Main

Utopia Planitia, the largest impact basin in the northern hemisphere of Mars1, is considered to be a Late Hesperian lowland unit2 (Fig. 1a). The northern lowlands were largely filled with materials of the Vastitas Borealis Formation (VBF)3 as a sublimation residue from frozen ponded bodies of water4 and subsequently modified by Amazonian resurfacing5, such as long-term weathering, aeolian deposition6,7 and impact remixing8. A large number of orbital and in situ geomorphometry measurements show that polygonal terrain9,10 (Fig. 1c–f) and other periglacial features11,12 are extensively distributed in southern and western Utopia Planitia, indicating the occurrence of water-related or ice-related activities13,14. Viking 2, a previous ground-based probe in northern Utopia Planitia (Fig. 1a), identified troughs that probably form a polygonal network15. In 2021, two rovers, Perseverance and Zhurong (Fig. 1a and Extended Data Fig. 1), landed on Mars almost simultaneously16,17. Both rovers are equipped with ground-penetrating radar (GPR), operating at a frequency range of 150–1,200 MHz for Perseverance and 15–95 MHz (low-frequency channel) and 450–2,000 MHz (high-frequency channel) for Zhurong18,19. These can detect, for the first time, the high-resolution subsurface structures of Jezero crater20 and southern Utopia Planitia21,22, respectively. As an important complement to orbital radar explorations23,24, in situ GPR surveying can provide critical local details of shallow structures and composition within approximately 100 m depth along a rover traverse.

Fig. 1: Zhurong rover landing site and images of polygonal terrain in Utopia Planitia.
figure 1

a, Topographic map of Utopia Planitia, showing the landing sites of the Zhurong rover, the Viking 2 lander and the Perseverance rover. The −4 km elevation contour is shown. Four local regions (c–f) with polygonal terrain are marked with white squares. b, The Zhurong rover traverse from Sol 11 through Sol 113 (HiRISE image: ESP_073225_2055). Green segments denote the wedges of buried polygons recognized from Fig. 2 (P1–P16). Purple segments denote the interiors of the polygons. c–f, Four representative HiRISE images of polygons in Utopia Planitia whose locations are marked in a: PSP_002202_2250 (c), PSP_006962_2215 (d), PSP_002162_2260 (e) and PSP_003177_2275 (f). Note the range of spatial scales for the sizes of the polygons. The average diameters of polygons shown in c–f are calculated in Extended Data Fig. 6. Credit for HiRISE images: NASA/JPL/University of Arizona.

Full size image

The Zhurong landing site is thought to be one of the best places for detecting ground ice at low-to-mid latitudes on Mars25. The GPR onboard Zhurong rover, thus, provides an unprecedented opportunity to illuminate subsurface structures and to investigate geological processes, particularly those associated with ancient or current water-related activities in southern Utopia Planitia. Subsurface layering in Utopia basin of Mars has been revealed by the radar of the Zhurong rover21,22, indicating the presence of sedimentation due to episodic hydraulic flooding that is interpreted to represent the basin infilling of Utopia Planitia during the Late Hesperian to Amazonian. However, previous works mainly focused on the vertically layered subsurface structure and less attention has been paid to lateral variations along the Zhurong radar profile.

Subsurface features potentially revealed in lateral variations are equally critical compared with the vertical layered structures for discovering the geological evolution of Mars. However, in the presence of strong scattering effects, preliminary attempts to extract features from the lateral variation of the Zhurong GPR profile proved unsuccessful. In an effort to unveil potential characteristics of the lateral variation of the subsurface surrounding the landing site, we conducted a comprehensive time–frequency analysis (Methods) of the Zhurong GPR data (Extended Data Figs. 2 and 3). We identified 16 nearly vertical bands dominated by anomalous low-frequency components at depths of 35–65 m along the rover traverse (Fig. 1b), which probably formed on ancient Mars and were buried by later geological processes.

Palaeo-polygons detected by the GPR on the Zhurong rover

Figure 2a shows the frequency distribution in the depth domain after random noise attenuation26 and time-to-depth conversion using the velocity model of ref. 21 (Methods). According to the features in the frequency variation with increasing depth, the subsurface structure can be divided into three layers: (1) the first layer (0–35 m) has uniformly distributed energy, indicating relatively homogeneous media; (2) the second layer (35–65 m) has a series of vertical bands with anomalous low-frequency components, indicating strong lateral variations; and (3) the third layer (65–80 m) is dominated by strong random noise, where the frequency increases anomalously and precludes further interpretation. The most notable feature in the GPR profile is the alternating occurrence of high- and low-frequency bands within the depth range 35–65 m in the second layer (Fig. 2a). The dominant frequency of these low-frequency bands (∼45 MHz) is ∼7 MHz lower than that of the background (∼52 MHz) (Extended Data Fig. 3). Along the 1.2-km-long rover traverse, as many as 16 such low-frequency bands were identified (Figs. 1b and 2b).

