Radar and Infrasound Observations of the 2026 May 30 Cape Cod Bay Superbolide

Introduction

Detailed measurements of metre-scale hypervelocity atmospheric entries are rare [8]. Observations of their trajectories, plasma wakes, fragmentation, and energy release are needed to test and improve atmospheric-entry models in this extreme high-Mach-number regime [59]. For planetary defense, accumulating pre-impact orbits of these high-energy impactors is also essential to characterizing their source populations and encounter conditions [8, 31].

Corresponding author: Juha Vierinen

Email: [email protected]

On 2026 May 30, a bright daytime bolide crossed the Massachusetts coast and ended over Cape Cod Bay. Public accounts reported a widely heard sonic boom, indicating a meteoroid large enough to generate a shock front. The NASA/JPL Center for Near-Earth Object Studies (CNEOS) reported a U.S. Government sensor detection, with peak brightness at 18:06:23 UTC near $42.0^{\circ}$ N, $70.5^{\circ}$ W and 32.0 km altitude, a total radiated energy of $4.21 \times10^{11}$ J, and an estimated impact energy of 1.1 kt of trinitrotoluene (TNT) equivalent ((NASA/JPL Center for Near-Earth Object Studies 2026)). Only three events in the CNEOS catalogue for 2025 had estimated impact energies exceeding the 1.1 kt reported for the Cape Cod Bay bolide. For scale, the reported impact energy is comparable to the 1.2 kt TNT-equivalent asteroid 2008 TC3 impact over Sudan and about one quarter of the 4 kt Sutter’s Mill event, both reported by P. Jenniskens et al. [33]. This places the Cape Cod Bay bolide in the metre-scale impactor regime.

Serendipitously, this event was captured by a wide range of sensors that could be used to track the meteoroid from atmospheric entry down to near sea level and elucidate the nature of the meteoroid: dashcam and security cameras recorded the path through openings in clouds, infrasound sensors recorded the acoustic signature generated by the shock wave, the Massachusetts Institute of Technology (MIT) Haystack Observatory specular meteor radar network detected echoes from the atmospheric entry plasma in the upper atmosphere, and four weather radars detected signatures of falling meteorites in the lower atmosphere. Together, these observations constrain the orbit, provide a detailed description of the atmospheric entry, and predict the meteorite strewn field. Figure 1 shows: (a) the bolide in a security-camera frame from Mechanicville, New York, and (b) GOES-19 Geostationary Lightning Mapper (GLM) observations. The location of the camera is shown in Figure 2. The GLM optical transient was reported by CIMSS ((Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin–Madison 2026)); the use of GLM observations for bolide detection is discussed by J. C. Smith et al. [65].

We combine meteor-radar, weather-radar, and infrasound observations to reconstruct the trajectory and pre-impact orbit, constrain fragmentation and energy release, and connect atmospheric entry to the predicted meteorite strewn field. We examine the unusually low-altitude plasma echoes and assess what the observations imply about the body’s mass, mechanical strength, and possible composition.

Methods

The primary trajectory data are obtained from raw complex voltages ([72, 73]) of the 31.65 MHz MIT Haystack Observatory Spread Spectrum Interferometric Meteor Observing Network (SIMONe) coded continuous-wave transmitters and meteor-radar network receivers [71, 9, 10]. The radar system employs five linearly polarized transmit antennas radiating 500 W each. The five channels utilize different pseudorandom 10 ms long continuous binary phase coded transmit waveforms with 100 $\mu$s bit lengths, allowing estimation of transmit-target-receiver propagation path-length, rate-of-change of path length, as well as interferometric angle-of-departure [9]. The radar receivers employ dual polarized antennae located in Beverly, Clinton, Pomfret, Siena University, and Amherst (Figure 2). At the VHF frequency used by the radar, the measured radar echoes arise from scattering by the plasma produced by the atmospheric entry of the meteoroid and plasma turbulence in its wake – not direct detections of solid meteoroids.

In contrast, the microwave weather radars, with a much shorter wavelength, are sensitive to small solid meteoroids in dark flight. This is possible because the meteoroids are comparable in size to the radar wavelength, allowing for resonant scattering. We used volume-scan data from the NOAA S-band WSR-88D weather radars KBOX, KENX, and KOKX, and from the FAA Doppler weather radar TBOS, which operates at C band. Radar-signature centroids and times constrain fragment descent. Figure 2 shows the weather-radar sites and the measured falling meteorite positions.

Infrasound measurements provide independent constraints on the bolide’s source direction and energy through the atmospheric pressure waves generated during entry and fragmentation. We surveyed stations within 2000 km of the CNEOS peak-brightness location, obtaining usable records from 199 sensors, including conventional microbarometers, the Elginfield Infrasound Array (ELFO), and citizen-operated Raspberry Shake/Boom pressure sensors. Figure 11 shows the infrasound station locations. Candidate arrivals were tested against acoustic travel-time windows and local background noise; array beamforming and inter-station envelope delays constrained the arrival direction, while dominant signal periods provided an empirical energy estimate.

Meteor radar

The SIMONe Haystack meteor radar raw voltage data were processed using range-Doppler matched filtering for the high-Doppler-shift meteor head echoes for the plasma surrounding the moving meteoroid, and compressive sensing radar signal processing [70] for the slowly moving plasma wake. The head-echo analysis estimated the transmitter–target–receiver path lengths and their rates of change. We used a matched filter to detect the five transmit signals and searched for the best-matching angle of departure (AoD) and polarization to maximize the signal-to-noise ratio and estimate an independent interferometric location for the head echoes.

