagate: the drag-based model carries the front through the solar wind (§4).
• Solve: a root-finder meets that trajectory with the planet's true, moving position (§5, §6).
• Estimate: the arrival speed maps to an expected Kp and NOAA G-level (§7).

Nothing is tuned per event. §8 tests the chain against ten years of arrivals.

2The measurements

A CME heading straight at the observer shows almost no sideways motion, so one viewpoint is not enough. NASA's DONKI analysts fit a cone to coronagraph images from two angles when they can: L1 (SOHO/LASCO) and STEREO-A, which orbits off the Sun-Earth line. When STEREO-A's images are missing, or it sits too close to that line to add a second angle (as around its August 2023 pass of Earth), the fit rests on one view and is weaker. Each analysis gives:

• Speed: radial speed of the leading edge, in km/s.
• Direction: latitude and longitude of the cone axis in Stonyhurst coordinates (0, 0 is the point on the Sun's equator facing Earth).
• Half-angle: half the cone's angular width, typically 20 to 45°.
• time21_5: when the leading edge crosses 21.5 solar radii (about 0.1 AU). The model starts there.

How much do these inputs move? The catalog gives no error bars, but NASA measured 682 of the 6,237 CMEs of 2014 to 2024 more than once, and the change from first measurement to final is a fair guide. Speed changed by a median of 14% (44% or more for one in ten). Direction moved 9° (one in ten: 36° or more). Half-angle changed 3° (one in ten: 12°). For 11% the revision changed whether our cone test called a hit or a miss. CMEs get measured again when they matter or when the first fit was in doubt, so these are the harder cases, and also the ones people watch.

A CME with no analysis (too faint, far-sided, or not measured yet) is drawn as a directionless shell and gets no arrival (§14).

3Will it hit? Cone geometry

A CME hits a planet when the angle between the cone's axis and the Sun-planet line is at most the half-angle. That angle is a great-circle separation, not √(lon² + lat²), which overstates the offset at high latitudes:

For planets other than Earth:

• The axis is frozen at launch. Stonyhurst longitude is relative to Earth at launch, so it is converted once to a fixed heliocentric direction (Earth's longitude at launch plus the CME's longitude). The cone does not co-rotate with Earth in flight.
• The planet moves during transit. Mercury covers about 4° a day, 16° over a four-day transit. The test uses the planet's position when the front reaches its orbit, not at launch.

4The drag-based model (DBM)

Beyond 21.5 R☉, a CME faster than the ambient solar wind is slowed by it and a slower one is pulled along (Vršnak et al. 2013):

v is the front's speed, w the ambient wind speed and γ the drag parameter. It integrates in closed form, so there is no numerical integration:

• Ambient wind w: the solar wind speed measured at L1 when the model runs, clamped to 280 to 750 km/s (430 km/s if unavailable).
• Drag parameter γ: banded by launch speed, lower for faster CMEs. The two fast bands sit below Vršnak's published floor of 0.2 × 10⁻⁷ km⁻¹ because the 282-event back-test favours them, so that test is not an independent check of them (§8):

  Launch speed	γ (km⁻¹)
  ≥ 1500 km/s	0.10 × 10⁻⁷
  1000 to 1499	0.15 × 10⁻⁷
  600 to 999	0.20 × 10⁻⁷
  < 600	0.30 × 10⁻⁷

Fast CMEs arrive much slower than they launch. In this model a 2,000 km/s CME reaches Earth at about 1,000 km/s, and a 1,000 km/s CME in a 430 km/s wind takes about two days. The model brakes the very fastest too hard: the August 1972 storm crossed in under 15 hours, which no setting of this drag reproduces. The arrival speed, not the launch speed, drives the storm estimate in §7.

5Solving the arrival time

Setting r(t) equal to Earth's distance has no algebraic solution, and that distance varies: Earth is about 3.3% closer to the Sun in January than in July, which moves a two-day arrival by about two hours. The tracker solves numerically against the true distance:

r(t) only increases, so bisection always converges. Other planets use their own varying distance (Mercury's runs from 0.31 to 0.47 AU, so a fixed radius would misplace arrivals by hours).

The ±12 h shown with each arrival is the median error from §8. It is widened to ±15 h for CMEs under 500 km/s, which are measured less well, and to ±18 h away from Earth, where nothing has been validated.

