So far, we have discussed that the Starlink satellite trajectory during a solar storm gives insight into the Starlink fleet operations. This does not shed light on the geospatial aspect of satellite trajectory change. This aspect is explored in a recent work (Kang et al., 2025 ) investigating the LEO satellite network during solar storms, which identified a ‘W’ shaped altitude change pattern across the orbits of the Starlink constellation. Their work does not explain the reason behind this pattern. In this work, using CosmicDancePro, we dive deeper into this mysterious geospatial pattern to find the root cause of this.
Authors in (Kang et al., 2025 ) have demonstrated systematic differences in altitude decay responses in satellites belonging to different orbits during the May 2024 superstorm. Their analysis shows that satellites in an orbit position inline between the Sun and Earth experience significantly larger absolute altitude changes over one day. Whereas satellites positioned approximately orthogonal to this direction (around ± 90 ∘ \pm 90^{\circ} phase offset) exhibit comparatively stable behavior. When all the orbital planes are aggregated in order, these variations form a ‘W’-shaped profile in absolute altitude change. They attribute this pattern primarily to differences in solar radiation exposure across the orbital planes, suggesting that satellites closer to the Sun experience stronger perturbations, while those farther away are less affected (Kang et al., 2025 ) . Note that, unlike high altitude GEO satellites, LEO satellites operate below the inner Van Allen belt and are largely shielded from direct extreme radiation exposure except for two anomalous regions: the South Atlantic Anomaly and the polar region. As we have already demonstrated in our prior discussion, the dominant driver of orbital decay in LEO is atmospheric drag, which is modulated by thermospheric density variations rather than direct solar proximity. The average distance between the Sun and Earth is approximately 150 million km. A LEO satellite at 550 km altitude while orbiting Earth will have distance variation with the Sun up to 0.009% of 150 million km. Which we believe is insufficient to produce meaningful differences in radiation exposure. This highlights a gap in understanding the key underlying mechanism behind the observed pattern. Specifically, we find that the following four question is still remain to be answered:
6.4.2. Decoding the root cause behind ‘W’ pattern:
To investigate the root cause of the observed ‘W’-shaped pattern, we perform a year-long analysis across orbital planes. For each satellite within a shell, we collect all available TLEs for a given day and compute the daily altitude change as the difference between the maximum and minimum semi-major axes reported in those TLEs. This captures a satellite’s net altitude change within a day. We then propagate satellite positions (latitude, longitude, elevation) using TLEs at 1-second granularity to obtain fine-grained orbital trajectories. Using these positions, we compute (i) the duration of solar exposure and (ii) the atmospheric density encountered along the orbit around the Earth. We use the atmospheric condition datasets to obtain density by mapping satellite positions to the nearest grid point. Since density measurements are available at a 20-minute cadence, satellite positions are temporally aligned to the closest available snapshot.
The pattern exists throughout the year - In Fig. 8, we summarize these measurements in one day per month for a one-year period. For May and October, we show the peak solar storm day (i.e., the 11th day of the month), while for all other months we show the first day. Notice that the x-axis in these figures represents the RAAN, ordering orbital planes. RAAN is the angular term that denotes where a satellite intersects the Earth’s equator while moving towards the geographic north. Therefore, all satellites in the same orbit will have the same or very similar RAAN values. Using this, we plot the left y-axis, which shows the distribution of daily altitude change experienced by the satellite across 72 orbital planes. On the right side, using a blue y-axis, we show the maximum difference in atmospheric density encountered by the satellites while orbiting the Earth in that orbital plane. Using another red y-axis beside that, the median of sunlight exposure duration on a day in that orbital plane. From this summary of one-year measurements, we make the following observations:
(a) 1st January, 2024 (b) 1st February, 2024 (c) 1st March, 2024 (d) 1st April, 2024 (e) 11th May, 2024 (Solar superstorm) (f) 1st June, 2024 (g) 1st July, 2024 (h) 1st August, 2024 (i) 1st September, 2024 (j) 11th October, 2024 (Solarstorm) (k) 1st November, 2024 (l) 1st December, 2024 Figure 8 . The distribution of Starlink satellite orbital altitude changes (left y-axis), along with the maximum atmospheric density change encountered (right y-axis, blue) and median sunlight exposure (right y-axis, red), follows a similar pattern throughout the year. This pattern gradually shifts to the right, with transitions in March–April and September–October. During periods of solar storms in May and October, the magnitude of altitude and atmospheric density variation is significantly higher than under baseline conditions.