The slowly moving, range-spread plasma structures formed in the meteoroid’s wake were analyzed using a compressive sensing approach, which simultaneously deconvolves the continuous coded transmit signals from the five transmitters [69]. This allowed three-dimensional localization of the meteor trail plasma turbulence using interferometric angles of departure and propagation delays [68].

The head-echo radar cross section (RCS) was estimated using the bistatic radar equation:

Here, $P_r$ is the received signal power, $P_t$ is the transmit power of one transmit antenna, $R_t$ is the transmitter–target distance, and $R_r$ is the target–receiver distance. The effective transmit and receive gains toward the target, $G_t$ and $G_r$, are linear power ratios, and $\lambda$ is the radar wavelength. The effective bistatic RCS, $\sigma_{\mathrm{RCS}}$, is expressed in square metres and corresponds to the transmitted linear-polarization state and the polarization-matched combination of the two receive channels.

The single mode noise power is

where $k_{\mathrm{B}}$ is Boltzmann’s constant, $T_{\mathrm{sys}}$ is the system noise temperature, and $B$ is the matched-filter noise bandwidth. We used $B = 100$ Hz for a single 10 ms code. At the radar frequency of 31.65 MHz ($\lambda= 9.47$ m), sky noise, primarily from Galactic synchrotron emission, dominates receiver noise. We therefore used $T_{\mathrm{sys}} = 1.54 \times10^4$ K, obtained from the LWA1 low-frequency sky model [16, 48], giving $P_n = 2.12 \times10^{-17}$ W.

Defining the linear signal-to-noise power ratio as $\gamma_{\mathrm{SNR}} = P_r/P_n$, the RCS estimator becomes

We evaluated $R_t$ and $R_r$ at each sample using the meteoroid trajectory. The conversion assumes $P_t = 500$ W, an effective transmit gain of 7.0 dB for trajectory-directed transmit combining, and an effective receive gain of 3.0 dB for the polarization-matched receiver. The power and gain values are first-order estimates rather than absolute calibrations, so the absolute RCS should be regarded as a rough order-of-magnitude estimate.

The trajectory was fitted in the Geocentric Celestial Reference System (GCRS) by combining timed head-echo range/Doppler measurements with non-specular trail positions. In the fit, the head echoes constrain position and velocity at the time of observation. The trail echoes, on the other hand, provide point-on-path constraints: the measured trail-echo positions constrain the meteoroid’s path, but not the precise times at which it passed those positions, because the echoes originate from the turbulent plasma wake left behind by the meteoroid. The trajectory model included atmospheric drag, Earth’s gravitational acceleration, and Earth’s rotation. The atmospheric density was taken from an NRLM SIS 2.1 altitude profile [18].

Because of the object’s large mass and the relatively short head-echo path between altitudes of 90 and 70 km, the measurements could not constrain the deceleration due to atmospheric drag. NASA’s Meteoroid Environment Office estimated a diameter of 1.6 m and a mass of 5600 kg [41]. For the fixed-drag calculation, we adopted a 1.6 m diameter sphere with a mass of 5000 kg and a drag coefficient of one, giving a fixed drag-area-to-mass ratio of $4.021 \times10^{-4}\ \mathrm{m^2\ kg^{-1}}$. The trajectory fit is not sensitive to this choice, as no observable deceleration could be detected from the head echo.

Head-echo range and Doppler measurements were selected independently by manual inspection of the station range–Doppler products. The fit retained 229 ranges, 137 Doppler measurements, and 146 trail positions. Trail selections required SNR $\ge8$ dB, altitudes of 20–115 km, and an AoD offset within $35^\circ$ of the search prior; an isolated Beverly AoD solution at 18:06:24.88 UTC was excluded.

The six inertial-state parameters were estimated by generalized least squares using head-echo range and Doppler residuals and three-dimensional trail residuals to the closest point on the propagated path. Head-echo error scales were estimated from residual scatter, giving $\sigma_R = 828$ m and $\sigma_f = 25$ Hz; no SNR dependence was retained. Trail uncertainties used the interferometric three-dimensional position error, bounded to 1.2–12 km, as an isotropic per-coordinate standard deviation. Different stations and observable components were treated as independent covariance blocks. The three-code incoherent head-echo averages, stepped by one 10 ms code, gave within-block correlations of 2/3 and 1/3 at lags of one and two codes. Trail correlations followed the overlap of eight-code coherent integrations and were restricted to the same station and raw 3 km range gate; no retained trail pairs met both conditions. The state covariance was $(J^{T}\Sigma^{-1}J)^{-1}$, where $J$ is the residual Jacobian and $\Sigma$ is the measurement covariance. The reduced $\chi^2$ was 1.29.

The fitted inertial state and covariance were sampled and transformed using the DE440 planetary ephemeris [46]. Samples were propagated backward with the REBOUND N-body integrator [49] until they were outside Earth’s sphere of influence, then converted to heliocentric osculating Keplerian elements referred to the mean ecliptic and equinox of J2000.0.

Weather radar and dark-flight modeling

The Jörmungandr2 (J2) dark-flight model combined the SIMONe path with the weather-radar signature centroids to define the post-fireball flight regime [21, 23]. The 30–34 km plasma-detection range set the luminous-flight terminus; each weather-radar centroid was back-propagated to 3 km s$^{-1}$ to estimate a common terminus. J2 then iteratively solved for the fragment mass that traversed the terminus-to-detection distance in the observed time under the measured winds. The wind profile was constructed from a distance-weighted average of NOAA radiosonde measurements from the three stations nearest the fall site. J2 also extrapolated each fragment to sea level and calculated an extended strewn field for generic masses from 0.1 g to 10 kg. Total fall mass was estimated both by assuming 10% survival of the pre-atmospheric mass and by integrating the J2 radar-derived mass–number distribution; the latter is a currently unvalidated method presented for informational purposes only.

