A Catalogue of Impact Craters and Surface Age Analysis in the Chang’e-6 Landing Area

Abstract

Chang’e-6 (CE-6) is the first sample-return mission from the lunar farside and will be launched in May of 2024. The landing area is in the south of the Apollo basin inside the South Pole Aitken basin. Statistics and analyses of impact craters in the landing area are essential to support safe landing and geologic studies. In particular, the crater size–frequency distribution information of the landing area is critical to understanding the provenance of the CE-6 lunar samples to be returned and can be used to verify and refine the lunar chronology model by combining with the radioisotope ages of the relevant samples. In this research, a digital orthophoto map (DOM) mosaic with resolution of 3 m/pixel of the CE-6 landing area was generated from the 743 Narrow Angle Camera of the Lunar Reconnaissance Orbiter Camera. Based on the DOM, craters were extracted by an automated method and checked manually. A total of 770,731 craters were extracted in the whole area of 246 km × 135 km, 511,484 craters of which were within the mare area. Systematic analyses of the crater distribution, completeness, spatial density, and depth-to-diameter ratio were conducted. Geologic model age estimation was carried out in the mare area that was divided into three geologic units according to the TiO₂ abundance. The result showed that the east part of the mare had the oldest model age of $\mu {3.27}_{-0.045}^{+0.036}$ Ga, and the middle part of the mare had the youngest model age of $\mu {2.49}_{-0.073}^{+0.072}$ Ga. The crater catalogue and the surface model age analysis results were used to support topographic and geologic analyses of the pre-selected landing area of the CE-6 mission before the launch and will contribute to further scientific researches after the lunar samples are returned to Earth.

1. Introduction

Lunar samples have been returned to Earth by six Apollo missions (Apollo 11, 12, 14, 15, 16, and 17), three Luna missions (Luna 16, 20, and 24), and the Chang’e-5 mission. The analysis results of these samples are invaluable to our fundamental understanding of lunar origin and evolution [1,2,3,4]. Based on the lunar sample radiometric ages, the Moon also provides the foundation of crater size–frequency distribution chronologies for the solar system [5,6,7,8]. However, all of the samples were returned from the lunar nearside, and the samples from the farside, especially from the lunar South Pole Aitken (SPA) basin, are crucial to address other fundamental questions such as the nature of the Moon’s asymmetry, the composition and structure of the lunar mantle, etc. The Chang’e-6 (CE-6) mission is scheduled to launch in May of 2024. The pre-selected landing area of CE-6 is in the south of the Apollo basin, a ~490 km diameter crater, within the SPA basin [9,10]. It is the first sample-return mission to the lunar farside and is expected to return the farside basalts.

Impact craters are one of the most prominent surface features of the lunar surface, and they are important in studying the subsurface structure of the Moon [11], tracing the sample source area [12], analyzing the impact flux of the meteorites [8], etc. For comprehensive studies of the impact craters, many global or regional crater catalogues have been established [13,14,15,16]. The completeness diameter of the current global crater databases is too large for the CE-6 landing area (e.g., 1 km in [13]) in carrying out scientific analysis on the lunar samples. Zeng et al. mapped 26,785 craters using the 7 m/pixel resolution Chang’e-2 (CE-2) digital orthophoto map (DOM) in the pre-selected CE-6 landing area [10]. However, their work mainly focused on studying the geologic background of the CE-6 landing area, with no particular analysis on the mapped craters such as the completeness diameter that would greatly hinder its further application. Therefore, a detailed study of the impact craters in the CE-6 landing area using higher-resolution images is still necessary.

In this research, we utilized a seamless DOM mosaic based on the Lunar Reconnaissance Orbiter Camera (LROC) Narrow Angle Camera (NAC) images and SLDEM2015 (a lunar digital elevation model generated by co-registration and combining SELENE TC digital elevation model with LRO laser altimetric data) [17], to obtain a crater catalogue in the pre-selected CE-6 landing area. The resolution of the mosaic was 3.0 m/pixel, and the original images were in suitable illumination conditions. As a result, a more reliable and accurate crater catalogue could be obtained. In addition, with the SLDEM2015, the depth of craters was also automatically measured. The obtained crater catalogue and the analysis results can contribute to the studying of the returned lunar samples and other important issues such as the lunar Late Heavy Bombardment. Combined with the radioisotope ages of the samples, the crater size–frequency distribution information can be used to verify and refine the lunar chronology model, which is of fundamental importance for lunar and planetary sciences.

