Orbital fit

The orbital solution was estimated through a least-squares fit52 to the available optical and radar astrometry. We corrected optical astrometry to remove star catalogue biases53 and weighted data according to a statistical analysis of past performance for the different observatories54. Our own Dk154 and radar observations were weighted based on the individual measurement uncertainty. The parameters estimated from the fit are the cometary orbital elements and the A2 parameter used to model non-gravitational perturbations as a transverse acceleration A2g(rH) where g provides the functional dependence on the heliocentric distance rH (ref. 55). In this paper, we set g(rH) = (1 au/rH)2, which is the typical choice to model the Yarkovsky effect28.

Optical observations

In early September 2025, 1998 SH2 was imaged during routine observations by the ATLAS survey29. Typical ATLAS survey exposures are sidereally tracked and 30-s long, using either the ATLAS o (~r + i) or c (~ g + r) filter, with 4 exposures taken over a 30-min interval at the same footprint on the sky. The ATLAS projected pixel scale is 1.86″ on sky.

The Dk154 observations were obtained with the Danish Faint Object Spectrograph and Camera (DFOSC)56, using either the Cousins (Bessell) R filter or no filter. The detector is an E2V231-42, with a pixel scale of 0.396″ on sky. The telescope was tracked at a half of the apparent rate of the observed target, resulting in same trailing of field stars and the target.

The CFHT observations were obtained with the MegaPrime instrument57, using the ‘gri.MP9605’ filter. While MegaPrime is a mosaic of 40 charge-coupled devices (CCDs), the detector used was ‘ccd23’, a Marconi/EEV CCD with a 0.187″ pixel scale.

The VLT observations were obtained with the Focal Reducer/low dispersion Spectrograph 2 (FORS2)58, using the clear filter, optimized for throughput. FORS2 is equipped with a 1 × 2 mosaic of CCDs. We used ‘chip 1’, an MIT/LL detector, with a pixel 0.126″ pixel read binned 2 × 2 resulting in a 0.252″ on sky. The observation circumstances are listed in Table 1.

Image processing

The data were processed using standard electronic bias subtraction and flat-fielding with averaged twilight or dome exposures. Images from the Dk154 and VLT were aligned with subpixel accuracy using the centroids of tens of background stars. Stars were identified using Source Extractor59 in the SEP implementation60 and cross-matched with Astroalign61, resulting in the accurate offset between frames. The star-aligned frames were then stacked using a sigma-clipping median rejection, creating a deep star background reference image. The object was rejected by the median due to its motion and is thus absent from this reference. In the case of CFHT, as only three frames were available, a master background stack could not be produced. The stars and background objects were therefore manually masked in the three frames.

The ATLAS exposures were reduced using its standard image reduction pipeline29. Transient sources were detected by subtracting the ATLAS o-band template image, and the asteroid was detected by linking catalogue detections using the ATLAS Moving Object Processing System. Eight of the subtracted images from 3 and 4 September 2025 were registered to a pixel scale of 2″ per pixel and further stacked using a ‘weighted median’ (50% quantile of cumulative weights) to search for a signal of cometary activity.

The ephemeris of 1998 SH2 was retrieved from Jet Propulsion Laboratory (JPL)’s Horizons62 for each frame’s epoch. Using the previous inter-frame offsets and the reference image’s astrometric calibration, the necessary offsets to align the images on 1998 SH2 were computed. The star-aligned reference image was also subtracted from the individual frames. This removed the extended signal (for example, stellar point spread function (PSF) wings and galaxies), leaving only small residuals near the star cores due to minor seeing variations, subpixel misalignment and VLT diffraction pattern rotation. These residual frames were then shifted and stacked. The sigma-clipping median rejected background star residuals, cosmic rays and blemishes, resulting in deep stacks showing the object on a clean, empty background (Fig. 3).

Cometary activity

The surface-brightness profile of the object was computed using the final stacks from Fig. 3, and compared with the profile of the PSF, evaluated by averaging several well-exposed field stars in the corresponding master background stack.

The flux was integrated in a series of circular annuli centred on either the object or the star template. The error was taken as the standard deviation of the individual pixels within each annulus. For the PSF profile of trailed images, we only used pixels in angular regions perpendicular to the star’s trailing direction. For the object’s profile, we rejected pixels within 5° of the tail’s position angle to avoid contamination. The PSF profile was normalized so that its peak flux matched that of the object. The resulting profiles are shown in Extended Data Fig. 4.