Infrasound Observations and Analysis

Infrasound observations were used as an independent constraint on the source direction and energy of the Cape Cod Bay bolide. The reference source location was the peak-brightness location reported by the National Aeronautics and Space Administration/Jet Propulsion Laboratory Center for Near-Earth Object Studies (NASA/JPL CNEOS; $42.0^\circ$ N, $70.5^\circ$ W; 32.0 km altitude), for which a calculated impact energy of 1.1 kilotons of trinitrotoluene (TNT) equivalent was reported (1 kt TNT $= 4.184 \times10^{12}$ J). The VHF radar terminus at $41.9105^\circ$ N, $70.4202^\circ$ W and 33.61 km altitude was retained as an independent spatial reference.

All openly available infrasound stations within 2000 km of the CNEOS location were surveyed. Usable waveform data were obtained from 199 sensors, including conventional single-station microbarometers, the four-element Elginfield Infrasound Array (ELFO) of Natural Resources Canada, and citizen-operated Raspberry Shake/Boom pressure sensors. All waveform data were accessed through the International Federation of Digital Seismograph Networks (FDSN).

Candidate single-station arrivals were restricted to physically admissible acoustic propagation windows following established bolide-infrasound screening practice [57, 76]. Arrival windows were defined using nominal tropospheric, stratospheric, and thermospheric celerity ranges [43]. To quantify the likelihood of coincident background signals, we developed a matched-window empirical false-alarm test in which the event window was compared with equal-duration off-source windows from the same station record. The empirical false-alarm probability was defined as the fraction of off-source windows whose peak envelope amplitude equaled or exceeded that of the candidate arrival. Candidates were subsequently inspected in the filtered waveform and spectrogram, and ambiguous arrivals were excluded from quantitative analysis.

ELFO was analyzed using sliding-window frequency-wavenumber (FK) beamforming to estimate back-azimuth and trace velocity, following standard infrasound array-processing principles [53, 13]. Element records were decimated to 25 Hz and bandpass-filtered from 0.14 to 2.0 Hz, with the passband selected from the array signal-to-noise spectrum. Bartlett FK processing was performed in 25 s windows advanced by 5 s over a propagation-slowness grid spanning $\pm3.6$ s/km in 0.05 s/km increments. For each window, time-domain semblance was calculated at the preferred FK slowness after alignment, demeaning, and root-mean-square normalization of the individual element records, providing an amplitude-normalized measure of inter-element waveform coherence [44]. Signal onset was defined as the first sustained, source-consistent window exceeding the pre-arrival semblance background at a physically admissible celerity [57]. Directional uncertainty was estimated by element jackknife applied to an independent cross-correlation delay solution [66, 24].

A spatially distributed cluster of six Raspberry Shake/Boom sensors, located 65–101 km from the source and separated by up to 36 km, was analyzed jointly as an array of opportunity. Because the inter-station separations substantially exceeded the acoustic wavelength of the observed signal, conventional phase-coherent FK processing was not appropriate. Inter-station delays were instead estimated from waveform-envelope correlations, and the preferred plane-wave solution was obtained by maximizing envelope coherence over back-azimuth and trace velocity. Detection significance was evaluated against randomized time-shift and off-source coherence distributions, and directional uncertainty was estimated by station jackknife. This approach extends previous use of spatially distributed single-sensor stations for bolide localization through multilateration [76] by obtaining a direct direction-of-arrival constraint from inter-station coherence.

Dominant signal periods were measured at maximum signal amplitude using the established zero-crossing approach for bolide infrasound, with spectral-peak periods used as an independent consistency check [19, 51, 57]. Source energy was estimated from the network-averaged dominant period using the bolide-specific multi-station period-yield relation of E. A. Silber et al. [62],

where $E$ is the source energy in kilotons of TNT equivalent and $P$ is the network-averaged dominant period in seconds. The energy estimate used the five retained single-station detections together with the ELFO beam period.

Results

VHF radar echoes

Figure 3 shows the range–time–echo power structure on all five VHF links. Range here is the total transmit-target-receiver propagation distance. The high-Doppler head echo is visible inside the cyan selection boxes at Beverly, Clinton, and Pomfret and follows the decreasing bistatic range of the fitted trajectory from approximately 90 to 71 km altitude. Clinton contains the shortest retained head-echo segment; Amherst and Siena did not detect head echoes. The brighter, more range-extended returns that follow are non-specular plasma trail echoes. At altitudes above approximately 72 km their onset is delayed by several seconds relative to passage of the body, whereas at lower altitudes the trail response develops progressively closer to the trajectory passage time. Trail echoes remain detectable far below the head-echo interval, down to approximately 30 km.

The head echo RCS measured with the Beverly receiver is shown in Figure 4. The RCS is around 20–30 dBsm during the strongest part of the head echo between 90 and 70 km, comparable to or larger than representative microwave-frequency RCS values for large commercial aircraft [64]. Measurement samples outside that interval are also shown, but these are due to receiver noise. The strongest trail echoes in Figure 3 reach SNRs near 20 dB, which gives order-of-magnitude peak bistatic RCS estimates up to approximately 40 dBsm per range–Doppler cell.