6Where the planets are

Planet positions come from classical orbital elements at the displayed time. Orbits are ellipses, so mean motion is corrected with the equation of center, to third order in eccentricity e:

This matters most for Mercury (e = 0.206), where a circular orbit would be wrong by up to 24° of longitude and 21% of distance, enough to flip a hit into a miss. With the correction, positions are good to about 0.1°. Earth uses the Almanac series, and the two agree to under half a degree.

STEREO-A is near-circular, so mean motion pinned to its August 2023 Earth flyby is good to a few degrees (labelled “approx” on the map). Parker Solar Probe (e = 0.88) and Solar Orbiter use a Newton-Raphson solve of Kepler's equation M = E − e sin E on JPL HORIZONS elements. Solar Orbiter's orbit changes at every Venus flyby, so its elements describe only the current leg and are refreshed after each one. The map projects its inclined orbit onto the ecliptic and prints its latitude on the label.

7Estimating storm strength

Storm strength depends on how hard the CME hits (speed and density) and on which way its magnetic field points on arrival. The first is known days ahead. The second cannot yet be measured from a distance, and no operational forecast has it until the cloud reaches L1, 30 to 60 minutes out. So the launch-time estimate is a statistical fit to observed storms (tested in §8), shown as a range:

The slope is shallow: a 1,000 km/s arrival lifts the median only to about Kp 5.2. Launch-time observables barely constrain storm size. So every alert and card shows a likely value and a ceiling together (for example "Kp ~4.8, up to 6.8"), detection triggers use the ceiling, and the single-level call waits for the L1 watch.

8Validation

How the arrival model is back-tested:

• Ground truth: NASA DONKI's interplanetary-shock catalog, instrument-timed arrivals at L1 from 2014 to 2024.
• Associations: NASA's own shock-to-CME links, not ours, so we cannot choose which CME "caused" which shock.
• Ambient wind: hourly OMNI measurements for each event.
• Code: the production functions, the same ones that draw the map.

Across 282 NASA-linked CME-to-shock pairs:

An earlier test on 11 famous storms scored 8.3 h. That was a selection effect, and the 282-pair figures replaced it.

The same test, made harder

That headline is generous to the model in three ways. It uses solar wind measured 12 to 30 hours before the shock arrived, which a forecast would not have. It drops pairs that miss by more than 48 hours. And it uses NASA's final measurement of each CME, often revised after the event. One step at a time:

The hindsight wind did not help. Keeping every pair and using the measurement a forecaster would have had costs about two hours. The last row is the fair description of the drag model as a forecast: a mean miss of 16.6 hours, a median of 12.6, and a worst case of nearly four days. The "±12 h" on every arrival is the median miss, so about half of real arrivals fall outside it.

WSA-Enlil, like for like. NASA ran Enlil for 264 of these pairs. Its first real-time run missed by 11.7 h on average (median 8.7 h) and ran 3.6 h early on average. The drag model on the same 264 pairs, with NASA's first measurement, missed by 15.8 h (median 12.1 h). Enlil is about four hours better. A published verification of Enlil's real-time runs from 2010 to 2016 found a mean miss of 10.4 h (Wold et al. 2018).

Not an out-of-sample test. The drag values in §4 were chosen with this record in view. Split in two on the third row's test, the mean miss is 16.5 h over 2014 to 2020 (100 pairs) and 15 h over 2021 to 2024 (192 pairs).

What this test cannot show. Every pair is a CME that arrived, with a shock, and that NASA's forecasters linked to that shock. They made those links with Enlil's forecast in front of them, which may favour CMEs that arrived near the predicted time. A CME that arrived without a shock, or was never linked, is not in the test. Whether a CME arrives at all is the next table.

Does the verdict hold? Hits, misses and false alarms

We took every CME from 2014 to 2024 with a measured direction (5,419 of them) and compared the cone test of §3 with what happened. A CME counts as arrived when NASA linked it to a shock at Earth 8 hours to 6 days later. 364 did.

A DIRECT call is right about six times in ten and catches 58% of what arrives. Count GLANCING as a yes and it catches nine in ten, but most of its yes calls come to nothing. On the CMEs both were run on, the cone test and Enlil sort hits from misses about equally well. Speed matters: a DIRECT call is right 43% of the time for a CME under 500 km/s, 74% from 500 to 999 km/s, and 83% at 1,000 km/s and over.

