Volume 49, Issue 16 e2022GL100783
Research Letter
Free Access

Present Day Endogenic and Exogenic Activity on Mercury

E. J. Speyerer

Corresponding Author

E. J. Speyerer

School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA

Correspondence to:

E. J. Speyerer,

[email protected]

Contribution: Conceptualization, Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration

Search for more papers by this author
M. S. Robinson

M. S. Robinson

School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA

Contribution: Conceptualization, Methodology, Formal analysis, Writing - review & editing, Supervision, Project administration, Funding acquisition

Search for more papers by this author
A. J. Sonke

A. J. Sonke

School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA

Contribution: Formal analysis, ​Investigation, Resources, Data curation, Visualization, Writing - original draft, Writing - review & editing

Search for more papers by this author
First published: 14 August 2022
Citations: 1


Exogenic and endogenic activities dictate how planetary surfaces evolve. However, the present-day influence of each of these activities is not well constrained. MESSENGER (MErcury Surface, Space ENvironment, GEochemistry, and Ranging) temporal imaging recorded surface changes associated with these activities between 2011 and 2015. Here we present the results of a temporal investigation that identified 20 reflectance changes. One change exhibited bright rays extending from the site, consistent with the aftermath of an impact event. If all changes result from impact events, the present flux rate is 1,000 times higher than models predict. Consequently, we also report changes on slopes in areas with concentrations of tectonic landforms and identify a subset of changes that are on or adjacent to hollows, consistent with present-day endogenic activities. Therefore, we conclude that these observations captured endogenic and exogenic modifications to the surface.

Key Points

  • Temporal imaging acquired over 4 years reveals 20 surface changes on Mercury providing evidence of present-day surface activity

  • One surface change exhibited ray-like features consistent with surface modification from an impact event

  • Other changes occurred on steep slopes near tectonic landforms or adjacent to hollows consistent with ongoing endogenic processes

Plain Language Summary

We use before and after (temporal) image pairs collected by the Mercury Dual Imaging System to identify 20 surface changes on Mercury that formed between 2011 and 2015. We identified at least one change likely resulting from a newly formed impact crater with bright rays that extend away from the site. If all the changes result from impact events, then the present-day rate of impactors striking the innermost planet is 1,000 times higher than models predict. Therefore, we investigate other sources for these detected changes. We located several changes on steep slopes near tectonic landforms, consistent with ongoing tectonic activity. Additionally, we identified several changes in areas adjacent to hollow formations, consistent with present-day activity. These detected changes will be critical targets for the upcoming BepiColombo mission. This work also provides a framework for detecting future changes using cross-mission image comparisons to constrain present-day surface processes further.

1 Introduction

The MESSENGER (MErcury Surface, Space ENvironment, GEochemistry, and Ranging) mission (Solomon & Anderson, 2018; Solomon et al., 2007) provided the first global look at the innermost planet enabling the identification of geologic units at the 100–200 m scale. A fundamental aspect of interpreting a geologic unit or individual surface feature is understanding its formation age and post-emplacement modification (Banks et al., 2015; Blewett et al., 2011; Braden & Robinson, 2013; Ernst et al., 2010; Fassett et al., 2012; Ostrach et al., 2015; Prockter et al., 2010; Watters et al., 2009). While the accumulation and measured density of impact craters are used to evaluate the surface age of geologic units (Le Feuvre & Wieczorek, 2011; Marchi et al., 2009; Michael & Neukum, 2010; Neukum, 1984; Neukum et al., 2001; P. D. Spudis, 1985; P. D. Spudis & Guest, 1988; Strom & Neukum, 1988), the formation rate of impact craters on Mercury is the least constrained of all the inner planets (Fassett, 2016).

Furthermore, the surface of Mercury is modified from endogenic activities, and the present-day rate and development of these processes are also not well constrained. For example, small, crisp scarp structures are consistent with ongoing tectonic activity in the last 50 Myr (Watters et al., 2016). Additionally, clusters of small enigmatic hollows with their characteristic shallow, irregular contours cover about 0.1% of the surface imaged at pixel scales <180 m/pixel (Thomas et al., 2014) and typically lack superimposed craters supporting a young age and possible continued development (Blewett et al., 20112013; Thomas et al., 2014; Xiao et al., 2013). In this study, we use temporal images collected by the Mercury Dual Imaging System (MDIS) (Hawkins et al., 2007) to identify surface changes during the MESSENGER mission to gain insight into the present-day endogenic and exogenic activity.