The head-echo Micro-Doppler spectra at Beverly, Clinton, and Pomfret are shown in Figure 5. The Doppler shift for all three links is shifted to appear at zero, showing only the deviation to a point-like target. During the strongest returns, the spectra contain a single narrow ridge around the fitted head-echo Doppler, with no resolved secondary tracks indicative of differential fragment velocities.

The head-echo measurement set consists of transmit-target-receiver path lengths and rate-of-change of path length measurements for the three stations that measured a head echo. The measurements and the ballistic trajectory fit model obtained by fitting the positions of the trail echoes are shown in Figure 6. The measured trail-echo positions and their relationship to the subsequent dark-flight analysis are shown in Figure 7.

The tri-static head-echo time-of-flight solution and the transmit interferometry provide complementary estimates of the meteoroid position. In the former, the position is constrained by simultaneous bistatic path lengths from multiple receivers; in the latter, the measured transmit AoD fixes a ray from Haystack and the bistatic path length fixes the distance along that ray. For the head-echo samples that passed the AoD selection, the straight-line fit to the AoD plus delay positions was separated from the tri-static trajectory by 5.1 km RMS. The median pointwise separation of individual AoD plus delay positions from the tri-static trajectory was 5.7 km.

Weather-Radar Detections and Dark Flight

Signatures of falling meteorites from this event are visible in data from WSR-88D weather radars in the National Oceanic and Atmospheric Administration (NOAA) NEXRAD radar network. The signatures are identifiable as such through a combination of factors: they appear at the time and place of a bolide, the ensemble of signatures appears as a vertical column consistent with material falling over a range of altitudes that include values higher than the ambient weather, and the earliest-arriving signatures include pronounced short-range turbulence consistent with high-velocity objects. These features have been used to identify meteorite falls in NEXRAD data both as they occur and in archived data, for a total of 61 falls as of July 2026, including the recovered Lake Huron event (M. Fries, personal communication, 20 July 2026) [21, 33, 34, 26, 35, 22, 54].

Mean / $10^{4}$				Covariance / $10^{4}$
Variable	Estimate	$x$	$y$	$z$	$v_x$	$v_y$	$v_z$
$x$	10.183	25.84	6.168	3.129	−4.214	0.5103	0.6241
$y$	487.82	6.168	19.05	16.55	−1.608	−5.345	−5.918
$z$	437.83	3.129	16.55	39.68	−0.4369	−5.837	−10.19
$v_x$	−0.87827	−4.214	−1.608	−0.4369	0.8368	0.1308	−0.07903
$v_y$	−1.588	0.5103	−5.345	−5.837	0.1308	1.788	2.072
$v_z$	−1.8737	0.6241	−5.918	−10.19	−0.07903	2.072	3.116

Meteorites are seen in data from the KBOX (Boston, Massachusetts), KENX (Albany, New York), and possibly KOKX (New York City, New York) radars. The KBOX data are the most numerous, with signatures evident in four radar sweeps in the 18:03 UTC data suite. Altitudes range from 13,595 m down to 7,600 m above mean sea level (AMSL) for the KBOX data. Signatures are also evident in data from the TBOS airport weather radar operated by the Federal Aviation Administration, at 15,178 m and 19,452 m AMSL. Clouds in the area at the time of the fall are found at up to approximately 6,500 m AMSL, well below the signatures of falling meteorites, so the two sources are distinct.

The SIMONe-constrained terminus and weather-radar centroids produce a J2 terminus location of $41.89491^\circ$ N, $70.41082^\circ$ W. The azimuth of the bolide ground track and trajectory depression angle are taken from SIMONe data as $115.0^\circ$ and $72^\circ$, respectively. The SIMONe path and the weather-radar signatures are in good spatial agreement. Figure 7 shows the positions of the VHF radar echoes of meteor plasma, weather radar echoes of falling meteorites, and the path hypotheses of the falling meteorites based on the $J_2$ model.

The $J_2$ results indicate that the meteorite strewn field is centered approximately at $41.865^\circ$ N, $70.385^\circ$ W. The strewn field is relatively small despite the high terminus altitude because of light winds, spanning approximately 6 km east–west and 5 km north–south when masses down to 0.1 g are included. The six weather-radar-constrained landing solutions shown in Figure 8 have modeled masses from 3.2 g to 1.425 kg. This does not define the complete range of meteorite masses in the fall, because larger masses tend to descend rapidly and can escape detection by the slow-turning, mechanically driven WSR-88D radars. $J_2$ also calculates an extended strewn field using a hypothetical mass range from 0.1 g to 10 kg. Figure 8 overlays the weather-radar-constrained landing solutions and the extended hypothetical-mass calculation.

Assuming 10% survival, conditional on an iron composition, gives an illustrative total fall mass of 560 kg based on the NASA Meteoroid Environment Office pre-atmospheric mass estimate of 5600 kg ([41]). Integrating the radar-derived mass–number distribution instead gives 82 kg when a dielectric factor $|K|^2 = 1.00$ is assumed, with the caveat that this calculation includes only meteorites observed by radar and not larger meteorites that escaped detection. This radar-reflectivity method is unverified, and the iron-composition survival assumption is not a composition determination. These calculations are presented here only to show that total fall-mass estimates vary widely and require further method development. Using the conservative 82 kg estimate, the modeled ocean-surface density ranges from approximately 1,000 meteorites km$^{-2}$ for the approximately 2 g mass range to 29 meteorites km$^{-2}$ for 985 g meteorites. Water depth at the fall site is approximately 34 m, so horizontal translation during descent to the seafloor should be negligible compared with the size of the strewn field.