1998 SH2 shows a clear flux excess over the PSF profile, extending beyond 10″. Quantitatively, this excess represented 0.06 ± 0.01, 0.210 ± 0.008 and 0.241 ± 0.008 mag for the Dk154, CFHT and VLT observations, respectively. The error bars seem small in absolute terms, but one must keep in mind that this is a relative measurement to the stellar profile. The total duration of each observation is unlikely to average out the possible rotational variability of the object, which could be in the 0.2–0.4 range63 for such a small object. An intrinsic variation of the nucleus of 0.2 mag would cause a change of the (unchanged) coma contribution of ±0.04 mag. The overall increase from 0.06 to 0.2 is therefore likely significant, but the variation from 0.21 to 0.24 could be caused by the nucleus rotation.

At radii r beyond the seeing disk, the excess flux follows a linear trend in the log–log plot, corresponding to a surface-brightness profile that evolves as r−n. A linear regression performed over 2 ≤ r ≤ 4 half-width at half-maximum yielded exponents n = 4.3, 3.8 and 3.6 for Dk154, CFHT and VLT, respectively. These values are considerably steeper than the n = 1 expected for a steady-state, isotropic coma31 or the n = 1.5 maximum expected when including the effect of solar radiation pressure32. Such steep slopes can be caused by the sublimation of icy grains over time, or by an increase in activity. Given that the fraction of extended flux relative to the total flux was possibly increasing with time, we favour the interpretation of increasing activity, with the rate of increase becoming slower with time.

In summary, the object was active on 13, 17 and 30 September 2025, and its activity level was possibly increasing throughout this period.

Analysis of the tail

The 1998 SH2 tail morphology was analysed using the Finson–Probstein method64, which models the motion of dust grains under the effects of solar gravity and solar radiation pressure. This analysis generates families of synchrones, connecting particles emitted at the same time, and syndynes, connecting particles with the same β, the ratio of the radiation pressure to the solar gravity. On the images, the synchrones appear as radial lines whose position angle (PA) is related to time of dust emission. To visualize and analyse these, the images were transformed to polar coordinates, shown in Extended Data Fig. 2, in which the nucleus appears as the broad bright band at low radii, and the tail as an horizontal feature.

The Dk154 image indicates that the dust in the tail was released between 30 August and 7 September 2025. The shallower CFHT data suggest a release window between 28 August and 7 September. Owing to Earth’s position nearly in 1998 SH2’s orbital plane during the VLT observations, the synchrones nearly collapse into a single line: at PA ≈ 65° for dust emitted after 9 September, and at PA ≈ 244° for pre-7 September emissions (matching the PA of the projected negative velocity vector). The entire observed tail lies along this position angle. In summary, the tail in each image corresponds to the same continuous activity event that took place between 30 August and 7 September 2025.

Although the Dk154 image had worse seeing, the closer proximity of 1998 SH2 to Earth provided the highest-spatial-resolution look at the object. Critically, the tail reveals additional information via the Finson–Probstein syndynes. The β value is related to the grain radius a (m) and density ρ (kg m−3) via:

$$\beta =5.74\times 1{0}^{-4}\frac{Q}{\rho a},$$

(1)

where Q ≈ 1 is the radiation pressure efficiency, which depends on the grain material64. While ρ can vary widely—from 1,000 kg m−3 (a traditional cometary value) to 1,900 ± 1,100 kg m−3 from in situ measurements65 on comet 67P, and up to 3,000 kg m−3 for S-type asteroids66—we adopt ρ = 2,000 kg m−3 for this analysis. Extended Data Fig. 3 shows a subset of the Dk154 image from Extended Data Fig. 2, with β and the corresponding radius a labelling the plotted syndynes. The PAs of the synchrones are also marked. To characterize the tail’s shape, a Gaussian profile was fitted to the tail at various distances from the nucleus, and its central PA and FWHM are marked as green symbols.

The peak of the tail is confined within the 0.0005 ≤ β ≤ 0.0010 range, corresponding to emission times from 29 August until 7 September 2025. The tail is too faint and diffuse to measure its position before 29 August with this method, and it is too close to the nucleus and lost in its glare after 7 September, but there is no indication of an abrupt change. The measured range of β corresponds to very large grains, on the order of 400 μm. Cometary dust grains typically follow a power-law size distribution67 with an index of about −4, and up to an upper limit amax. The fact that no grains appear below the β = 0.0005 syndyne suggests an upper size limit of amax ≈600 μm. As grain brightness scales as a2, the observed grains follow a brightness power-law distribution with an index of about −2. This implies that grains smaller than ~400 μm should be more numerous and brighter, which is not the case. This discrepancy strongly suggests that the actual grain size distribution is narrowly limited to the 300–600 μm range.