2 Materials and Methods

MESSENGER was launched on 3 August 2004 and was the first spacecraft to enter orbit around Mercury (Solomon et al., 2007). It remained in orbit from 18 March 2011 to 30 April 2015 and provided close-up and synoptic views of the innermost planet in the Solar System. The spacecraft was equipped with the MDIS, which included a monochrome Narrow-Angle Camera (NAC) and multispectral Wide-Angle Camera (WAC) (Hawkins et al., 2007). From an altitude of 200 km, the cameras had a ground sampling distance of 5.1 and 35.8 m for the NAC and WAC, respectively. However, the image pixel scale ranges widely across the surface due to the elliptical nature of the orbit (200 × 9,300 km). Additional information about the camera specifications can be found in Hawkins et al. (2007), and background into the orbit is documented in Solomon & Anderson (2018).

The MDIS instrument acquired over 270,000 images of Mercury during the orbital phase. For temporal image analysis, we selected a small subset of images using the following criteria:
  • Incidence angle <75°

  • Emission angle <30°

  • Smear magnitude <2.0 pixels

  • Horizontal pixel scale <1 km

  • Overlapping image footprints

  • Time difference between observations >30 days

  • Same Sun direction (east vs. west)

  • Incidence angle difference between observations <4°

  • Phase angle difference between observations <6°

  • Pixel scale of the before image within 25% of the after image

  • Same filter (if applicable)

We identified 58,552 time-separated observation pairs (temporal pairs) with similar photometric angles well suited for detecting any changes in the intervening time (minimizing the possibility of false positives due to shadow changes or poor image registration). The average time between temporal observations was 303 Earth days with a standard deviation of 228 Earth days. The resulting spatial distribution of the temporal pairs was not random (Figure S1a in Supporting Information S1). Therefore, we calculated the time each point on the surface was observed and did not double count areas observed simultaneously by two or more temporal image pairs (Figure S1b in Supporting Information S1). This removed biases introduced by the dense overlap in some regions (up to 700 overlapping temporal image pairs).

Additionally, since the time between temporal observations varies, we normalized the search area by time to calculate the rate of surface change. For example, a 1 km2 area imaged a year apart is equivalent to a 0.25 km2 area imaged 4 years apart. Normalizing the surface coverage to one Earth year results in an area of 6.27 × 107 km2 (Figure S2 in Supporting Information S1). This metric is key in determining the annual rate of activity (see Section 4).

Each image pair was processed using the calibration routines included in Integrated Software for Imagers and Spectrometers (ISIS) (Anderson et al., 2004). Once calibrated, we aligned the images into the same reference frame and co-registered the pair within a pixel. A ratio image was calculated by dividing the after image by the before image. We produced an animated Graphics Interchange Format (GIF) file showing the blinking before and after image on the left and a static ratio image on the right. Once created, we manually searched each animated GIF file and cataloged any changes in the image. Each detected change was verified by reviewing all other image pairs of the same location, and potential false positives were removed (Figures S3 and S4 in Supporting Information S1). Example animations of temporal image pairs are found in Movie S1 in Supporting Information S1.

3 Results

Following the methods described above, we identified nineteen quasi-circular surface changes and one linear feature (Table 1; Figure 1a). Most of the detections are in the northern hemisphere, likely due to an observational bias from the eccentric orbit of the spacecraft (Mazarico et al., 2014; Solomon & Anderson, 2018; Solomon et al., 2007) with a northern hemisphere periapsis (and thus higher spatial resolution). The nineteen quasi-circular albedo anomalies range in size from 400 to 1,900 m in diameter and formed between 18 March 2011 to 14 March 2015 (Figure S5 in Supporting Information S1). The center of each surface change has reflectance on average 11% (±4%) higher than the same area observed in the “before” image. Unfortunately, even at this scale, we could not definitively resolve a crater rim or detailed morphology within these albedo features, likely due to spatial resolution limitations. However, one of the changes located west of Caloris Planitia (38.02°N 115.18°E) has bright rays (reflectance increase of 4%) extending ∼20 km SW, consistent with an impact origin (Figures 1b–1d).