Orbit

Figure 9 shows the estimated orbit based on the VHF radar path. The estimated pre-Earth-encounter orbital elements are given in Table 2. The uncertainty is Monte-Carlo sampled from the initial state vector uncertainty covariance matrix. For comparison, the orbit plot shows a representative subset of 57 recovered-meteorite orbits with complete orbital elements from the meteorite orbits compilation of [39].

Quantity	Symbol	Unit	Value
Epoch		UTC	2026-05-30 00:53:16.260
Epoch		MJD	61190.03699375
Semimajor axis	$a$	AU	$2.44 \pm0.10$
Perihelion distance	$q$	AU	$0.6045 \pm0.0030$
Aphelion distance	$Q$	AU	$4.27 \pm0.20$
Eccentricity	$e$		$0.752 \pm0.010$
Inclination	$i$	deg	$19.21 \pm0.18$
Argument of perihelion	$\omega$	deg	$93.4 \pm0.6$
Longitude of ascending node	$\Omega$	deg	$68.99209 \pm0.00016$
True anomaly at epoch	$\nu$	deg	$85.9 \pm0.6$
Jupiter Tisserand parameter	$T_{J}$		$2.99 \pm0.08$
Corrected radiant right ascension	$\mathrm{RA}$	deg	$59.12 \pm0.34$
Corrected radiant declination	$\mathrm{Dec}$	deg	$45.85 \pm0.18$
Time at 100 km	$t_{100}$	UTC	2026-05-30 18:06:19.819
Geodetic latitude at 100 km	$\phi_{100}$	deg	$41.99$
Geodetic longitude at 100 km	$\lambda_{100}$	deg	$-70.69$
Entry azimuth at 100 km	$\mathrm{Az}_{100}$	deg	$111.4$
Entry elevation at 100 km	$\mathrm{El}_{100}$	deg	$-70.2$
Ground-relative speed at 100 km	$v_{100}$	$\mathrm{km\,s}^{-1}$	$26.00$
Terminus altitude	$h_{\mathrm{term}}$	km	$33.61$
Terminus geodetic latitude	$\phi_{\mathrm{term}}$	deg	$41.9105$
Terminus geodetic longitude	$\lambda_{\mathrm{term}}$	deg	$-70.4202$
Terminus azimuth, velocity at 100 km	$\mathrm{Az}_{\mathrm{term}}$	deg	$111.6$
Terminus elevation, velocity at 100 km	$\mathrm{El}_{\mathrm{term}}$	deg	$-70.0$

Infrasound Results

Infrasound detections were retained at five single-sensor stations located 358–1649 km from the source. In addition, two geographically and instrumentally independent multi-sensor systems resolved the signal direction. A representative retained arrival at SSPA is shown in Figure 10, illustrating the waveform and associated spectral energy. The spatial distribution of detections, ELFO, and the Raspberry Shake/Boom cluster is shown in Figure 11(a).

Analysis of the ELFO array data at a range of 894 km identified a coherent infrasonic wavetrain with a stable apparent trace velocity near 0.37 km/s and a back-azimuth of $101.4^\circ\pm5.3^\circ$ (Figure 12). The observed back-azimuth differs by $+6.6^\circ$ from the bearing to the CNEOS peak-brightness location and by $+6.0^\circ$ from the bearing to the VHF radar terminus. An independent cross-correlation delay solution gave a back-azimuth of $100.8^\circ$. The magnitude of the ELFO residual is within the range of directional deviations previously observed for bolide infrasound propagating over regional to long distances, where atmospheric winds and propagation structure can modify the apparent arrival direction ([57]).

The nearby citizen-operated Raspberry Shake/Boom cluster provided an independent short-range directional measurement. The six sensors gave a back-azimuth of $129.4^\circ\pm1.1^\circ$, differing by $+2.0^\circ$ from the bearing to the CNEOS peak-brightness location and by approximately $-1.0^\circ$ from the bearing to the VHF radar terminus. The cluster geometry and recovered source direction are shown in Figure 11(b). Although none of the six sensors constituted a statistically significant detection individually, their joint envelope-coherence measurement was highly significant, with the observed array statistic exceeding all 2000 randomized time-shift realizations and all 728 off-source windows ($p < 0.001$).

A first-order assessment of the source-region stratospheric winds placed all retained detections along downwind or crosswind propagation paths, with no upwind detections. The network-averaged dominant period for the five retained single-sensor detections and the ELFO beam was 2.92 ± 0.41 s. Applying the multi-station bolide period-yield relation of E. A. Silber et al. [62] gives an infrasonic energy estimate of 0.45 kt TNT equivalent. Using the five single-sensor detections alone gives a mean period of 3.02 s and a corresponding energy estimate of 0.51 kt TNT equivalent.

Discussion

Radar Signatures

Several aspects of the meteor-radar observations support relatively high mechanical strength of the meteoroid. Despite its large inferred diameter and mass (1.6 m and 5600 kg, estimated by NASA’s Meteoroid Environment Office [41]), the meteoroid produced a detectable head echo only below approximately 90 km. By comparison, the much smaller meteoroid observed by K. S. Obenberger et al. [45] with a similar meteor radar system was detectable at about 120 km. The entry speeds differed substantially: approximately 25 km s$^{-1}$ for the present event and 72 km s$^{-1}$ for the K. S. Obenberger et al. [45] event, which makes a direct comparison harder.