Two cautions. The false alarms are overstated: a slow CME can arrive with no shock and never enter NASA's shock list, and one shock is often linked to several CMEs. And 818 more CMEs had no measured direction and get no verdict from us; 3 of those arrived.

Storm strength

The Kp fit in §7 was calibrated on 221 CME-to-shock-to-Kp events from 2014 to 2024, with GFZ-measured Kp as truth and 53 shocks left out because a second shock fell in the same window. The displayed band is the measured spread of that fit's errors.

It shows no skill out of sample. Fitted on 2014 to 2020 only (74 events) and scored on 2021 to 2024 (147), the same formula lands within one G-level 84% of the time, with a mean error of 1.2 Kp. Saying "Kp 4.3" every time, the median of the training years, scores the same 84% and 1.2. The training years alone give no dependence on speed at all. The slope in §7 appears only when all eleven years are fitted together. At launch the estimate says what CME storms are usually like and almost nothing about this one.

The headline table and histogram render from data/model-skill.json, written by scripts/validate-model.mjs. The stricter and verdict tables come from scripts/validate-strict.mjs (data/strict-skill.json), and the storm-strength test from scripts/calibrate-kp.mjs (data/kp-skill.json). Both were run on 2026-10-06.

The live scoreboard

The back-test is history. This is the running record: every arrival window the site has issued since the ensemble went live in September 2026, scored against the solar wind that reached L1. Misses stay on it.

Loading the scoreboard…

How it is scored. The unit is the forecast, not the arrival.

• It must be a forecast. A window counts only if it was issued before it opened. The four made on the feature's first night, after their CMEs had arrived, are drawn in grey and never scored.
• Scored 60 hours after its window closes: a day's allowance either side, plus the day and a half the detector needs to tell a CME's body from a coronal-hole stream.
• Caught means a front fell inside the 80% window. Near means the closest front was within a day outside it. Nothing arrived is counted too.
• The error is arrival minus forecast median, for caught and near alike, so a miss just outside the window still raises the average.
• Fronts are detected once, over the whole record, from our own 10-minute L1 archive: shocks (a 30 km/s jump with density and field both rising) and gradual arrivals (80 km/s inside six hours that falls back within a day and a half). In a 125-day check the shock detector found 12 of the 23 CME-linked shocks NASA listed, the clear ones, and four in five of its detections matched a listed shock.
• The drag model is scored on the same forecasts, run with the wind measured in the six hours before launch.

What this cannot show yet. The sample is small. A window is about 40 hours wide, and in an active week a front reaches L1 every day or two, so catching one is not yet evidence of skill. Several CMEs often share one arrival, and nobody can say which of them the front was. Testing the 80% claim needs several dozen forecasts, including quiet stretches. Until then this is a record, not a result.

9The L1 nowcast

Spacecraft at the Sun-Earth L1 point sit about 1.5 million km upstream and sample the wind that reaches Earth minutes later:

The tracker polls the feed every 90 seconds and runs a detector on it:

• Shock: median wind speed in ~15-minute windows before and after each point. A jump above 45 km/s with a magnetic-field jump (or above 80 km/s alone) flags a shock and starts a countdown to Earth. Medians, so one bad telemetry point cannot fake a shock.
• Southward field: a 30-minute median Bz at or below −8 nT, with at least four in five of the readings in that half hour at or below −8. One southward spike does not count.

This is a measurement with a travel time, not a model, and it is the most reliable product on the site. The same detector runs on the server every two minutes for subscriber alerts: when the motional electric field (v·B_south, 4 mV/m or more here; Gonzalez & Tsurutani put intense storms at about 5) reaches that level with the field held south as above, a single-level "conditions arriving now" alert goes out.

The alert rule was back-tested on hourly L1 data from 2015 on. Hourly averages cannot test the half-hour condition, so this checks the thresholds, not the timing. It would have fired 210 times, about 18 a year. A G2 storm or stronger followed within six hours 80% of the time, and at least a G1 97% of the time. It is a late warning by design, with a median lead of half an hour, and it catches only 31% of G2 episodes: many storms build without the field ever holding that far south for half an hour.

10The Earth globe

The globe is drawn from the Sun's point of view, so the face you see is the one an arriving CME strikes.