Table 5 . Range of RAAN degrees where significantly high and low altitude variations are observed per day over the span of one year. Month High variation RAANs Low variation RAANs Range 1 Range 2 Range 1 Range 2 January 240–330 30–90 150–200 350–0 February 300–360 120–150 60–90 210–240 March 330–360 180–210 120–150 250–270 April 210–240 0–30 90–150 290–310 May 180–270 0–60 120–150 330–360 June 270–300 30–60 120–180 330–0 July 290–330 60–90 180–240 0–30 August 90–150 310–340 210–260 0–60 September 120–150 350–30 60–90 240–270 October 150–210 30–60 90–120 360–30 November 170–240 0–30 60–90 240–270 December 240–270 30–90 180–210 360–20
(1) This ‘W’-shaped pattern in altitude variation exists throughout the year, including days without any significant solar activity, observes the other months in Fig. 8 except May and October. We summarized RAAN degrees of high and low variation orbital planes in Table 5. Also, notice that the magnitude of change increases significantly during solar storms, reaching up to 700 meters per day in May and October, while in non-storm periods, it is typically within 150 meters per day, and still exhibits a ‘W’ structure. This indicates that the pattern is not solely induced by extreme solar events, but reflects an underlying systematic phenomenon. (2) Altitude variation across orbital planes strongly correlates with the atmospheric density difference encountered along the orbit around the Earth. Orbital planes with high altitude variance show a larger difference in atmospheric density. In contrast, sunlight exposure duration shows an inverse relationship with altitude variation. Orbits with longer cumulative sunlight exposure tend to exhibit lower altitude changes. (3) The spatial structure of the pattern evolves continuously over the year, with two prominent transition periods, one in March–April and another in September–October. Between these transitions, the ‘W’-pattern shifts rightward progressively over the RAAN degree. This indicates a tight correlation among the orientations of the Sun, Earth, and the orbital planes.
Decoding the ‘W’ pattern - To further understand the origin of the observed pattern, we analyze the geospatial states of satellite and atmospheric density conditions at their altitude. In Fig. 9, we visualize this state on 1st January at two timestamps, 06:00 UTC and 18:00 UTC, respectively. The background color in these figures represents atmospheric density at the Starlink shell’s operational altitude of 550 km, while the yellow marker denotes the subsolar point. Satellite positions at that particular timestamp from a few orbital planes are marked using the red and blue circles. Where red circles correspond to satellites in five consecutive orbital planes closest to RAAN 300 ∘ 300^{\circ} that experience the maximum altitude changes, notice in Fig. 9(a). Similarly, the blue circles correspond to satellites in five consecutive orbital planes, with the RAAN at 180 ∘ 180^{\circ} experiencing the minimum altitude change. Further, the solid and hollow red/blue circles indicate that satellites are in sunlight and Earth’s shadow, respectively.
At 06:00 UTC in Fig. 9(a), notice a strong day–night asymmetry in atmospheric density. The dayside shows significantly higher density than the nightside, up to a factor of four at the same altitude. This gradient directly impacts the drag experienced by satellites along their orbit. Notice the satellites in the red orbital group traverse through regions spanning the subsolar point (maximum density) and the midnight sector (minimum density) within a single orbit. As a result, these satellites encounter large density variations along their orbits around Earth. This leads to varying drag forces. Consequently, larger cumulative altitude changes over a day are reflected in TLE-derived measurements. In contrast, satellites in the blue orbital group remain close to the terminator (dawn–dusk line), where atmospheric density remains relatively stable due to the transition between day and night conditions. This results in near-constant drag along the orbit around Earth, producing minimal altitude variation across consecutive TLEs. This behavior is not transient. The same spatial situation persists 12 hours later, notice the Fig. 9(b). As shown earlier in Fig. 8, it remains consistent throughout the year.