2. Data and Methods

2.1. LROC NAC Images and SLDEM2015

A high-resolution DOM mosaic of the landing area was generated using 743 LROC NAC images as a base map to extract and measure craters in the CE-6 landing area (Figure 1). The LROC NAC images covered most of the lunar surface with a horizontal resolution of 0.5–2 m and were the highest-resolution orbiter images currently available in the pre-selected landing area [18]. A method for large-area DOM generation was proposed and used to produce the high-resolution orthophoto mosaic of the CE-5 landing area [19] and was also used in this research. According to this method, 743 NAC images were selected first, and the Planetary Data System (PDS) experiment data record (EDR) files of them were downloaded from the PDS Geosciences Node Lunar Orbital Data Explorer website. Next, the SPICE kernels were attached, and radiometric corrections were performed for each image using the Integrated System for Imagers and Spectrometers (ISIS) software (v7.2.0). The planar block adjustment was then utilized to improve the relative positional consistencies of the LROC NAC images of the sub-areas to subpixel level and simultaneously align the NAC images to the control source (SLDEM2015). Subsequently, TPS (Thin Plate Spline)-based image registration was used to reduce geometric inconsistencies between the adjacent sub-area DOM mosaics, ensuring that the final DOM was seamlessly mosaicked over the entire area. Finally, a geometrically seamless DOM mosaic was produced, covering 149°–160° west longitude and 39.7°–44.1° south latitude with a resolution of 3 m/pixel. The detailed mapping method can be found in [19].

SLDM2015 covered latitudes within ±60° with a horizontal resolution of approximately 60 m in the equator area and a vertical accuracy of 3–4 m. It can be downloaded at PDS LOLA data node (http://imbrium.mit.edu/EXTRAS/SLDEM2015/, accessed on 10 May 2023). In this research, SLDM2015 provided necessary elevation information when generating the DOM mosaic and measuring crater depths in the CE-6 landing area.

2.2. Crater Extraction and Mapping in CE-6 Landing Area

In this research, the impact craters within the CE-6 landing area were extracted using a deep-learning-based automated method followed by manual inspection. Our method was based on the YOLOv8 model (https://github.com/ultralytics/ultralytics, accessed on 15 June 2023), which combines object classification and localization in the same network, directly regressing the position and class information of the object in an end-to-end manner [20]. Our improved YOLOv8-LCNET (YOLOv8-lunar crater net) integrated more pyramid levels to extract craters with a wide range of diameter spans. Comparing with the former versions of YOLO, our improved YOLOv8-LCNET demonstrated higher recall rate in the lunar crater extraction task. As the focus of this paper was the analysis of the extracted craters of the landing area, the technical details of YOLOv8-LCNET will not be elaborated. A comparison of the performance of our improved method with the original method can be found in Appendix A. The training of the network and validation of the results are briefly described below.

Within the CE-6 landing area, a local mosaic DOM map with a resolution better than one meter was produced, based on which the crater sample dataset was established using the same approach of automatic extraction followed by manual inspection. The local DOM mosaic was cropped to sub images of 512 × 512 pixels to form the dataset, which contained 66,030 impact crater samples with a range in diameter from 3.7 m to 609 m. In addition, in order to improve the generalization of YOLOv8-LCNET, random data augmentation techniques such as flipping, rotation, adding noise, and image enhancement were applied to generate a dataset comprising 209,627 sample images. The dataset was split into training and validation datasets in the ratio of 8:2 for model training and performance evaluation.