Table 1. Location and Properties of Detected Surface Changes
Location Formation date Size (m) Estimated crater diameter (m) Reflectance change (%) Formed on slope (i.e., crater wall) Formed within 50 km of tectonic landform Distance to LRM (km) Formed near existing hollow
20.91°N 2.98°E 7/28/2013 to 7/14/2014 930 290 6.3 ± 4.4 No No 70 Yes
20.43°N 11.66°E 2/1/2013 to 7/12/2014 1,050 320 5.4 ± 1.5 No Yes 15 No
36.32°S 82.72°E 1/8/2012 to 6/29/2012 1,050 320 25.3 ± 4.6 Yes Yes 50 No
18.85°N 99.08°E 8/10/2011 to 2/1/2012 930 280 12 ± 4.6 No No 200 Yes
38.02°N 115.18°E 6/25/2012 to 6/11/2013 1,030 320 17.9 ± 3.1 Yes Yes 25 No
41.27°N 117.22°E 12/1/2011 to 5/25/2012 1,170 360 6.2 ± 6.6 No Yes 20 Yes
36.02°N 135.04°E 12/15/2012 to 6/9/2013 1,260 390 18.5 ± 5.6 No Yes 30 No
33.58°N 156.65°E 12/11/2012 to 6/5/2013 500 150 8.7 ± 3.9 No Yes 150 No
50.56°N 166.25°E 11/8/2012 to 10/12/2014 550 170 7.7 ± 2.3 No No 20 No
32.65°S 183.91°E 6/10/2012 to 11/7/2012 450 140 8.8 ± 2.5 No Yes 0 No
62.75°N 208.52°E 12/19/2011 to 6/10/2012 830 250 5.9 ± 5.0 No Yes 20 No
1.77°S 241.72°E 4/8/2012 to 10/1/2012 610 190 6.1 ± 3.0 Yes No 80 No
51.27°N 267.16°E 4/18/2011 to 10/11/2011 4,410 n/a 14.3 ± 7.1 Yes No 0 Yes*
42.37°N 274.80°E 4/2/2012 to 3/19/2013 510 160 11.0 ± 4.2 Yes Yes 10 Yes
43.00°N 288.43°E 3/30/2012 to 3/17/2013 950 290 11.2 ± 2.9 No No 10 Yes
13.99°N 312.63°E 3/13/2013 to 9/4/2013 1,920 590 7.7 ± 2.3 No No 0 Yes*
2.57°N 316.32°E 9/20/2014 to 3/14/2015 790 240 32.7 ± 7.7 No No 0 No
40.30°N 332.06°E 3/23/2012 to 3/9/2013 870 270 6.9 ± 1.7 No Yes 0 No
42.41°N 333.44°E 3/23/2012 to 3/9/2013 390 120 4.5 ± 2.0 No No 0 No
32.70°N 357.18°E 2/3/2013 to 7/14/2014 1,030 320 6.6 ± 2.4 Yes No 15 No
  • Note. The formation date ranges are constrained by the temporal images covering the site. Distance to low reflectance material (LRM) is derived using the maps presented in Klima et al. (2018). Criteria for the change occurring near pre-existing hollow formation (Thomas et al., 2014) are based on occurrence within the crater hosting hollows or within one crater diameter (i.e., ejecta deposit) of the rim of the host crater. The two surface changes that occurred directly on existing hollows are marked with an asterisk. Four additional changes identified in this study that lack additional confirming observations can be found in Table S1 in Supporting Information S1.
Details are in the caption following the image

Distribution of identified surface changes and example temporal image sets. (a) Spatial distribution of surface changes with the red dots indicating focused reflectance anomalies and the blue dot indicating a linear surface change. (b) Before, (c) after, and (d) ratio (after image/before image) of a surface change resulting from an impact event (38.02°N 115.17°E). (e) Before, (f) after, and (g) ratio of a quasi-circular surface change observed at 33.58°N 156.64°E.