Fig. 2: Time-varying average frequency distribution of the GPR data revealing buried polygons.
figure 2

a, Time-varying average frequency of the denoised GPR profile using the f–x regularized nonstationary autoregression method26. Green segments (P1–P16) denote the wedges of buried polygons. b, Stacked amplitude spectrum (amp. spec.; black curve) produced by summing the time-varying average frequency along each trace for depths of 35–65 m, which is roughly the depth range between the tops and bottoms of the observed polygons (black arrows in a). Green rectangles marked at both local minima of the stacked amplitude spectrum and the lower frequencies in a indicate the polygon wedges. c, Sketch map as an example of what the buried polygonal terrain could look like from an aerial view. The rover traverse might be parallel to, perpendicular to or intersecting with the wedges of different polygons (Fig. 1b). d, Sketch map showing the definitions of polygon features. e, Histogram of polygon diameter recognized from b. f, Comparison of polygon size between Earth and Mars (literature references provided in Extended Data Fig. 7). Some data are presented as mean values ± standard deviation (s.d.). The Zhurong landing site data are from the measurements made in this study.

Source data

Full size image

We conducted a series of analyses to make sure that these low-frequency bands were not artefacts (Methods). First, the time-varying average frequency of the original GPR data without random noise attenuation (Extended Data Fig. 2c) shows similar low-frequency bands. Second, different denoising methods (Extended Data Fig. 2d, ref. 27) and segmentation methods (Extended Data Fig. 4) show consistent low-frequency bands (Fig. 2a), suggesting that none of these low-frequency bands is an artefact from improper data processing. Additionally, the low-frequency bands generally start from a depth of ∼35 m, not from the surface, indicating that they are associated with underground structures rather than surface objects. Furthermore, the positions of these low-frequency bands along the rover traverse (green segments in Fig. 1b) do not show any correlation with the distribution of dunes or rocks on the surface (Extended Data Fig. 1), suggesting that the low-frequency bands are not caused by surface-related features. Therefore, we can confirm that the low-frequency bands faithfully reflect the lateral variations of subsurface structures. Considering that these anomalous structures systematically occur every few tens of metres and are nearly vertical in orientation (Fig. 2), we interpret them as the infilled wedges between columns of a polygonal terrain buried under ∼35 m of overlying materials (Fig. 3).

Fig. 3: Schematic model of the polygonal terrain formation process at the Zhurong landing site.
figure 3

a, The origination of thermal contraction cracking on the surface. b, The formation of cracks infilled by water ice or soil material, causing three types of polygonal terrain (ice-wedge, composite-wedge and sand-wedge polygons). c, The stabilization of the surface polygonal terrain in the Late Hesperian–Early Amazonian, possibly with the cessation of an ancient wet environment. d, The palaeo-polygonal terrain, either with or without being eroded, was subsequently buried by deposition of the covering materials in the Amazonian. The Mars surface image was acquired by the Navigation and Terrain Camera (NaTeCam).

Full size image

Tens of giant troughs have been identified near the Zhurong landing site28. These troughs are part of the polygonal trough system in southern Utopia Planitia29. Compared to polygonal terrain that is in the form of a network, isolated troughs usually exhibit linear shapes. Several isolated troughs have been observed with widths >100 m around the Zhurong landing site (Extended Data Fig. 5). However, no polygonal terrain has been identified from surface observations or orbital imagery (Fig. 1b) within several kilometres of the Zhurong landing site (Extended Data Fig. 1). Thus, the buried polygons observed here from the GPR profile exclusively represent a palaeo-polygonal terrain. The average polygon diameter extracted from the GPR profile (Fig. 2c,d) is ∼67 m (Fig. 2e), which is within the typical range of previously reported Martian polygons (Fig. 2f) and is comparable to that of the surface polygons observed in western Utopia Planitia (Fig. 1c–f) within the latitudinal range between 40° and 50° (Methods and Extended Data Figs. 6 and 7). Considering that the direction of the rover track could be randomly oriented either perpendicular, parallel or oblique to the orientation of the polygon wedges (Extended Data Fig. 8), the average apparent width of the polygon wedges (∼27 m) is regarded as an overestimate so that the actual average width should be narrower. The average height of the polygon wedges (the absolute elevation difference between the bottom of a wedge and the shoulder of a polygon) is ∼30 m, corresponding to a polygon diameter/wedge height ratio of ∼2.2, which is generally consistent with the theoretical polygon diameter/wedge height ratio (∼3.0) of polygonal terrain30. The materials within the polygon wedges, possibly unconsolidated soil-rock mixtures, are more likely to absorb or scatter high-frequency components of radar waves, thus producing local low-frequency anomalies. In contrast, the polygon interiors are probably composed of well-consolidated materials, thus the high-frequency components can be well retained (Fig. 2a), leading to a weaker attenuation of radar waves.

