Starship super heavy acoustics: Far-field noise measurements during launch and the first-ever booster catch

SpaceX's Starship spacecraft and Super Heavy booster (together referred to by SpaceX as Starship) are currently being flight-tested to become a rapidly reusable, multipurpose launch vehicle capable of crewed Mars missions (SpaceX, 2024). Because of its thrust, anticipated launch cadence, and booster return-to-pad maneuver (i.e., “flyback”), Starship has generated significant public interest. This interest has been fueled by recent high-profile news articles (Lipton, 2024; Pulliam and Maidenberg, 2024) and easy access to 24-h livestreams of Starbase operations (e.g., NASASpaceflight, 2024). Because Starship is the most powerful launch vehicle ever built, questions naturally arise regarding its noise: What are the sound levels during liftoff and its booster landing? How do these levels compare to predictions in environmental assessment (EA) documents? How does the noise compare to NASA's Space Launch System (SLS)—now the world's second most powerful, successfully launched rocket? How do the launch noise and flyback sonic boom compare to SpaceX's Falcon 9—the world's most frequently launched orbital rocket?

This letter discusses far-field sound levels and other noise characteristics measured during Starship's fifth flight test and compares them to EA predictions (FAA, 2022, 2024) as well as to SLS (Gee , 2023; Kellison , 2023) and Falcon 9 (Mathews , 2021; Anderson , 2024) measurements. As the need to understand potential impacts from rocket noise rapidly grows (Lubert , 2022), we hope these data will support both EA studies and model validation for launch noise (James , 2020b; Plotkin , 1997; Plotkin, 2010) and sonic boom (e.g., Lonzaga , 2022).

As Starship lifted off from Starbase's Orbital Launchpad A (OLP-A) [see Fig. 1(b)], Super Heavy's 33 Raptor 2 engines produced 74.4 MN (16.7 × 10⁶ lb) of thrust, almost twice that of SLS (39.1 MN) and nearly ten times that of Falcon 9 Block 5 (7.6 MN). In Fig. 1(a), relative scales of Falcon 9, SLS, and Starship are shown, along with a standard American football field for context. During liftoff and ascent, the close spacing of the Raptor 2 engine nozzles results in early plume interaction and coalescence, as evidenced by a vehicle-width shock-cell diamond in Fig. 1(c). Approximately 7 min after liftoff, the Super Heavy booster returned [Fig. 1(d)] to OLP-A, resulting in a widespread flyback sonic boom, propulsion noise during landing and booster catch [Fig. 1(e)], and an additional sonic boom due to the vehicle's hot-staging ring falling from high altitude into the ocean several kilometers from OLP-A. (The hot-staging ring is a removable, vented heat shield that prevents the booster from being damaged by upper-stage ignition.)

Acoustic pressure data were recorded at eight sites ranging from 9.7 to 35.5 km. Station numbers, locations, and distances from OLP-A are shown in Fig. 2. Station 1 was located on the Margaritaville Hotel roof, while the other stations were located at ground level. Stations 3 and 6 were placed within national wildlife refuge (NWR) areas with lower ambient noise levels, while other stations were closer to spectators viewing the launch (e.g., Station 5). As with prior rocket launch and sonic boom measurements, an in-house data acquisition system was used (Gee , 2020). The Portable Unit for Measuring Acoustics (PUMA) consists of a weatherproof case containing a ruggedized computer, Masterclock GPS-500 clock for synchronization, NI 9250 24-bit/5-V data acquisition module sampling at 102.4 kHz, and battery. Connected to the PUMAs were 12.7 mm (1/2″) GRAS 47AC (0.09 Hz–20 kHz) microphones at all stations except for Station 8, which used a GRAS 146AE (3.15 Hz–20 kHz) microphone. Digital pole-shift filtering (Rasband , 2023; Marston, 2008) was used to compensate for the low-frequency roll-off of the 9250 modules (0.43 Hz cutoff) and the 146AE microphone in particular. The microphones were housed in a weather-robust ground plate/windscreen system developed for launch vehicle measurements (James , 2020a) and further improved for sonic boom measurements (Gee , 2020; Anderson , 2022).