• Sunlit face: the sub-solar point. Latitude is the solar declination, δ = arcsin(sin ε sin λ) with obliquity ε = 23.44°. Longitude comes from UTC corrected by the equation of time (up to ±16 minutes, or 4° of longitude).
• Compression of the magnetosphere: solar-wind dynamic pressure:
  P_dyn = 1.6726×10⁻⁶ · n · v²  nPa (n in protons/cm³, v in km/s)
• Aurora: live, the globe paints NOAA's OVATION map (§13). In replays, forecasts and data gaps the ovals are sized by Kp: each step moves the boundary a few degrees toward the equator.

Inside measured data (live, recent past or a replay) the globe is driven by measurements. For a CME that has not arrived it uses a synthetic profile: a dynamic-pressure spike at the shock that decays over about an hour, a dense sheath, and a southward-field pulse scaled to the Kp ceiling, all relaxing over 18 hours. The caption says which mode is showing.

11The Parker spiral & radiation storms

The Sun rotates (sidereal period 25.38 days) while the wind carries its magnetic field radially outward, which winds the field into a spiral. A field line's longitude lags by:

Solar energetic particles follow field lines, so an eruption within roughly ±45° of the footpoint can dose Earth in minutes to hours wherever the CME itself goes. That is why the radiation card can light up for a CME that misses.

The connectivity overlay draws the Earth-connected field line from the wind speed measured at the displayed moment. The footpoint sits near W35 at 700 km/s and W70 at 350 km/s, with the ±45° band shaded round it, and each selected CME's source is flagged connected or not by that geometry. A measured proton event outranks the geometry. When the CME's linked SEP begins within −3 to +24 hours of launch, the map shows the observed radiation with its real arrival lag even if the source is outside the band, since wide shock fronts can put particles onto Earth-connected lines from well beyond it. A catalogued proton event outside that time gate is not attributed to the eruption.

12HF radio & the ionosphere

The radio conditions page is built from numbers the site already tracks.

• F10.7 and the sunspot number. The F-layer (about 150 to 400 km) does the long-haul reflecting and is ionized by solar EUV, for which the 10.7 cm flux is the standard proxy. More flux means higher usable frequencies: at SFI 150+ the 10 m band carries worldwide, and at SFI ~70 it is dead.
• Flares. A flare's X-rays over-ionize the D-layer (60 to 90 km), which absorbs HF. The sunlit side loses the lower bands within minutes and recovers within the hour after the flare fades. This is NOAA's R scale, graded from the GOES X-ray flux.
• Geomagnetic storms. High Kp and A-index mean a depleted mid-latitude F-layer, auroral absorption and closed trans-polar paths, typically for 24 to 48 hours after arrival. The aurora itself can reflect VHF and open 6 m and 2 m paths.

The band table on /radio is the standard operator heuristic: a base rating per band group from SFI thresholds (30 to 20 m opens at about SFI 90, 17 to 15 m at 115, 12 to 10 m at 150; the low bands depend more on darkness than flux), minus a geomagnetic penalty (one step when Kp ≥ 4 or A ≥ 20, two when Kp ≥ 6 or A ≥ 30). It is not an ionosonde measurement or a ray-trace, and it knows nothing of season, path or antenna.

13The other models on the site

Each is listed with what it was tested against, or a statement that it has not been tested.

The arrival window

For each Earth-directed CME the server runs the fluid model below 200 times, each with the inputs nudged at random: speed by 15%, direction and width by 8°, launch time by 45 minutes, background wind by 8% (one standard deviation each). The card shows the 10th to 90th percentile of the arrival times.

Input error alone is not enough. On the back-test storms, a window built only from those nudges held the real arrival 44% of the time, where it should hold 80%. So each run's arrival also gets random timing error with a 14 hour standard deviation, the amount that brings the window to 79% on that record, at a mean width of 38 hours. That calibration used 60 runs per CME, and the 214 storms for which the model produced a window. The width is fitted to past misses, not derived from the physics.

The card also shows the share of runs in which the cloud crosses Earth. That is not a probability of impact. It cannot be calibrated against a list that only holds CMEs that arrived, and 42 of the 282 real arrivals in the back-test were given under 10%.

The fluid wind model

The Fluid layer is a reduced-physics solar wind model in the style of HUXt (Owens et al. 2020), solved in the browser. Each radial line in the ecliptic plane is solved on its own as a gas that conserves mass and momentum, with a polytropic pressure. The inner boundary at 21.5 R☉ is built from our archive of wind measured at L1 over the previous solar rotation, mapped back to the Sun. Each CME goes in as a cone-shaped pulse of faster, denser wind.