Assuming an impactor origin for these 19 surface changes, we estimated the crater diameters of each surface change by evaluating the relationship between the diameter of the high-reflectance ejecta deposit versus the diameter of the primary impact crater for resolved craters seen elsewhere on Mercury. Previous studies analyzing craters ≥30 km in diameter found that the continuous ejecta deposit on mercurian craters typically extends 0.34–0.55 crater diameters from the rim (Gault et al., 1975; Xiao et al., 2014). However, the features observed in this study are at least an order of magnitude smaller, and it is not known if the ejecta of smaller craters should follow the same trend. Therefore, we evaluated the extent of the ejecta around 52 resolved, Kuiperian-era craters with diameters between 100 and 400 m using a subset (n = 648) of the highest resolution MDIS images (pixel scale <30 m) with incidence angles between 25° and 50° and emission angles <15° to determine a scaling relationship for smaller impact craters. We found that the crater diameter was 30% ± 4% of the proximal high reflectance zone (inferred to be the continuous ejecta deposit; Figure S6 in Supporting Information S1). Therefore, we assume the quasi-circular albedo anomalies correspond to the proximal high-reflectance zone found around fresh craters and use the scaling relationship to approximate the crater diameter (Table 1).

While one surface change had clear ray patterns, the remaining features only exhibit a quasi-circular area of increased surface reflectance lacking any ray-like expressions (Figures 1e–1g). Therefore, while some of the remaining changes may be unresolved impact events, others may be the surface expression of endogenic processes such as faulting-induced landslides or enigmatic hollow formation. Five observed changes occurred on large (>10 km) craters or basin walls. Seismic shaking from nearby, active fault structures could induce a landslide on a slope that results in a change in the observed surface reflectance. For example, the change detected at 36.33°S 82.72°E occurs on the wall of a 21 km diameter crater superimposed on the 720 km Rembrandt basin, which is one of the most tectonically complex basins on Mercury (Figure S7 in Supporting Information S1). The basin contains a series of extensional and compressional structures and is crosscut by Enterprise Rupes, which is the largest thrust fault on Mercury (Ferrari et al., 2015; Watters et al., 20092016).

A second endogenic landform unique to Mercury is the rimless depressions with flat floors and rounded contours termed hollows (Blewett et al., 20112013; Thomas et al., 2014). Hollows preferentially occur in low reflectance material (LRM; ∼30% lower than the already dark global average reflectance) (Robinson et al., 2008) and on the floor, central peak, rim, or the ejecta deposit of large craters or basins. Previous studies suggested that hollows could result from the sublimation of volatile-bearing materials such as sulfides, space weathering, outgassing, and pyroclastic volcanism (Barraud et al., 2020; Blewett et al., 20112013). Additionally, their crisp features and lack of superimposed craters indicate that they are the youngest non-impact features on Mercury (Blewett et al., 2018).

One such hollow is located along the western terraced wall of Sholem Aleichem crater, which was imaged in 2011 (18 April and 11 October). The temporal ratio image reveals a linear surface change emanating from a hollow and extending down a steep slope (Figures 2a–2d). The linear change remains away from the shadow boundary, which remains stationary in the image pair due to the nearly identical lighting geometry (< 1° difference in incidence angle). This region was previously mapped as potentially impact-excavated LRM material or a dark spot, a subset of LRM with the lowest average reflectance (0.1–0.2) on Mercury, and proposed to form during the initial stages of hollow formation (Xiao et al., 2013). A high-resolution MDIS observation (Figure 2e) acquired on 6 November 2011 shows the hollow and higher reflectance material extending downslope.

Details are in the caption following the image

Linear feature showing putative mass wasting stemming from a hollow. (a) Context image (EW0211633839G) shows the rim of Sholem Aleichem (box, lower right). A closer look at the (b) before (EW0211633839G), (c) after (EW0226795417G), and (d) before/after ratio image enhancing a distinct linear reflectance feature (white arrow). The difference in incidence angle between the before and after image is 0.7°, and the reflectance feature falls outside of the shadowed area in both images. (e) Highest resolution Mercury Dual Imaging System (MDIS) observation (EN0229105236M) showing the location where the hollow material was deposited downslope toward the crater floor. The location of (b–d) is highlighted with the white box in (a).