As our purpose is to document Starship far-field noise as it relates to EA noise studies and other launch vehicles, we focus first on a discussion of the significant acoustical events during Flight 5 and then report on unweighted and weighted levels generated during launch and booster flyback. For the launch noise, single-launch metrics from the Federal Aviation Administration (FAA) EA document (FAA, 2022) are calculated. The maximum unweighted (Z-weighted) level, $LZ,max$⁠, identifies the one-second period with the most acoustical energy from the running level, $Lz$⁠, whereas $LA,max$ corresponds to the maximum A-weighted level. While maximum levels are important, quantifying the total launch noise energy using the SEL is also valuable. The SEL calculation is sensitive to both the sound level and its duration. Although only the launch A-weighted SEL, $SELA$⁠, is found in EA documents, we also calculate the unweighted $SELZ$ for completeness and because the suitability of A-weighting to describe launch vehicle noise perception is uncertain (Gee , 2023). While A-weighting is used to correlate with human noise exposure and annoyance for numerous types of sounds, rocket noise is unique in that it is high-amplitude and includes significant low-frequency rumble and transitory crackle. It is possible that filtering out the low and high-frequency energy through A-weighting (designed for relatively quiet sounds) makes it insufficient to capture all relevant perceptual aspects of rocket noise.

For the booster flyback sonic boom, three metrics are obtained. First, measured sonic boom overpressure enables a direct quantitative comparison against the updated EA (FAA, 2024). Additionally, calculated perceived level (PL) (Stevens, 1972) from Flight 5 booms can be compared against other sonic booms and impulse noise events. Finally, calculating $SELA$ for the boom allows it to be directly compared against the launch noise in terms of total A-weighted energy.

Figure 3 shows the pressure waveform from Station 3 (10.5 km), along with the 1 s-averaged running Z (⁠ $LZ$⁠) and A-weighted (⁠ $LA$⁠) levels. Colored arrows show the timing of 1-s waveform segments that describe four events of acoustical significance: launch $LZ,max$ (orange), flyback sonic boom (blue), landing burn (red), and the sonic boom from the free-falling hot-staging ring (green). During maximum liftoff noise, the $LZ,max$ waveform segment [Fig. 3(b)] is shocklike and is heard as distinct crackle during liftoff and ascent. These shocks can be intense; at Station 2, car alarms went off during the maximum launch noise period. It is surprising that the maximum noise during the landing burn [Fig. 3(d)], which involves igniting the inner 13 booster engines before shutting off all but three for the booster catch, is on par with the maximum liftoff noise when all 33 engines are firing. More detailed investigations are needed as the fluid mechanics and aeroacoustics of a supersonic plume firing forward of the vehicle and in the direction of travel are complex (e.g., Vos , 2022).

The highest-amplitude event in Fig. 3(a) and at all eight stations is the flyback sonic boom [Fig. 3(c)], which set off car alarms at Stations 2 (10.1 km) and 4 (16.6 km). A prior Falcon 9 study (Anderson , 2024) shows near the landing pad, maximum launch noise exceeds the flyback boom, but that there is a range (∼2 km for the Falcon 9) beyond which the cylindrically spreading boom's overpressure becomes larger in amplitude than the spherically spreading launch noise. The booster's flyback boom's overpressure of 7.1 psf (0.34 kPa) in Fig. 3(c) is part of a clean triple-shock waveform that is similar to Falcon 9's signature (Anderson , 2024), despite the fact that Falcon 9 has a different geometry. An ongoing investigation into the aeroacoustic origins of Falcon 9's triple boom, when complete, should also provide insights into the Super Heavy flyback boom.