Possible origin of the buried polygons

Polygonal terrain has been reported mainly in cold regions on Earth and mid-to-high latitudes on Mars9,11,31,32. Previously, Martian polygonal terrain has been observed only on the surface, mainly distributed at latitudes >30° (Fig. 1c–f; ref. 33), with diameters ranging from centimetres to kilometres9,34. Large Martian polygons (usually kilometre-scale) are widespread in the northern lowlands of Mars35,36. They were potentially caused by contraction jointing from lava cooling, contraction cracking from clay desiccation, thermal contraction, tectonic fracturing or the coalescence of smaller polygons9,37,38. In contrast, small-scale polygonal terrain (centimetres to tens of metres) was first found in situ by Viking 2, where the polygonal diameter near the lander was <10 m (ref. 31).

For polygons with diameters from centimetres to tens of metres, possible formation mechanisms31,39,40 may include contraction from desiccation of wet sediments producing mud-cracks, contraction from cooling lava producing columnar jointing, faulting creating a jointing system in rock and thermal contraction cracking. Polygonal cracks of desiccation-induced contraction are dominated by the evaporation of water in the soil, usually with a ratio of the crack width to the polygon diameter of <0.1 (refs. 41,42), which is much smaller than 0.4, the ratio of the wedge width (∼27 m) to the polygon diameter (∼67 m) detected in this paper. Consequently, desiccation as a contraction mechanism can be ruled out. In addition, if the buried polygons were caused by the contraction of cooling lava, the reflections from the interfaces between the lava flow(s) and underlying and overlying sediments should be notable due to a strong dielectric contrast43. However, no such strong reflection interfaces were observed in the low-frequency GPR data (Extended Data Fig. 2a) or in the SHARAD data (Extended Data Fig. 9). In addition, there is no evidence for the presence of basaltic extrusions in the Zhurong landing area, suggesting that volcanic columnar jointing is an unlikely explanation. Faulting-generated jointing systems are typically long and linear in shape40. Moreover, the low-frequency bands appear intermittently over relatively short segments of the rover path (such as P10 to P11, P13 to P14, and P14 to P15 in Fig. 1b), instead suggesting a polygonal terrain. Additionally, the lengths and widths of jointing systems due to faulting are usually kilometres in scale, but the SHARAD profile across the Zhurong landing site (Extended Data Fig. 9) does not show any evident reflection in this region. Consequently, fault jointing as a cracking mechanism can also be ruled out. Therefore, by a process of elimination, the buried polygons are interpreted to have most likely formed by thermal contraction cracking. The cracks generated in the ground may be infilled by water or soil material, causing three types of polygonal terrain (ice-wedge, composite-wedge and sand-wedge polygons44, Fig. 3b). Ice-wedge polygons usually develop in permafrost, with ice infilling the wedges45. The ice in a wedge could sublimate and local gravel, sand and clay particles could then partially fill in the wedge46,47, so that composite-wedge polygons form. Sand wedges usually form in cold regions from initial thermal contraction with subsequent aeolian deposition in the wedges48. We next consider the implication of the proposed explanation.