These figures also explain the inverse relationship between sunlight exposure and altitude variation. Satellites in dawn–dusk aligned orbital planes remain illuminated for most of their orbital period, whereas satellites in subsolar aligned orbital planes undergo extended eclipse phases, typically 30–40% of their orbital period. Observe this illustrated in Fig. 9(a), satellites in blue orbits are predominantly sunlit except a few between 120∘-170∘W, while a large fraction of satellites in red orbits are in Earth’s shadow, notice between 30∘-180∘W.
(a) At 06:00 UTC, 1st January, 2024 (b) At 18:00 UTC, 1st January, 2024 Figure 9 . Color mesh in background represents the global atmospheric density at 550 km altitude, showing that daytime density is about four times higher than nighttime levels. Red markers indicate satellites passing through regions of maximum and minimum density, leading to significant altitude variations over the course of a day. Blue markers represent satellites traveling along dawn–dusk paths, where atmospheric density remains nearly constant, resulting in minimal altitude changes.
What happens during a solar storm? - During solar storms, the spatial distribution of atmospheric density exhibits a strong temporal and geographic heterogeneity, rather than a uniform global increase. To illustrate this behavior, in Fig. 10, we present two snapshots from the May 2024 solar superstorm and the October 2024 solar storm. At the onset of the solar storm, the atmospheric density inflation is concentrated in the polar regions. As the storm evolves, we observe high-density patches propagate towards the equator like waves, forming a non-systematic dynamic temporal density distribution. In Fig. 10(a) at 02:00 UTC on 11th May, we can observe a high-density patch originating in the polar region moving towards the equator in the nightside hemisphere. All the states in Fig. 10 illustrate this complex scenario of non-systematic dynamic temporal density distribution in atmospheric conditions during the solar storm. Despite these complex dynamics of localized high-density patches, the overall day–night asymmetry in atmospheric density remains preserved. As shown in the snapshot at 04:20 UTC for both events, the dayside continues to exhibit overall higher density than the nightside. Consequently, the key driver of ‘W’ pattern formation in altitude changes remains intact even during storm conditions, as reflected in Fig. 8(e) and (j). However, the absolute magnitude of atmospheric density increases substantially during the storm, with typical values rising by a factor of 6–8 compared to quiet periods. This amplification directly translates into stronger drag forces, resulting in a corresponding increase in altitude decay from approximately 150 meters under nominal conditions to up to 700 meters during the solar storm.
(a) At 02:00 UTC, 11th May, 2024 (b) At 04:20 UTC, 11th May, 2024 (c) At 02:00 UTC, 11th Oct, 2024 (d) At 04:20 UTC, 11th Oct, 2024 Figure 10 . Color mesh in background represents dynamics of spatiotemporal global atmospheric density distribution at 550 km altitude during (a)-(b) May 2024 solar superstorm and (c)-(d) October 2024 solar storm. The overall atmospheric density increased by up to 8 times compared to typical conditions. Despite that, overall dayside density remains about 4 times higher than nightside density, so satellite altitude variations become larger in magnitude, while the underlying pattern of behavior remains unchanged.