Finally, a relatively unique event captured as part of this launch is the hot-staging ring sonic boom, which occurs about 30 s after booster catch. This boom has a classic N-wave signature, with greater evidence of atmospheric turbulence affecting its shape (possibly due to its landing in the ocean several kilometers from OLP-A) and producing a maximum overpressure of 0.70 psf (34 Pa). This boom was also detected at all eight stations, including the north Brownsville station (Station 8) at a distance of 35.5 km. At Station 8, the boom's rise time was significantly increased and its amplitude was about 0.27 psf (13 Pa). While an interesting acoustical event, the ring boom is not discussed further in this letter.

One final comment regarding Fig. 3(f) has to do with the launch noise duration. The running $LA$ reaches ambient by ∼350 s, before increasing for the sonic boom and other landing-related noise. However, this is not true for $Lz$⁠, which is dominated by infrasound that propagates near-losslessly. An examination of several recordings shows that it takes 16–17 min (960–1020 s) for the Z-weighted levels to reach prelaunch levels, depending on the ambient environment. Low-frequency launch noise persistence for several minutes has been noted previously for Falcon 9 by Mathews (2020) and SLS by Kellison (2023), and long-range infrasound propagation from a large number of rocket launches has been studied by Pilger (2021).

We now focus on an analysis of relevant noise metrics at all eight stations, first for the launch noise and then for the booster flyback sonic boom. Shown in Figs. 4(a)–4(d) are launch $LZ,max$⁠, $LA,max$⁠, $SELZ$⁠, and $SELA$⁠, respectively, for all eight stations as a function of OLP-A range. Logarithmic curve fits are used to show the approximate decay trend over this distance range and aid in discussion. Along with the measured data are levels extracted from the coarse contour plots found in FAA (2022) for $LZ,max$⁠, $LA,max$⁠, and $SELA$⁠, labeled here as predicted data. In addition to the data being independently valuable, this analysis represents a direct validation of the Starship EA launch noise modeling at Boca Chica, which uses the RNOISE model based on the work of Sutherland (1993), Plotkin (1997), and Plotkin (2010).

In Fig. 4(a), there is considerable scatter for $LZ,max$ relative to any reasonable decay trend across distance. Examination of the stations' waveforms suggests short-term, high-amplitude events within the liftoff noise drive the variation in $LZ,max$⁠. While the measured levels around 10 km are only 2 dB greater than the EA prediction, by 20 km this difference has grown to 4–5 dB. Most markedly, Station 8 (north Brownsville) measured a $LZ,max$ 9 dB greater than predicted in the EA.

The change in $LZ,max$ only reveals part of the overall picture. For example, $LA,max$'s decay trends in Fig. 4(b) match the EA model well, despite a scatter of about $±$5 dBA. The town of Port Isabel (∼10 km) sees a maximum of ∼105 dBA, equivalent to an average rock concert and chainsaw levels (Berger , 2016), whereas $LA,max$ in Laguna Vista (∼20 km) is also near the EA-predicted value of 90 dBA. This weighted sound level is associated with table saws, snow blowers, and loud shouting (Berger , 2016). Finally, the maximum levels in Brownsville (30–35 km) are ∼75–80 dBA, equivalent to a hair dryer or vacuum cleaner. However, we note the noise quality will be quite different given the greater low-frequency energy present in the rocket noise.

The SEL metrics in Figs. 4(c) (⁠ $SELZ$⁠) and 4(d) (⁠ $SELA$⁠) consider not only the maximum level of the sound but the total sound duration. Because $SELZ$ does not appear in the EA, there is no comparison curve, but there is a remarkably smooth decay in measured level out to Station 8 in Brownsville (35.5 km). A similarly smooth decay in measured $SELA$ also occurs over the 9.7 – 35.5 km range. For context, Starship's 113 dBA $SELA$ at 10 km is similar to the 110–115 dBA measured 61 m (160 times closer) from large commercial aircraft (A380, 747-400, 777-300ER) at takeoff (EASA, 2023).