To assess the rim accuracy of the impact craters automatically extracted by YOLOv8-LCNET, the diameters of the extracted craters and the manually confirmed true values were compared (Figure 2). As shown in Figure 2, an optimal straight line fitted from the scattering points plotted by the diameters automatically measured from the extracted craters and the true diameters had a slope of 1.034 and an intercept of 0.411 in meters. Additionally, the mean relative error (MRE) of the diameters was calculated to be 0.1042 (i.e., 10.42% diameter difference). These findings indicated that the automatically extracted impact crater diameters were consistent with the true diameters. This consistency verified the agreement between automatic extraction and manual extraction methods, thus demonstrating that the YOLOv8-LCNET algorithm could replace most of the manual efforts in impact crater extraction. It is worth mentioning that for manual crater mapping, different workers would generate different results, with up to 10% diameter differences [21]. In this research, the approach of automatic crater extraction and manual checking proved to be practical and time-saving.

Note that the rims of the craters were first extracted and then transformed into local projection coordinate systems, ensuring that the crater measurements were not affected by map projection. After the automatic extraction by YOLOv8-LCNET, manual inspection work was conducted to erase some false detections and add missing detections from the DOM mosaic with an ArcGIS extension named “CraterTools” [22]. All crater extraction results were manually examined using this tool, especially for impact craters larger than 200 m for dating purposes. Due to significant differences in the solar azimuth and elevation angles of some images forming the DOM mosaic, the manual examination was mainly needed in regions with extreme solar elevation angles or where large craters were in two images with contrary azimuth angles.

To calculate the crater rim-to-floor depth, an automatic tool developed by [23] was utilized. For each crater, profiles crossing the crater center were generated every 22.5° to form a total number of 8. Then a local maximum of the topographic curvature in each profile was found within 10% of the diameter on both sides of the crater circle. The resultant 16 rim points were used to represent the crater rims in order to obtain an averaged rim-to-floor depth of the crater. Note that in some cases such as craters intersecting with other craters, the interfered rim points were eliminated automatically. If fewer than 6 rim points were left, manually added rim points were needed. More details about the automatic crater rim-to-floor tool can be found in [23]. As the resolution of SLDEM2015 constrained small crater rim-to-floor depth calculation, craters larger than 8 pixels were selected for this automatic depth measurement [15,24].

In order to carry out surface dating, secondary craters needed to be removed from the dataset. Based on the appearance of clusters and chains around a primary crater with ejecta patterns, secondary craters were recognized and removed manually.

2.3. Surface Dating with Craters

Absolute ages can provide vital information in establishing the geologic history of the Moon, and currently, the widely used surface dating method is based on the crater size–frequency distribution (CSFD). The method was elaborated by [5] and was the subject of many papers (e.g., [25,26]). The rationale of the method was to fit the observed CSFD of a counting area to a known crater production function (PF) (e.g., [27,28]), which was further used to derive the absolute age along with a chronology function (CF) calibrated to radiometric dating of the returned lunar samples [5]. The detailed steps of the method can be summarized as follows: (1) obtaining the CSFD of a counting area belonging to the same geologic unit, (2) comparing it with a known crater production function (PF) [27,28], and (3) deriving the absolute age according to a CF that has been calibrated to radiometric dating of the returned lunar samples [5]. In practical work, these calculations are often completed with the aid of CraterStats software (v2.0) [28].

The titanium dioxide (TiO₂) content of lunar mare basalt represents an important indicator in magma evolution and is widely used to classify lunar basalt types [29]. According to the TiO₂ abundance derived from LROC WAC imageries by [30] and the basalt classification schema [29], three areas (i.e., A1, A2, and A3) within the basalt unit [31] for the CE-6 potential landing and sampling region inside the Apollo basin were outlined for crater dating in this study (shown in Figure 3), with the TiO₂ abundances of them being 4.6 ± 1.0%, 6.2 ± 1.3%, and 3.8 ± 0.8%, respectively.

3. Results

3.1. Catalogue of Craters in CE-6 Landing Area

In this 246 km ∗ 135 km landing area, a total number of 770,731 craters were extracted using the approach introduced in Section 2.2. According to the properties of the area surface, the whole area could be divided into mare and highland. The completeness diameter size was calculated in the mare area and is shown in Figure 3.