Decoding pattern transition - The Earth revolves around the Sun at an angular rate of 0.986 ∘ 0.986^{\circ} per day, which can be approximated as 1 ∘ 1^{\circ} per day for simplicity. Therefore, the relative angular rotation of Earth around the Sun is roughly 30 ∘ 30^{\circ} per month. Notice this value is consistent with Fig. 8, where the ‘W’ pattern shows a systematic rightward shift of 30 ∘ 30^{\circ} in RAAN across successive months. In addition to this longitudinal shift, the latitudinal position of the subsolar point varies over the year due to the Earth’s axial tilt of approximately 23.5 ∘ 23.5^{\circ} . As a result, the subsolar point migrates between 23.5 ∘ 23.5^{\circ} N in summer and 23.5 ∘ 23.5^{\circ} S in winter over an annual cycle. This seasonal movement is reflected in the geospatial visualizations, where the subsolar point appears in the southern hemisphere in Fig. 9(a)-(b), while it shifts to the northern hemisphere in Fig. 10(a)-(b). The transition of the subsolar point across the equator occurs around the equinox periods (March–April and September–October). These transitions correspond directly to the shifts observed in the structure of the ‘W’-pattern in Fig. 8(c)–(d) and (i)–(j), indicating that the evolution of the pattern is tightly coupled to seasonal changes in Sun–Earth geometry.
Universal LEO Trait? - The LEO altitude range is generally considered to be 160-2,000 km. All the Starlink operational shells are within a range of 540–570 km. This ‘W’ pattern altitude variation is visible in all the Starlink shells. The pattern becomes more pronounced in shells with a larger number of orbital planes, whereas higher-inclination shells (e.g., 97.6 ∘ 97.6^{\circ} ), which contain fewer orbits, exhibit a weaker manifestation of the pattern. In the prior work (Kang et al., 2025) , this was already shown.
To examine whether this behavior generalizes across LEO constellations at different altitudes, we analyze the OneWeb constellation, which operates at 1,200 km, roughly twice the altitude of Starlink. In Fig. 11, we present the corresponding analysis. Note that OneWeb satellites are deployed in Walker-star configuration (Wood, 2012; Basak et al., 2025a) with 12 orbital planes uniformly across 180 ∘ 180^{\circ} of RAAN degree. Also, we do not account for atmospheric density data in this analysis, as the TIE-GCM model (Center, 2026) is limited to altitudes below 1,000 km. At higher altitudes, the atmosphere becomes too thin to manifest general fluid-dynamic properties. Consequently, Fig. 11 shows marginal altitude differences on the order of 2–3 meters across orbital planes. Furthermore, even during the May 2024 solar superstorm in Fig. 11(c), no significant amplification is observed. This indicates that, at higher LEO altitudes, atmospheric drag is too weak to induce the structured variations seen in lower-altitude constellations, resulting in relatively stable orbital behavior.
Between these regimes, the Amazon Leo constellation is being deployed at altitudes ranging from 590 to 650 km (Basak et al., 2025a) , directly above the Starlink shells. However, the current deployment is of 212 satellites, with most still in the orbit-raising phase. In Fig. 12, we show the altitude change since the first launch and the distribution of all the satellites at the current state, indicating that only a fraction have reached operational altitude. As a result, the current state is insufficient to evaluate the presence or absence of the ‘W’-pattern in this constellation. There are Earth observation satellites operating at altitudes around 800 km. Again, these satellites are typically placed in sun-synchronous orbits with a narrow range of RAAN. This limited angular coverage prevents meaningful analysis of cross-orbital variation, making them unsuitable for validating the observed pattern.
{keybox} Key takeaway (5):- The root cause of the ‘W’ pattern in altitude variation (up to 150 meters) across the orbits is the geospatiality of atmospheric density distribution between day and night. This pattern exists throughout the year and continuously shifts in synchronization (1∘/day) with Earth’s rotation around the Sun. During the solar storm, this pattern amplifies significantly, reaching up to 700 meters.
(a) 1st March, 2024 (b) 1st April, 2024 (c) 11th May, 2024 Figure 11 . Altitude variation of OneWeb satellites at 1,200 km shows that the ‘W’ pattern nearly disappears. The difference in altitude change is only about 2–3 meters in (a) March and (b) April. Even during the May 2024 solar superstorm (c), there is no noticeable difference.