From the gas production rate computed above, \({Q}_{{{\rm{H}}}_{2}{\rm{O}}}\approx 1.2\times 1{0}^{24}\) molecules per second, we estimate the largest dust grain that can be lifted from the nucleus. For that critical radius, the gas drag equals the weight of the grain. Assuming a grain density of 1,000 kg m−3 and a nucleus density of 500–1,000 kg m−3, the critical radius can be estimated68 as a ≈ 1.6 mm, confirming that the large grains observed can easily be lifted by the gas. Alternatively, assuming that the largest grains observed, with a ≈ 600 μm, correspond to the critical radius, the density of the nucleus would be ~1,300 kg m−3. We can therefore use this value as an upper limit to the nucleus density.

While typical cometary dust is in the micrometre range, very large grains (up to centimetere scale) have been detected. Such large grains were observed remotely, for instance69,70, near comet C/2001 A2 or in situ near comet 67P. Laboratory simulations suggest that these large particles are ejected when ice sublimation occurs below the surface, leading to a buildup of pressure that explosively expels the material30. A fast rotation could also contribute to the ejection of large particles, for example, similar to one of the hypotheses formulated for 133P/Elst–Pizarro71. Notwithstanding the origin of the grains, they were continuously released from 29 August until 7 September 2025, with no indication of an abrupt start or stop at either these dates.

Radar

Radar observations of 1998 SH2 occurred at Goldstone (8,560 MHz, 3.5 cm) on 26 August and 2 September 2025, dates that straddled the closest approach within 0.02 au on 31 August, when the asteroid was too far south for Goldstone to track. The 26 August observations did not produce a detection but observations on 2 September were successful. The radar observations used standard data acquisition and reduction techniques72,73. At the time of the observations, problems with one of the klystron amplifiers limited the transmitter power to 240 kW, or slightly more than one-half of the nominal value.

On 26 August, we estimated that signal-to-noise ratios (SNRs) would be strong enough to obtain an echo within a few minutes. Given the diameter, and the fact that nearly all near-Earth asteroids (NEAs) >0.15 km in diameter have rotation periods slower than 2.1 hours, we expected an echo bandwidth of less than about 20 Hz. After observing for about 20 minutes without detecting an echo, we checked different frequency resolutions in case the echo was much narrower or wider than expected. After 40 minutes, there was still no echo, so we abandoned 1998 SH2 and observed a different asteroid. Earlier during the observing session on 26 August, we detected radar echoes from 1997 QK1, and after we stopped the 1998 SH2 observations, we also detected echoes from 2025 QX4, so we knew that the radar system was functioning well and suspected that the pointing was off for 1998 SH2, which was later confirmed after the observations on 26 August concluded.

We began on 2 September with continuous-wave observations and saw an echo within 2 minutes. The echo has a bandwidth of 7 Hz and is centred on the Doppler frequency predicted by the ephemeris (Extended Data Fig. 1). We then transmitted coded waveforms with time delay resolutions of 10 μs, 11 μs and 1 μs (distance resolutions of 1,500 m, 1,650 m and 150 m) to estimate the range. The entire sequence of Goldstone radar observations spanned about 67 min and is summarized in Extended Data Table 1.

On the basis of infrared data obtained by the NEOWISE mission, the diameter and albedo of 1998 SH2 are estimated24 as of 380 ± 60 m and 0.058 ± 0.024. The width of a radar echo is given by:

$$B=\frac{4{\rm{\pi }}D\cos (\delta )}{\lambda P},$$

(2)

where B is the bandwidth or Doppler broadening of the echo, D is the diameter, δ is the subradar latitude, λ is the wavelength and P is the rotation period. If the rotation period is known, then equation (2) constrains the pole-on extent of the asteroid. For 1998 SH2, a rotation period has not been reported but the bandwidth, diameter and equation (2) allow us to estimate the period. Given the bandwidth of 7 Hz and a diameter of 380 m, equation (2) places an upper bound on the rotation period of 5.4 h under the assumption that the diameter is correct and that 1998 SH2 is not considerably elongated.