Geological age constraints indicate that the previously reported polygons at the surface of Utopia Planitia mainly formed in the Hesperian3,31. Near the latitude of the Zhurong landing site, polygons possibly formed in the Hesperian and were then covered by materials from the Late Hesperian to Amazonian plains1,2. At the Zhurong landing site, the material at depths of 30–80 m could have formed in the Late Hesperian–Early Amazonian, consistent with crater-counting ages estimated over various spatial ranges in southern Utopia Planitia5,28,29,49. In addition, the dielectric permittivity in this depth range is like that of materials of the VBF, indicating that this layer may represent an upper portion of the VBF deposits21. As the wedges of the polygons occur at depths of 35–65 m (Fig. 2a), the buried polygonal terrain probably developed from the sedimentary materials of the VBF, under dramatic changes in surface temperature on early Mars50. Polygonal terrain is distributed on the surface of present-day Mars mainly in high-latitude regions (generally >30°; ref. 31), whereas the buried polygonal terrain detected by the Zhurong rover occurs at low-to-mid latitudes (25° N; Fig. 3d). This latitudinal contrast may indicate that the Zhurong landing site had a cold environment that is found only at high latitudes on present Mars, but in the Late Hesperian–Early Amazonian, allowing for the formation of the palaeo-polygonal terrain at low-to-mid latitudes.

Implications for the palaeoclimatic conditions on Mars

The above-mentioned formation mechanism for the buried palaeo-polygonal terrain requires a cold environment and might be related to water/ice freeze–thaw processes in southern Utopia Planitia on early Mars. The detected buried polygons, which indicate that freezing occurred at low-to-mid latitudes, require strong palaeoclimatic variability, potentially due to the higher obliquity than today51. The possible presence of water and ice required for the freeze–thaw process in the wedges (Fig. 3a) may have come from cryogenic suction-induced moisture migration from an underground aquifer on Mars52,53,54, snowfall from the air55 or vapour diffusion for pore ice deposition.

The contrast in the lateral frequency-variation patterns above and below ∼35 m depth (Fig. 2a) suggests that for the polygons to form and become buried, there was a critical transformation of aqueous activity or thermal conditions in the Late Hesperian–Early Amazonian. This stark environmental transition at ∼35 m, thus, may indicate both the cessation of an ancient wet environment (Fig. 3c) as well as that unknown notable geological events occurred after the formation of the polygonal terrain56. The depositional thickness and the age of the present surface materials at the Zhurong landing site could be roughly estimated by a geological survey of this region5. However, the role of erosion in the area is difficult to constrain. The continued acquisition of in situ data by the Zhurong rover will help better constrain the local dynamics of deposition and erosion. The tops of the polygons are at different depths (Fig. 2a), and smooth lateral changes in depth from the top of one polygon to the next exhibit broad peaks and valleys that may imply erosion before they were buried. Whether the buried polygonal terrain experienced subsequent erosion or not, the ∼35-m-thick overlying materials provide a new constraint for estimating the deposition rate in southern Utopia Planitia.

Lateral variations in the subsurface structure at the Zhurong landing site provide evidence of a buried palaeo-polygonal terrain that formed in the Late Hesperian–Early Amazonian from periglacial processes. Occurring at low latitudes (∼25° N), the polygonal terrain, which is interpreted as having most likely formed by thermal contraction cracking, makes a compelling case for the high obliquity of early Mars. The subsurface structure with the covering materials overlying the buried palaeo-polygonal terrain suggests that there was a notable palaeoclimatic transformation some time thereafter.

Methods

Time–frequency analysis and time-varying average frequency

GPR is an ideal instrument for exploring subsurface structures on Earth and extraterrestrial bodies. Electromagnetic waves are emitted on the surface and reflections are received from subsurface interfaces where the dielectric permittivity or conductivity changes. Although rover-based GPR has a limited detection range and penetrating depth, it is an effective tool for detecting shallow subsurface structures and has been successfully applied to both the near and far sides of the Moon57,58. The GPR data employed in this study are the low-frequency channel data with a frequency range of 15–95 MHz, which can penetrate a depth of 80 m below the Martian surface. The local time–frequency decomposition method is an effective form of time–frequency analysis. It has a higher temporal resolution than the widely used short-time Fourier transform method and S-transform method59. The main idea of local time–frequency decomposition is to use a Fourier basis to match nonstationary signals by solving a regularized least-squares minimization problem. A casual nonstationary signal f(t), t ∈[0, L], can be expressed as a Fourier series as follows:

$$f({\rm{t}})=\mathop{\sum }\limits_{{\rm{n}}=-\infty }^{\infty }{A}_{n}(t){\psi }_{n}({\rm{t}}),$$
(1)

where An(t) are the Fourier coefficients and \({\psi }_{n}({\rm{t}})\,{={\rm{e}}}^{i(2\pi nt/L)}\). We can obtain An(t) by solving the least-squares minimization problem:

$$\mathop{\min }\limits_{{{\rm{A}}}_{n}}{\Big\Vert\, f(t)-\sum _{n}{A}_{n}(t){\psi }_{n}({\rm{t}})\Big\Vert }_{2}^{2}.$$
(2)