For fixed cross-sectional area and drag coefficient, the kinetic-energy loss rate scales as $\rho_a v_{\mathrm{air}}^3$. Equal loss rates therefore require an atmospheric density approximately 24 times larger at 25 km s$^{-1}$ than at 72 km s$^{-1}$. The MSIS atmospheric density ratio between 90 and 120 km is approximately 170, substantially larger than the factor required to compensate for the lower speed in this scaling. Entry speed and neutral density alone therefore do not explain the low detection altitude. Head-echo detectability also depends on composition-dependent ablation and the speed- and species-dependent efficiency of collisional ionization [74], as well as entry angle. A less volatile composition could contribute to the lower initial detection altitude for the head echo.

The micro-Doppler spectra (Figure 5) show no resolved secondary tracks indicative of fragmentation during the head-echo observation, in contrast to the clear fragmentation signatures reported by K. S. Obenberger et al. [45]. This again supports a high mechanical strength meteoroid.

The modeled dynamic pressure reached approximately 7 MPa near the radar terminus (Figure 7c), using $p = \rho_a v_{\mathrm{air}}^2$ following J. Borovička et al. [7]. This pressure exceeds the 0.9–5 MPa second-stage fragmentation pressures reported for ordinary-chondritic meteorite falls by J. Borovička et al. [7].

Together with the weather-radar evidence for surviving fragments, these observations support relatively high mechanical strength of the meteoroid and the presence of material capable of withstanding substantial aerodynamic loading.

The effective bistatic head-echo RCS is 100–1000 m$^2$ (Figure 4). These values greatly exceed the projected area of the solid meteoroid and are 100–1000 times larger than the largest RCS reported for faint head echoes observed with a MST radars with a similar frequency [55]. The large return may arise from plasma in the shock layer and near wake of the moving metre-class meteoroid. The high Doppler shift and agreement with the modeled range migration show that the dominant scattering plasma most likely is moving together with the meteoroid. At the upper end, the RCS corresponds to an equivalent circular diameter of 30–40 m, but this is not a physical plasma diameter: the estimate is conditional on the transmitted and received polarization, and the RCS depends on the electron-density distribution and scattering regime [38]. At 31.65 MHz, the cold-plasma critical electron density is approximately $1.2 \times10^{13}\ \mathrm{m}^{-3}$. Simulations of shock-generated plasma surrounding much slower moving hypersonic vehicles and their near wakes yield electron densities well above this threshold [20], supporting the possibility of an overdense shock-layer contribution to the head echo.

The head echo disappears at an altitude of approximately 72 km. The reason for its disappearance is not entirely clear. This altitude also coincides with the onset of a promptly generated non-specular trail echo, with a delay of less than 1 s. The localized trail positions agree, within their uncertainties, with the trajectory inferred independently from the tri-static head-echo ranges and Doppler shifts. Their short duration and subsonic Doppler motion favor scattering from transient plasma turbulence in the supersonic shock wake rather than from a long-lived, kilometre-scale specular trail. The relevant flow regime should be examined more carefully; in particular, the atmospheric mean free path and Knudsen number may ultimately provide a lower bound on the body size at the onset of shock-generated trail scattering.

Below approximately 70 km altitude, ordinary meteor-train plasma evolution is not governed by ambipolar diffusion alone. Ion–molecule chemistry can remove free electrons through dissociative recombination and attachment, particularly in dense, low-altitude trails [6, 4, 5]. These processes can shorten radio-echo lifetimes and bias diffusion coefficients inferred from echo decay times. The ion chemistry likely explains why the low-altitude trail echo is short lived. The observed prompt echoes can thus be interpreted as transient, shock-coupled wake plasma rather than as classical long-lived specular train scatter, although a quantitative plasma-chemistry model that incorporates the hypervelocity shock front and wake turbulence is still needed to reach further conclusions.

The lack of detectable deceleration is qualitatively consistent with a large meteoroid, but it does not provide a strong quantitative constraint in this case. A sweep over ballistic coefficients indicates that the smallest object that would avoid detectable deceleration within the observed head-echo segment would have a mass of only about 0.2 g. Since the actual object was clearly much larger than this, the non-detection of deceleration does not provide a useful lower limit on the meteoroid mass.

Weather radar and dark flight modeling

We identify the lowest retained VHF trail echo, at 33.61 km, as the radar terminus: the lower end of the path along which the entry produces plasma structures detectable by the VHF system. The J2 dark-flight model assumes that the disappearance of the echo beyond this point results from breakup into smaller fragments, which decelerate rapidly and are too small to sustain a sizeable shock-clad plasma structure. In the J2 analysis, the weather-radar meteorite centroids are back-propagated to a 3 km s$^{-1}$ transition speed, yielding a terminus at the same location. Although the radar terminus need not coincide with an optical endpoint or the time of maximum luminosity, the CNEOS peak-brightness altitude of 32.0 km (NASA/JPL Center for Near-Earth Object Studies 2026) lies only about 1.6 km below it.

The radar observations follow the event from the head echo near 90 km, through prompt shock-wake echoes down to 33.6 km, and then through hard-target meteorite signatures to 7.6 km above mean sea level. The weather-radar detections were identified independently of the meteor-radar observations, whereas the dark-flight model uses the meteor-radar trajectory and terminus as inputs. Agreement between the detections and the modeled dark flight therefore supports the reconstructed fall sequence, although it does not provide a fully independent validation of the meteor-radar trajectory. Together with the CNEOS optical detection and the infrasound observation, these measurements provide a consistent account of atmospheric entry and the subsequent fall of surviving fragments.