The discrepancy between the measured $SELA$ and EA prediction in Fig. 4(d) is one of the major results of this letter. Unlike $LZ,max$⁠, for which the EA underpredicted Starship levels, the EA overpredicts $SELA$ measurements. The difference is ∼10 dBA in Laguna Vista (20 km) and ∼18 dBA in north Brownsville (35.5 km). After the relatively good agreement between measurement and model for $LA,max$⁠, this large difference for an integrated A-weighted metric is surprising. Although this difference is a welcome finding in terms of assessing potential Starship noise impacts, it also means there is a shortcoming in the EA model's ability to handle changes in spectrum and level that occur during launch. Modeling improvements should be sought in the future to obtain more accurate EA predictions near Boca Chica, Cape Canaveral, and elsewhere.

In addition to launch noise, Fig. 4 also displays flyback sonic boom metrics at the eight stations. Figure 4(e) shows the flyback boom overpressure. The 2024-updated EA overpressure contour plot (FAA, 2024) was fit using ellipses at each of the contours to create a function that was then evaluated at the measurement station locations to produce the predicted level markers in Fig. 4(e). At Station 3 (10.5 km) and beyond, the updated EA predictions closely follow the measured boom overpressures. However, for the first two stations, the measured overpressure deviates appreciably from predictions. Because overpressures near and exceeding 10 psf (0.48 kPa) at Port Isabel and the south end of South Padre Island pose greater risk of structural damage, such as glass breakage and falling bric-a-brac (Haber and Nakaki, 1989), additional measurements closer to OLP-A should be made.

The boom impacts were also asymmetric. Whether due to trajectory or meteorology, boom overpressures were sufficient to set off car alarms at Stations 2 (10.1 km northwest of OLP-A) and 4 (16.6 km north-northwest of OLP-A), but no car alarms were set off at Station 5 (14.5 km southwest of OLP-A) despite it having a slightly greater overpressure than Station 4 and several dozen vehicles present. Perhaps in the future, the difference between boom impacts at these stations represents a convenient opportunity to study boom characteristics that cause car alarm responses.

Beyond the overpressures in Fig. 4(e), Fig. 4(f) displays two sonic boom metrics—NASA's implementation (Shepherd and Sullivan, 1991) of Stevens Mark VII PL (Stevens, 1972) and $SELA$⁠. These metrics are two of several that have been adopted by NASA to assess human perception and annoyance due to sonic booms for quiet supersonic flight research (Loubeau and Page, 2018). Used here, PL allows for a direct comparison to other noise sources, while $SELA$ enables a direct comparison between launch noise in Fig. 4(d) and the flyback boom. At 10 and 20 km, the booster flyback boom results in a PL of about 125 and 110 dB, respectively, based on the curve fit in Fig. 4(e). Per Doebler and Rathsam (2020), who compiled values of PL for different impulsive noise sources, 125–135 dB represents a gunshot recorded at 2 ft, 115 dB is like a firework at 500 ft, and 103 dB is the PL of “nearby thunder.” Notably, a supersonic Concorde flying at an altitude of 18 km produced a sonic boom with a PL of ∼105 dB at the ground. With a difference of 5 dB, one approximation for a Starship flyback boom at 20 km is a ∼50% increase in loudness over the Concorde boom (where 9 dB represents a loudness doubling; see Stevens, 1972).

Comparing the Super Heavy booster to SpaceX's other reusable rocket, Falcon 9, is also instructive. Falcon 9's PL at 10 km is about 110 dB and, by 20 km, has fallen to around 90 dB (Anderson , 2024). Thus, another way to describe Starship's noise is to say that Starship's PL at 20 km is equivalent to Falcon 9's PL at 10 km. The distance-doubling for equal level also seems to hold from 20 to 40 km, given Starship's PL at 35.5 km is around 91 dB. [Note that we caution against trying to extrapolate this rule close to landing zones, given a plateauing of boom metrics in these regions caused by vehicle slowing (Anderson , 2024).] Regarding the $SELA$ curves in Figs. 4(d) and 4(f) for Starship's launch noise and booster flyback boom, the launch noise is about 5 dB greater in overall A-weighted noise energy. This offset appears to be similar for Falcon 9, which shows a difference of 5–12 dB (depending on the particular launch and distance), but more data points are needed for both vehicles to solidify a trend (Anderson , 2024).