On the same back-test storms, with the boundary built only from wind measured before each CME left, its mean miss was 11.8 hours (median 9.2 h, bias −0.8 h), against 13.6 hours for the drag model on the same storms. That covers 171 of the 282. The model works only in the ecliptic plane, so it has nothing to say about the other 111, where the cone as measured never crosses Earth in that plane. It has no magnetic field, so it says nothing about storm strength, and its densities are right in pattern more than in number.

The magnetosphere

The magnetopause is the empirical surface of Shue et al. (1998), set by the solar wind's dynamic pressure and Bz at the displayed moment:

The bow shock sits ahead of it by the Farris and Russell (1994) stand-off ratio, which depends on the wind's Mach number. Both are average shapes fitted to many spacecraft crossings. They are symmetric about the Sun-Earth line, with no cusps, no dipole tilt and no dawn or dusk difference, and the Mach number assumes a typical 5 nT of field beside Bz. When the nose falls inside 6.6 Earth radii the page says geosynchronous satellites are outside the shield. That is the model's statement, not a measured crossing. The field lines are a dipole clipped to this surface, drawn for shape.

The radiation belts on that view are an illustration, not a radiation model. One measurement sets them: the flux of electrons above 2 MeV at geosynchronous orbit from GOES. The outer belt's brightness follows that reading and its outer edge is cut where the magnetopause has moved in. The inner belt is drawn the same at all times.

Satellite drag

The Atmosphere view uses a small model of the upper air, not NRLMSIS. An exospheric temperature is set by the Sun's F10.7 radio flux and by Kp, with heating that fills in over about six hours and cooling that takes about two days:

The constants were fitted to measured decay of the ISS, Tiangong and Hubble since 2015, taken from their published orbits: 7,116 satellite-days as of October 2026. On quiet days the model has no overall bias (median ratio 1.02), but single days scatter widely. Half fall between 0.74 and 1.41 times the measured loss, and 40% are within 25%. Some of that is the measurement, since a day's loss is a few tens of metres. For the ISS in storms it gives 3.2 times the quiet rate where 3.3 was measured. These scores are on the same days the constants were fitted to (scripts/score-drag.mjs, data/drag-skill.json). It has no local time, latitude or composition, and a satellite that changes its attitude or fires its engines breaks the comparison for that day.

Fast streams and their return

A stream is found in our L1 archive when the hourly wind rises to 500 km/s or more, at least 100 km/s up within a day, and holds a 12 hour mean of 480 or more. It ends when the wind stays under 450 for 12 hours. A stream that begins beside a fast Earth-directed CME's arrival is flagged, because a CME's wake also raises the wind. The "due back" forecast is the same stream one solar rotation later, 27.27 days as seen from Earth. The return forecast has not been scored yet. Coronal holes grow, shrink and close, so a stream may come back early, late, weaker or not at all.

Radio bursts and provisional CMEs

A Type II radio burst marks a shock leaving the Sun, and it often reaches the catalog before the CME does. Across 486 Type II bursts reported by ground stations from 2015 to 2026, 87% had a catalogued CME between 30 minutes before and three hours after, and the burst came before the fastest of them by a median of 27 minutes. So when one fires with no CME listed yet, the map draws a provisional, dashed cone from the flare or region it came from.

Textbooks derive a shock speed from how fast the burst drifts down in frequency. We tested that against the 393 bursts with a usable drift and a measured CME and found no skill: the correlation is 0.02 (data/type2-skill.json). The provisional cone therefore uses a fixed typical speed and half-angle (670 km/s, 36°) and is dropped when a measured CME replaces it. The reverse is weak too: only 17% of CMEs faster than 800 km/s came with a reported Type II.

The aurora maps

The oval on the aurora page is NOAA's OVATION nowcast, a model driven by the wind measured at L1 that gives the chance of aurora overhead for the next half hour to hour. It is drawn as NOAA publishes it. The Earth globe uses the same map when it is live. In replays, forecasts and data gaps its oval is sized from Kp alone, which is a rough rule and says nothing about where along the oval the display is bright.

The oval's edge is where aurora may be overhead. A display can be seen low on the horizon from farther away, and the page draws that line about 5° of latitude beyond the edge. That margin is a rule of thumb. The site knows nothing of your clouds, moonlight or city lights, and it has never been checked against what people saw.