The modeled masses range from 3.2 g to 1.425 kg, with tens to hundreds of grams represented prominently; this is atypical compared with the more usual 1–100 g range. The total meteorite observation time is also anomalously short, only 2 min 2 s rather than the more typical 8–12 min, indicating that small meteorites are scarce in this fall. J2 also calculates the slope of log mass versus log $N$, where $N$ is the number of meteorites in a given signature estimated from total reflected radar energy. The slope of a linear fit to these data measures the prevalence of small meteorites relative to large ones and is a proxy for parent-body mechanical strength. The Cape Cod Bay fall exhibits the flattest slope seen to date, −1.024, indicating a mechanically robust body that survived the bolide with a large population of relatively large meteorites. The radar and dark-flight evidence are consistent with a dense, mechanically strong meteoroid, possibly iron-rich, but the composition remains to be confirmed by recovered material or independent compositional constraints.

Source Association

The Jopek–Williams sporadic-meteoroid classification criteria T. J. Jopek & I. P. Williams [37] give conflicting dynamical classifications. Of the 2000 sampled orbits, 63.3%, 100.0%, and 97.3% satisfy the cometary side of the $T_J-i$, $K-i$, and $P-i$ criteria, respectively, whereas 94.7% and 100.0% satisfy the asteroidal side of the $Q-i$ and $\varepsilon-i$ criteria. The Jupiter Tisserand parameter, $T_J = 2.99 \pm0.08$, lies near the classification boundary of 3. These fractions describe the sampled orbits relative to empirical boundaries and do not provide probabilities of cometary or asteroidal origin. The orbital elements fall within the range reported for recovered meteorites by P. Jenniskens & H. A. R. Devill epoix [31]; M. M. M. Meier [39]. The nominal semi-major axis exceeds those of 55 of the 75 meteorite orbits (73%) listed in Table 2 of P. Jenniskens & H. A. R. Devill epoix [31].

P. M. Shober et al. [56] identified numerous sporadic meteoroids on Jupiter-family-comet-like orbits whose atmospheric penetration indicated relatively durable material. Their strength classification used the empirical $P_E$ criterion, which accounts for luminous-endpoint atmospheric density, initial mass, entry speed, and entry angle. They proposed that these objects originated in the main asteroid belt and evolved onto comet-like orbits. Cape Cod Bay similarly combines an orbit near the cometary–asteroidal classification boundary with evidence for mechanically resistant material.

A catalogue-neighbour search against Minor Planet Center near-Earth-object and comet orbit catalogues G. V. Williams & T. B. Spahr [75]; Minor Planet Center [40] and established IAU Meteor Data Center showers [40, 32, 30] shows no convincing shower, comet, or near Earth asteroid association. The closest catalogue neighbours under the Jopek $D_{\mathrm{H}}$ orbital dissimilarity criterion T. J. Jopek [36] are near-Earth asteroids rather than comets. The closest near-Earth asteroid is 2024 XR16 with $D_{\mathrm{H}} = 0.093$, whereas the closest comet is P/2009 WX51 (Catalina) with $D_{\mathrm{H}} = 0.308$. The closest established shower, the omicron-Eridanids, has $D_{\mathrm{H}} = 0.212$ and a corrected-radiant separation of about $48^\circ$, making it a poor association to any known active stream.

Infrasound Source Interpretation

Bolides generate infrasound through the continuous ballistic shock associated with hypersonic flight and through localized energy release during fragmentation. Gross fragmentation can produce a strong, quasi-spherical shock from a confined portion of the trajectory, distinct from the cylindrical shock generated continuously along the flight path [50, 59, 67, 63]. As these disturbances decay into infrasound, atmospheric temperature and wind structure control refraction and ducting through tropospheric, stratospheric, and elevated acoustic waveguides over regional to long distances [17, 1, 57, 52].

The source region is constrained independently by radar, space-based optical observations, and infrasound. CNEOS places peak brightness at 32.0 km altitude, only 1.6 km below the 33.61 km VHF radar terminus, after which weather radar detected surviving fragments in dark flight. The Raspberry Shake/Boom back-azimuth differs by only about $1.0^\circ$ from the bearing to the VHF radar terminus and by $2.0^\circ$ from the bearing to the CNEOS peak-brightness location. This agreement associates the regional infrasound with the terminal portion of the luminous trajectory. A similar combination of seismoacoustic, weather-radar, and optical constraints was used to resolve trajectory and fragmentation for the 2025 Alaska fireball (L. T. Scamfier et al. 2026).

CNEOS coordinates, however, should not be treated as exact ground truth. Event-specific measurement uncertainties are not reported, and comparisons with independently constrained fireballs show event-to-event variability in the reported parameters [15, 47, 76]. The $2.0^\circ$ acoustic offset is therefore consistent with the CNEOS source location at the precision relevant here. Radar traces the meteoroid and its plasma wake to the end of the detectable entry path and subsequently detects surviving material in dark flight; the space-based observation identifies the strongest optical energy release at nearly the same altitude; and infrasound records the atmospheric shock radiated from that region. These independent measurements favor gross fragmentation near the end of luminous flight as the dominant source of the regional infrasound.