The completeness value of a crater dataset means that all the craters with diameters larger than this value are included [33]. Previous studies developed three different approaches of estimating the crater diameter completeness [33,34,35]. Salamunićcar et al. considered the completeness diameter as the inflection point of the slope in the cumulative CSFD curve of impact craters [34]. Robbins and Hynek adopted the value of the completeness diameter from the diameter of the bin one size larger than the bin with the highest number of impact craters in the incremental CSFD [35]. According to Robbins [33], the value was estimated by the Nth diameter bin larger than where the derivative of the slope was zero in an incremental CSFD.

Figure 4a,b illustrate the incremental and cumulative CSFD, respectively, of the craters in the mare area. In Figure 4a, the diameter of the crater cluster with the maximum number is approximately 25 m, yielding a completeness diameter of approximately 30 m according to the definition of Robbins and Hynek [35]. In Figure 4b, the inflection point of the slope is approximately 20 m. In Figure 4c, the value given by Robbins’ method is approximately 170 m [33]. Robbins’ value was larger compared with the other two, which may have been due to crater saturation equilibrium in the mare region. As the resolution of the mosaic map was 3 m/pixel, the completeness analysis result illustrated by Figure 4a,b indicates that the craters with diameters no less than 10 pixels were extracted. This dense crater catalogue can provide rich information for multiple applications such as landing safety evaluation, crater formation mechanisms, and impact gardening effects, etc. [36].

In this crater catalogue of the whole area, 38,471 possible secondary crater clusters were identified and eliminated by the easily applicable criteria of irregular shape, clustered distribution, and auxiliary information of shallow depth. In the original crater catalogue, there were 511,484 craters with diameters larger than 30 m, 15,575 craters with diameters larger than 200 m, and 456 craters with diameters larger than 1 km. The mapping results showing craters with diameters ≥200 m are illustrated in Figure 5. After secondary crater elimination, there were 484,410 craters with diameters larger than 30 m, 14,142 craters with diameters larger than 200 m, and 415 craters with diameters larger than 1 km.

Spatial density analysis was conducted on the catalogue after exclusion of secondary craters and on two subsets of greater than 30 m and greater than 1 km, respectively. As shown in Figure 6a,b, the spatial density of impact craters ranging from 30 m to 1 km was greater in the mare region at the center of the landing area than in the highlands. For craters with diameters greater than 1 km, they were more commonly found in the highlands than in the mare. This was because some of the small impact craters in the highlands were expunged due to topographic relief, while the large impact craters were preserved.

The estimation of crater depth was based on SLDEM2015, which is described in Section 2.2. Considering the grid size of SLDEM2015, only craters with diameters larger than 480 m were considered for depth estimation. Craters with depths larger than 12 m were utilized for subsequent analysis, considering the vertical accuracy of SLDEM2015 to be 3–4 m. A total of 1547 craters were ultimately measured for depth, with depths ranging from 12.1 to 2583.8 m. The majority of craters (94.2%) had depths less than 200 m. Six craters were deeper than 1 km, and nearly all of these craters were located in the geological units of the highlands. Figure 7 shows the crater depth distribution with a depth interval of 50 m.

In this work, the depth-to-diameter ratio was used to assess the morphology of craters. The depth-to-diameter ratios of a total number of 1,547 craters were calculated in the mare and highlands (as shown in Figure 3). The statistical results of the depth-to-diameter ratios are shown in Figure 8. As shown in Figure 8a, the depth-to-diameter ratio in the mare region was higher than that of the highland, which was because of the younger age of the mare, as well as the crater size difference illustrated in Figure 6. In the highlands, there were 1132 craters with diameters ranging from 480 to 19,882 m and depths from 12.1 to 2583.8 m, resulting in an average depth-to-diameter ratio of 0.058. In the mare area, there were 415 craters with diameters ranging from 481 to 6179 m and depths from 12.3 to 313.2 m, resulting in an average depth-to-diameter ratio of 0.071. Although the average depth-to-diameter ratios differed, the ranges of them were almost the same in both regions as shown in Figure 8b. Qualitative morphological analyses were also carried out, some examples can be found in Appendix B.