However, the minimization problem is ill-posed because it is severely underconstrained. To solve this problem, a regularization term is needed. After adding a regularization operator R, the formal solution \({\tilde{A}}_{n}(t)\) is given by:

$${\tilde{A}}_{n}(t)=\mathop{\min }\limits_{{{\rm{A}}}_{n}}{\Big\Vert\, f(t)-\sum _{n}{A}_{n}(t){\psi }_{n}({\rm{t}})\Big\Vert }_{2}^{2}+{\rm{R}}.$$
(3)

The absolute value of \({\tilde{A}}_{n}(t)\) is the time–frequency representation of the signal f(t), which we refer to as the time–frequency map. Additionally, the time–frequency map can be converted to a time-varying average frequency according to

$$f_\mathrm{a}(t)=\frac{\int f{F}^{2}(\;f,t)\,\mathrm{d}f}{\int {F}^{\,2}(\;f,t)\,\mathrm{d}f},$$
(4)

where fa(t) is the time-varying average frequency, F(f,t) is the time–frequency map, and f and t are the frequency and time, respectively. The time-varying average frequency can show well the spatial-temporal distribution characteristics of the main frequency components and, thus, is widely used to extract subsurface attributes60,61,62.

Potential causes of the anomalous low-frequency bands

To ensure that the identification of the anomalous low-frequency bands (Fig. 2a) was robust, we consider all their potential causes in this section. There are three possibilities for a horizontal discontinuity of the time-varying average frequency: (1) numerical artefacts caused by improper data processing, (2) an energy change in the GPR profile caused by surface anomalies, such as undulating terrain or surface rocks and (3) subsurface high-frequency-absorbing materials with an uneven transverse distribution. To avoid potential numerical artefacts and maintain the original proportion of the GPR profile energy in the horizontal direction during data processing, we used the same data processing methods, such as decoding, denoising and amplitude compensation, for all GPR data. The f–x regularized nonstationary autoregression method26 and the streaming orthogonal prediction filter method27 are both commonly used denoising methods in exploration seismic data processing, as they are effective in suppressing random noise and preserving weak signals. Therefore, item (1) should not be the case.

Furthermore, the travel path of the Zhurong rover was generally flat in terms of topography63. Within the first kilometre of the rover traverse, the fluctuation (within the local 3 m area covered by the rover) was no more than 0.1 m (Extended Data Fig. 1; ref. 64). Moreover, the engineering team guiding the rover tried to avoid rocks, grooves, pits and other terrain during path planning, so the impact of surface rocks and undulating terrain was mostly eliminated. The images taken by the Navigation and Terrain Camera (NaTeCam)65 show no evident variation in the terrain along the rover path except for several relatively bright white dune structures. This analysis shows that the horizontal discontinuity of the GPR profile was negligibly affected by data processing and surface factors, so that it faithfully reflects the high-frequency attenuation or absorbing effects of subsurface materials.

Determination of the polygon diameters

To determine the polygon diameters in the GPR profile (Fig. 2a), the recognition process was based on the width of the anomalous low-frequency bands as follows. Step 1: According to Fig. 2a, the depth range of these anomalous low-frequency bands is roughly 35–65 m. Thus, the frequency values within this depth range were stacked and smoothed to obtain a stacked amplitude spectrum by summing the time-varying average frequency (Fig. 2a). Step 2: Local minima in the stacked amplitude spectrum curve were identified. The left and right boundaries of the low-frequency bands were determined using the frequency distribution diagram in Fig. 2a. Step 3: The polygon diameter and the width of the wedge between two adjacent polygons were calculated using the left and right boundaries of the low-frequency bands. The diameter of a polygon was defined as the distance between the middle positions of two adjacent low-frequency bands (Fig. 2b), and the width of the wedge between the polygons is the width of a low-frequency band.

For the High Resolution Imaging Science Experiment (HiRISE) images (Extended Data Fig. 6), the polygons were recognized using the following five steps. Step 1: Calibrate the scale of the HiRISE images. Step 2: Identify the boundaries of polygons in the HiRISE images. Step 3: Mark the polygons using the imaging processing technology of the crack network to quantify crack patterns66,67. Step 4: Calculate the average diameter (unit: pixel) by averaging the maximum and minimum Feret diameters68 of each polygon cell. Step 5: Calculate the normal distribution statistics for the diameters of all polygons to obtain their mean value and standard deviation (Extended Data Fig. 6).