Reviewing the other 19 surface changes, 12 occurred in areas mapped on or near (<25 km) LRM, and six occurred within or on the continuous ejecta deposit of craters containing hollows. One example is a surface change on the ejecta deposit of Lermontov crater, which formed between 13 March and 4 September 2013 and increased the surface reflectance by 8% ± 2% over a 2.9 km2 region (Figures 3a and 3b). If this feature was formed from an impact event and we detected the ejecta deposit, we would expect a primary crater 590 m in diameter (based on the crater diameter to bright ejecta relationship discussed above). However, an MDIS-NAC observation acquired in 2014 with a 4 times smaller pixel scale (55 m/pixel) than the original temporal pair shows a cluster of bright haloed hollows and a lack of any resolvable impact crater within the zone of increased reflectance (Figures 3c and 3d), consistent with an endogenic change associated with the hollows.

Details are in the caption following the image

Change on the rim of Lermontov crater. (a) Before, (b) after, and (c) ratio of the temporal image pair. (d) A closer look at the location of the change (red outline) in a higher resolution Mercury Dual Imaging System (MDIS) observation acquired after the original temporal pair. The location of (d) is highlighted with the white box in (c). No resolvable impact craters are found within the red outline, consistent with the hypothesis that the change is due to the growth of the hollow.

4 Discussion and Conclusions

Temporal imaging analysis of Mercury revealed a collection of new surface changes that formed during the orbital phase of the MESSENGER mission. While some surface changes are the result of exogenic impacts, the location of others suggests an endogenic origin. To evaluate the validity of this hypothesis, we examine several other mechanisms and causes for these detections:
  1. Artifacts related to the MDIS instrument

  2. Significantly higher (1,000 times) small impactor population during this period near 0.3 to 0.5 AU than current estimates

  3. Secondary impacts from a series of much larger impact events located outside of the temporal coverage area

  4. Extensive, temporary bright halo surrounding fresh impacts that is rapidly removed by space weathering

First, we considered if these anomalies were artifacts generated within the MDIS instrument and not changes on the surface. We verified each detection with multiple temporal pairs to rule out high-energy particle (cosmic rays) interactions with the detector (Figure S3 in Supporting Information S1). We also considered that the point-spread function of the MDIS optics is enlarging the apparent diameters of the ejecta deposit. However, the ejecta deposit would need to be enlarged by 25 times to align with current cratering models. Additionally, 15 of the 19 detected changes were observed with the Wide Angle Camera, which has a relatively narrow point spread function (FWHM = ∼1 pixel; MTF(Nyquist) = ∼0.3) (Hawkins et al., 2007), so we rejected this possibility.

Next, we evaluated the hypothesis that we witnessed only exogenic impact activity and no present-day endogenic activity. With crater diameter approximations for the 19 surface changes (125–535 m) based on the size of the continuous ejecta around fresh craters of similar size, we estimated the annual crater size-frequency distribution by accounting for the surface area observed, the time between observations, and the pixel scale of each image pair. The resulting analysis (Figure 4) indicates a substantially higher impact flux (3 orders of magnitude) than models predict (only one crater >100 m forming annually (Le Feuvre & Wieczorek, 2011; Marchi et al., 2009; Neukum et al., 2001; Strom & Neukum, 1988)). These models assume only asteroid impactors and ignore near-Sun comets with a perihelion distance less the perihelion distance of Mercury (0.307 AU) (Jones et al., 2018). However, even including a significant contribution of cometary impactors (up to 50%) (Borin et al., 2017; Rickman et al., 2001) our conclusion would not change. Due to this seemingly unjustifiably high cratering rate, we reject this hypothesis.

Details are in the caption following the image

Comparison of the estimated crater size-frequency distribution with prior models estimating the annual crater flux. The derived flux is based on crater size estimation, temporal coverage, periods between each observation, and the spatial resolution of the image set.

Next, we investigated evidence consistent with the impact rate hypothesis. Prior studies have shown that fresh, simple Kuiperian craters (225 m–14.4 km) originate with a depth/diameter (d/D) ratio of ∼0.2 (Pike, 1988). However, comparisons of lunar and mercurian craters on similarly aged terrain show that mercurian craters have a shallower d/D ratio (median = 0.13) than their lunar counterparts (median = 0.21) (Fassett et al., 2017). One hypothesis for the shallower d/D ratio on Mercury could be a higher flux of smaller impactors (2 times) degrading craters at a more rapid pace (Fassett et al., 2017). However, this rate is still 500 times lower than the spike in flux we estimated, and this dramatic flux increase is not seen in other temporal studies conducted in the inner Solar System (Daubar et al., 2013; Grier & McEwen, 1997; Speyerer et al., 2016).