ELFO provides an independent directional constraint from a source-receiver distance of nearly 900 km. Back-azimuth deviations can arise from both source geometry and atmospheric refraction [58]. The Cape Cod Bay bolide trajectory is steep, for which geometry-induced directional offsets are expected to be substantially smaller than for shallow entries [58]. The $\sim6^\circ$ ELFO residual is therefore compatible with a contribution from atmospheric winds and propagation structure along the long path [17, 27, 3, 57] (C. D. de Groot-Hedlin et al. 2010); quantitative separation of source-geometry and propagation effects would require trajectory-dependent ray tracing.

The network-averaged dominant period corresponds to an infrasonic energy estimate of $\sim0.5$ kt using the multi-station bolide period-yield relation of E. A. Silber et al. [62]. The CNEOS estimate is 1.1 kt, and the difference lies within the established dispersion of empirical bolide period-yield scaling [19, 62]. The two estimates are based on independent observables: the CNEOS energy is inferred from optical radiated energy, whereas the infrasonic estimate is derived from the dominant acoustic period. Interpreted as pre-atmospheric kinetic energy at the radar-derived entry speed of 26.0 km/s, the 0.5 kt infrasonic estimate corresponds to an equivalent spherical diameter of $\sim1.5$ m for a chondritic body or $\sim1.1$ m for iron, using representative meteorite bulk densities [11].

The Raspberry Shake/Boom result extends previous use of spatially distributed single-sensor stations for bolide localization through multilateration [76]. In this case, inter-station envelope timing provides a statistically significant direct back-azimuth even though the individual citizen-operated sensors are inconclusive. Distributed pressure-sensor networks can therefore provide directional constraints without a purpose-built infrasound array when sufficient spatial coverage and timing coherence are available.

The same seismoacoustic principles apply to artificial atmospheric entries and space debris. For example, the Origins, Spectral Interpretation, Resource Identification, and Security-Regolith Explorer (OSIRIS-REx) sample return capsule provided a controlled hypersonic source with a known trajectory for testing infrasound source localization, signal evolution, and atmospheric propagation [61, 60]. Recent reconstructions of a Soyuz-2.1b, and Falcon 9 reentries further demonstrates that meteor-derived optical and seismoacoustic methods can be applied to space-debris trajectories [25, 28]. These applications link natural bolide monitoring with characterization of artificial reentries and other unplanned hypersonic atmospheric events.

Conclusions

The combined observations demonstrate that meteor-radar plasma tracking can bridge the upper-atmospheric trajectory and the onset of meteorite dark flight even when precise optical meteor camera observations are not available. Together with weather-radar and infrasound measurements, this permits reconstruction of the pre-impact orbit, terminal fragmentation region, plasma production, and subsequent fall.

The weather-radar-inferred fragment mass distribution, the modeled aerodynamic pressure of approximately 7 MPa near the radar terminus, and the absence of resolved fragmentation signatures in the head-echo micro-Doppler spectra together support a mechanically strong meteoroid.

The reconstructed orbit lies near the cometary–asteroidal classification boundary, with no convincing association with a catalogued parent body or established meteor shower. This combination of orbital characteristics and material durability is consistent with the population of sporadic meteoroids on Jupiter-family-comet-like orbits described by P. M. Shober et al. [56], who argued that most originated as main-belt asteroidal fragments whose orbits subsequently evolved into this regime.

The measured trajectory and height-dependent echo strengths and lifetimes, together with the infrasound energy estimate of approximately 0.5 kt TNT equivalent and evidence for a source near the radar terminus, provide constraints for modeling the hypersonic shock flow, its plasma wake, and acoustic radiation from the flow and terminal fragmentation. Such models should explain why the head echo disappears near 72 km while prompt, short-lived non-specular trail echoes remain detectable down to 33.6 km, and test whether shock-generated ionization, plasma chemistry, and wake turbulence can account for their rapid growth and decay.

Data Availability

The source radar data set is archived on Zenodo as the May 30th 2026 New England Bolide Radar Dataset R. Volz & J. Vierinen [73]. Infrasound waveform data used in this study were accessed through the International Federation of Digital Seismograph Networks (FDSN) and are freely available through its data services. We also provide a Zenodo repository with raw infrasound data (DOI: 10.5281/zenodo.22732819). The repository will be made public upon acceptance.

Acknowledgments

EAS and SA were supported by the Laboratory Directed Research and Development (LDRD) program (project number 236896) at Sandia National Laboratories, a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

JLC and MC acknowledge support from the European Office of AFOSR 717 (EOARD) grant FA8655-23-1-7017.

RV, PJE, FDL, JS, and AW acknowledge support for MIT Haystack Observatory participation under the NSF grants AGS-2504079 and AGS-1933007.

The authors acknowledge support from Colin Lonsdale, Will Rogers, Aleks Pop Stefanija, Chris McKenney, Steve Holmberg (MIT Haystack Observatory) and George Hassel (Siena University) for radar implementation, technical assistance, and hosting of nodes for the SIMONe Haystack multistatic meteor radar system.

Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC (NTESS), a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration (DOE/NNSA) under contract DE-NA0003525. This article has been authored by an employee of National Technology & Engineering Solutions of Sandia, LLC under Contract No. DE-NA0003525 with the U.S. Department of Energy (DOE). The employee owns all right, title and interest in and to the article and is solely responsible for its contents. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this article or allow others to do so, for United States Government purposes. The DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan https://www.energy.gov/downloads/doe-public-access-plan. This paper describes objective technical results and analysis. Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the U.S. Department of Energy or the United States Government. OpenAI Codex (https://openai.com/codex/) assisted with manuscript editing and the development of data analysis and figure-generation software.