Comparing the far-field sonic boom levels for Starship and Falcon 9 suggests a distance-doubling for an equal level. A launch noise comparison of Starship vs SLS and Falcon 9 for the four metrics in Figs. 4(a)–4(d) at 10 and 20 km is also revealing. Because measurement locations and distances were different for each vehicle, curve fits similar to those in Fig. 4 were obtained for each dataset to provide the most consistent comparison. For SLS, data from Gee (2023) and Kellison (2023) were used. For Falcon 9, the launch noise data from the three launches discussed by Anderson (2024) were analyzed and curve-fit to produce all four metrics. Table 1 shows $LZ,max$⁠, $LA,max$⁠, $SELZ$⁠, and $SELA$ for SLS and Falcon 9 at 10 and 20 km, relative to the levels of Starship. As expected from relative thrust [see Fig. 1(a)], the differences in level between Starship and Falcon 9 are greater than Starship and SLS.

Although it is difficult to identify trends across metric, vehicle, and distance, the most important conclusion from Table 1 is that a Starship launch represents a substantially greater noise source than SLS and Falcon 9. A look across the four metrics and two distances for SLS relative to Starship suggests that a Starship launch is the equivalent of 4–6 SLS launches in terms of noise impact, despite SLS producing more than half of Starship's thrust. The difference may be partly due to nozzle configuration, as hypothesized by Kellison and Gee (2023) as part of a recent comparison of SLS and Saturn V sound power level. They found that the Saturn V, with its clustered nozzle configuration, produced 2 dB greater power level despite lesser thrust. While the launch cadence of SLS will remain low, the same is not true for Falcon 9: ∼400 launches have already occurred over 14 years, with half of them in the past two years. Even a conservative interpretation of Table 1 indicates that a Starship launch produces more than ten times the noise of a single Falcon 9 launch at the distances considered in this letter.

We have described far-field (9.7–35.5 km) noise measurements associated with SpaceX's Starship fifth test launch and first booster catch. Although many other analyses remain, this letter communicates the following findings. First, acoustical events of significance include maximum liftoff noise, booster flyback sonic boom, landing maneuver noise, and hot-staging ring sonic boom. The flyback sonic boom is the event with the largest overpressure. Additionally, intense crackle during liftoff and landing can be tied to shocklike waveform characteristics. Second, although EA models are fairly accurate for some metrics at these distances, seemingly self-inconsistent discrepancies remain. The measured Z-weighted maximum level is 9 dB greater than the EA prediction at 35.5 km. On the other hand, the measured A-weighted SEL (⁠ $SELA$⁠) is ∼18 dB lower than the EA prediction at the same distance; this difference is surprising after the measured maximum A-weighted level was predicted reasonably well in the EA. The reasons for this discrepancy are unclear but suggest the need for additional near-source measurements and refined modeling. Third, sonic boom overpressures around 10 km are 1–4 psf greater than modeled, with the possibility of exceeding 10 psf (0.48 kPa) and increasing structural damage claims. By 20 km, these flyback booms have similar perceived levels as Falcon 9 at 10 km and are about 50% louder than a Concorde boom. Fourth, a launch noise comparison between Starship, NASA's SLS, and Falcon 9 suggests Starship is significantly louder by both unweighted and A-weighted metrics. At these distances, the noise from one Starship launch is equivalent to around 4–6 SLS launches and at least 10 Falcon 9 launches. Although these are relative comparisons, they provide greater insight into a vehicle that may soon launch more than 100 times per year. To understand Starship's potential noise impacts on sensitive structures, communities, and the environment, additional measurements are needed to verify this letter's findings and to extend them to closer ranges.

The authors have no conflicts to disclose.

Some data that support this study's findings are available by request to the corresponding author.