14Rules the map follows

• No invented aim. A CME with no measured direction is a dashed, directionless shell. It never gets a cone, an arrival, a banner or a storm on the globe.
• Linear scale, true positions. Distance to pixels is linear. Planets sit at their true positions on eccentric orbits and Earth at its real seasonal distance. One exception: the Sun is never drawn under 9 pixels (12 in the cinematic view), because at true scale it would vanish.
• Wedges match the math. The drawn cone uses the same launch-frozen axis, half-angle and drag-model radius as the hit test and the arrival solve.
• Time-aware instruments. The spacecraft layer shows only craft that existed at the displayed time. A 2012 replay has no DSCOVR, and the 1859 Carrington replay has none at all.
• Provenance labels. Replays separate measured archive data from reconstruction, arrivals away from Earth are tagged as unvalidated estimates, and the model-skill card shows the whole error distribution.

15Known limitations

The rule: an approximation is acceptable when its error is small against the roughly ±12 h and ±20° already in the inputs.

16References

• Vršnak, B., et al. (2013). Propagation of Interplanetary Coronal Mass Ejections: The Drag-Based Model. Solar Physics 285, 295–315. The propagation model.
• NASA CCMC, DONKI. CME analyses, flares, SEPs, and the interplanetary-shock catalog used for validation.
• NOAA SWPC, Space Weather Scales. The G and S scales.
• The Astronomical Almanac. Low-precision solar coordinates, Earth–Sun distance, and equation of time.
• Standish, E.M. (JPL). Mean orbital elements for the inner planets; JPL HORIZONS for Parker Solar Probe.
• Parker, E.N. (1958). Dynamics of the Interplanetary Gas and Magnetic Fields. The spiral.
• Davies, K. (1990). Ionospheric Radio. The standard reference for HF propagation physics; NOAA Space Weather Scales (R/G/S definitions).
• Millward, G., et al. (2013). An operational software tool for the analysis of coronagraph images: Determining CME parameters for input into the WSA-Enlil heliospheric model. Space Weather 11, 57–68. How the cone parameters in DONKI are measured.
• Odstrcil, D. (2003). Modeling 3-D solar wind structure. Advances in Space Research 32, 497–506. The Enlil model.
• Wold, A. M., et al. (2018). Verification of real-time WSA-ENLIL+Cone simulations of CME arrival-time at the CCMC from 2010 to 2016. Journal of Space Weather and Space Climate 8, A17. The published Enlil score quoted in §8.
• Riley, P., et al. (2018). Forecasting the arrival time of coronal mass ejections: Analysis of the CCMC CME Scoreboard. Space Weather 16, 1245–1260. How arrival forecasts from many models compare.
• Dumbović, M., et al. (2018). The Drag-based Ensemble Model (DBEM) for coronal mass ejection propagation. The Astrophysical Journal 854, 180. The drag model run as an ensemble.
• Owens, M. J., et al. (2020). A computationally efficient, time-dependent model of the solar wind for use as a surrogate to three-dimensional numerical magnetohydrodynamic simulations. Solar Physics 295, 43. HUXt, the model the fluid layer follows.
• Shue, J.-H., et al. (1998). Magnetopause location under extreme solar wind conditions. Journal of Geophysical Research 103, 17691–17700. The magnetopause shape.
• Farris, M. H., & Russell, C. T. (1994). Determining the standoff distance of the bow shock: Mach number dependence and use of models. Journal of Geophysical Research 99, 17681–17689. The bow shock distance.
• Gonzalez, W. D., & Tsurutani, B. T. (1987). Criteria of interplanetary parameters causing intense magnetic storms (Dst < −100 nT). Planetary and Space Science 35, 1101–1109. The southward-field criterion behind the L1 alert.
• Newell, P. T., et al. (2009). Diffuse, monoenergetic, and broadband aurora: The global precipitation budget. Journal of Geophysical Research 114, A09207. The basis of NOAA's OVATION aurora model.
• Newkirk, G. (1961). The solar corona in active regions and the thermal origin of the slowly varying component of solar radio radiation. The Astrophysical Journal 133, 983. The coronal density model behind the Type II drift test.

Related: Where the Sun shoots, the catalog's ten years of launch points on one unrolled Sun.

← Back to the tracker   ? Help & guide