Alternatively, we could be identifying secondary impacts from one or more large newly formed primary crater(s) outside our study area. One limitation of change detection is that the time of formation is not well constrained; the estimate is based on the temporal spacing between observations, varying from several months to several years. However, 10 observed changes had formation time windows overlapping between 3 February and 9 March 2013 (Table 1; Figure S5 in Supporting Information S1). While changes could be secondary craters, comparisons of global mosaics created from images acquired early and late in the mission do not reveal any new large craters (>5 km diameter); formation of craters of this size is exceedingly rare on a yearly time scale. Additionally, we did not observe clustered impacts (Figure 1a) or herringbone patterns commonly seen with secondary impacts (Oberbeck & Morrison, 1973; Robinson et al., 2015).

Next, we examined if we identified transient reflectance halos significantly larger than the continuous ejecta deposit. Newly formed lunar impacts exhibit broad distal reflectance zones in temporal ratio images, some extending in ray-like patterns outward of >1,000 crater diameters (Speyerer et al., 2016). These distal zones are thought to result from melt and vapourized rock jetted from the impact site during the initial contact and compression phase (Johnson et al., 2014; Melosh, 1989; Speyerer et al., 2016). In the lunar case, the distal high reflectance zones can increase the observed reflectance by only 1%–4% and extend on average 0.52D2.5 from the center of the crater where D is in meters (Speyerer et al., 2016). However, with the higher average impactor speeds (42.6 km/s vs. 19.7 km/s) (Strom & Neukum, 1988), there is the potential that jetted material may smooth the upper regolith layer and increase the observed reflectance around the site to a greater lateral extent on Mercury relative to the Moon. This delicate layer would then be subjected to the rapid space weathering environment and micrometeoroids that quickly degrade the Kuiperian craters discussed in the prior study (Braden & Robinson, 2013; Fassett et al., 2017). If this is the case, the bright halos would only be present for a short period (10s–1000s of years?) and would not be resolvable in slightly older Kuiperian craters, such as those in our analog study. Interestingly, unlike the lunar craters that exhibited broad distal low reflectance zones (Robinson et al., 2015; Speyerer et al., 2016), none of the observed changes in this study exhibit lower reflectance in the after observation, suggesting a difference due to the increased impactor speed (Strom & Neukum, 1988), greater impact melting and vapourization (Cintala, 1992), space weathering (Braden & Robinson, 2013), and/or regolith composition (Nittler et al., 2018).

Therefore, the observations and our analysis are consistent with the surface changes resulting from a mixture of present-day endogenic and exogenic activity. If these events were stochastic and representative of normal surface activity, 99% of the entire surface could be modified in ∼25M years. Additionally, smaller and more numerous surface changes (<500 m) are likely occurring, resulting in an even faster modification rate. However, since some of these changes are likely occurring due to localized endogenic activity (tectonic and hollows), some regions may be modified more rapidly than this average rate while other regions located outside of LRM and away from active tectonic activity may have a lower turnover rate due to only impact events.

The BepiColombo spacecraft (Benkhoff et al., 2021) will enter orbit around Mercury in 2025 and is equipped with SIMBIO-SYS (Cremonese et al., 2020), a collection of high-resolution imagers. These observations can provide additional detailed images of the changes detected in this study and spur future temporal analysis by comparing SIMBIO-SYS images to the MDIS images collected during the MESSENGER mission, providing a much longer baseline between observations and greater overall temporal coverage.


This work was supported by Arizona State University Internal Investigator Incentive Award PG04817. The authors would like to thank Dr. Brett Denevi for her review of an earlier draft of this work as well as Dr. Thomas Watters and Dr. Simone Marchi for their beneficial comments during review.

    Data Availability Statement

    The image data used for the change detection analysis in the study is available at the Cartography and Imaging Sciences Discipline Node of the Planetary Data System with open and free access (Hash, 2008). We processed the data using an open-source planetary cartographic software package called ISIS (Anderson, 2008), which is available on GitHub (Adoram-Kershner et al., 2020).