The echo in Extended Data Fig. 1 shows a dip at frequencies near the middle of the echo that is consistent with a concavity, but due to the relatively weak SNRs, the dip is also consistent with receiver noise. The narrow spikes also resemble echoes seen from satellites of binary systems, where the broad echo is from the primary and the narrow echo is from the secondary. To check, we processed the continuous-wave data at different frequency resolutions but did not find convincing evidence for a companion. We also checked the 1 μs ranging data by processing it at four different frequency resolutions and summing all the runs. Radar observations of binary NEAs observed previously at this delay resolution often show echoes from two separate objects. The summed images at 1.0 μs × 0.5 Hz resolution show a small number of pixels with SNRs ~3.5 in delay-Doppler locations expected for an object in orbit relative to the main echo, but the pixels are also consistent with noise (we expect ~30 noise pixels this strong), so the evidence for a satellite, although intriguing, is not convincing.

We searched for rotational variations in the bandwidths and spectral shapes by summing groups of 5 runs (spanning about 5 min each) processed at 0.5 Hz resolution. We did not see any variations that are statistically significant, so evidently the bandwidth did not change significantly over an interval of 1.06 h.

For 1998 SH2, we estimate a circular polarization ratio, that is, the ratio of the echo power in the same circular (SC) polarization state to that in the opposite circular (OC) polarization state, of SC/OC = 0.09 ± 0.03, which is lower than the average of ~0.3 seen for hundreds of other NEAs observed with radar74,75. This ratio is also lower than those observed for (433) Eros (0.28 ± 0.06), (25143) Itokawa (0.27 ± 0.04), (4179) Toutatis (0.29 ± 0.01), (101955) Bennu (0.18 ± 0.01) and (65803) Didymos (0.20 ± 0.02), which have been visited by spacecraft. The circular polarization ratio of 1998 SH2 could indicate that the near-surface is less rugged at decimetre spatial scales than the surfaces of the asteroids imaged by missions. However, modelling results76,77 indicate that surface texture and composition also strongly influence circular polarization ratios so roughness is not the only possibility. The low ratio of 1998 SH2 is inconsistent with those seen for V-, E- and some X-class NEAs (SC/OC > 0.6). The ratio is consistent with values estimated for the other spectral types, particularly a small sample of M types, and also with the lower end of the distribution for optically dark BC types and bright SQ types. The circular polarization ratio has been measured for 8 comets78,79,80,81,82 and ranges between 0.105 and 0.59; 1998 SH2’s value is lower than this range.

Radar echoes from some comets show a wide ‘skirt’ caused by centimetre- to decimetre-sized coma particles surrounding the nucleus. Coma echoes have been seen in radar echoes of numerous comets that were very active (for example, C/1996 B2 (Hyakutake)), but are not always detected from some comets that show a coma at optical wavelengths. We searched but do not see a wide coma echo for 1998 SH2, which is consistent with the low level of activity observed in the optical images. Thus, we conclude that cometary activity by 1998 SH2 on 2 September was too low to detect with radar observations at Goldstone.

The 1-μs echo occupies 2 rows, and given that the radar could illuminate only ~1/2 of the surface if the object were a sphere, this establishes that the diameter of 1998 SH2 is <600 m, a result that is consistent with the value of 380 m from NEOWISE24. The diameter of 380 m also indicates that 1998 SH2 is the smallest comet ever observed by radar. We used echo power spectra processed at 0.5-Hz resolution (Extended Data Fig. 1) to estimate a radar cross-section of 0.0048 km2 ± 35%, where the uncertainty accounts for systematic pointing and calibration errors. If we adopt the diameter of 380 m, then we obtain a radar albedo of ~0.04, which is lower than most observed among NEAs but overlaps many estimated for comet nuclei80. The radar albedo is a function of the near-surface bulk density80,83,84 and a value of 0.04 suggests a surface with porosity that is not highly compacted. The implication is that the radar albedo more closely resembles those seen from comet nuclei than from NEAs.

Meteoroid stream

Given that 1998 SH2 closely approaches Earth, the possibility of a meteoroid stream giving rise to a meteor shower does exist. Assuming any meteoroid ejection occurs at relatively low velocity, a meteoroid stream would be expected to move in parallel to 1998 SH2. On 30 August 2025, when the orbit of 1998 SH2 is 0.02 au from Earth, any potential shower would appear to originate from the geocentric radiant αg ≈ 172.3°, δg ≈ −0.1°, with a geocentric speed vg ≈ 17.2 km s−1. However, as this date corresponds to the start of the current activity, any meteoroids that may have been observed must have been released during previous activity. Also, while ~400 μm meteoroids would produce optical meteors, this radiant is close to the helion direction, and thus, probably would be only visible by meteor radars. An in-depth simulation to better model any potential meteoroid stream is beyond the scope of our work here.