3.2. Surface Model Ages of CE-6 Landing Area

By inputting impact crater extraction results into the CraterStats software, the absolute model ages (AMAs) of the A1, A2, and A3 basalt units were $\mu {2.62}_{-0.095}^{+0.093}$ Ga, $\mu {2.49}_{-0.073}^{+0.072}$ Ga, and $\mu {3.27}_{-0.045}^{+0.036}$ Ga, respectively (Figure 9a–c). The AMAs of A1 and A2 were within each other’s uncertainty range, suggesting there might be either two basalt flooding events or only one flooding event, but the surface of A1 was masked/contaminated by materials with low TiO₂ content (e.g., ejecta) that resulted in a decrease in TiO₂ abundance. If the latter was the case, dating on the A1 + A2 unit would indicate the flooding happened around $\mu {2.58}_{-0.064}^{+0.063}$ Ga (Figure 9d).

4. Discussion

4.1. Comparison with the Craters in CE-5 Landing Area

The completeness analysis of the mare region using three different methods provided varying values, which is introduced in Section 3.2. The value provided by Robbins’ method was much larger than the other two methods. This was because the method adopted the production function, which took the crater saturation equilibrium into account [33]. From Figure 9a–c, the three basalt units, covering most of the mare region, also showed that the craters were essentially saturated at diameters greater than 100 m.

As described in Section 3.2, the depth–diameter ratios of the mare and highland areas in the CE-6 landing area were 0.071 and 0.058, respectively, which coincided with the conclusion from [39] that small craters (less than ∼4 km in diameter) on the lunar maria were found to be deeper than those on the highlands. The highland area in the CE-6 landing area contained 415 craters larger than 1 km, which was almost twice the amount of the CE-5 landing area; therefore, the overall depth–diameter ratio distributions showed smaller values than that of the CE-5 landing area.

4.2. Dating Results Compared with Previous Study

Ref. [10] also carried out crater dating on the basalt units of the CE-6 landing area, with approximately similar dating areas of A2 and A3. They used the classic chronology system, i.e., [5]. If the chronology system from [5] was used for dating in this study, the AMAs of the A1, A2, A3, and A1 + A2 basalt units were $\mu {2.51}_{-0.094}^{+0.092}$ Ga, $\mu {2.38}_{-0.07}^{+0.07}$ Ga, $\mu {3.25}_{-0.056}^{+0.043}$ Ga, and $\mu {2.46}_{-0.062}^{+0.062}$ Ga, respectively (Figure 10a–c). Compared with their dating results (i.e., ${2.40}_{-0.11}^{+0.11}$ Ga for the F_A unit and ${3.43}_{-0.061}^{+0.044}$ for the L_A unit), the AMA of the A2 unit was about the same in both studies, while the AMA of the A3 unit estimated in this study was younger than that of [10]. The difference might be because the dating regions and data sources for crater counting were not completely consistent. Overall, as we mapped craters based on higher-resolution DOM and used more craters in dating the surface units after eliminating secondary craters, our surface dating results should be more reliable.

5. Conclusions

We established a catalogue of 770,731 craters with the completeness diameter of 30 m in the CE-6 landing area based on an LROC NAC DOM mosaic with resolution of 3 m/pixel. The craters were extracted by our improved YOLOv8-LCNET deep-learning algorithm and examined manually. The morphology and distribution of all the craters and the depth analysis of craters larger than 480 m were included in this catalogue. The crater size-frequencies of different surface units delineated by TiO₂ abundance were analyzed to obtain the absolute model ages. The results showed that the oldest unit lay in the right of the mare and had a model age of $\mu {3.27}_{-0.045}^{+0.036}$ Ga, and the youngest unit was the middle part of the mare, which had a model age of $\mu {2.49}_{-0.073}^{+0.072}$ Ga. The established crater catalogue was used to support topographic and geologic analyses of the pre-selected landing area of the CE-6 mission for landing safety evaluation and scientific research before the launch. The catalogue and the derived surface ages have multiple potential applications after the lunar samples are returned. This catalogue and the CSFD analysis results will be helpful for the interpretation of the samples and understanding their provenances, the lunar chronology model refinement when combined with the radioisotope ages of the relevant samples, and better understanding of the geologic properties of the landing site, the Apollo basin, and the SPA basin.