Evaluation of wet and dry recurring slope lineae (RSL) formation mechanisms based on quantitative mapping of RSL in Garni Crater, Valles Marineris, Mars

Icarus 335 (2020) 113420

Contents lists available at ScienceDirect

Icarus journal homepage: www.elsevier.com/locate/icarus

Evaluation of wet and dry Recurring Slope Lineae (RSL) formation mechanisms based on quantitative mapping of RSL in Garni Crater, Valles Marineris, Mars

T

⁎

David E. Stillmana, , Brian D. Bueb, Kiri L. Wagstaffb, Katherine M. Primma, Tim I. Michaelsc, Robert E. Grimma a b c

Dept. of Space Studies, Southwest Research Institute, 1050 Walnut St. #300, Boulder, CO 80302, USA Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109, USA Carl Sagan Center, SETI Institute, 189 Bernardo Ave Suite 200, Mountain View, CA 94043, USA

A R T I C LE I N FO

A B S T R A C T

Keywords: Mars, surface Geological processes Mars, climate

Recurring slope lineae (RSL) are narrow (0.5–5 m) low-albedo features that incrementally lengthen down steep slopes during warm seasons, fade in colder seasons, and recur each Mars year. To reduce the effort involved in manually mapping and analyzing each RSL in Garni crater, we developed Mapping and Automated Analysis of RSL (MAARSL) to analyze a set of orthorectified High Resolution Imaging Science Experiment (HiRISE) images and a digital elevation map. MAARSL along with manual mapping allowed us to detect RSL, compute descriptive statistics, and characterize changes over time. We mapped 2910 RSL in 22 orthoimages, from Mars Year (MY) 31 solar longitude (Ls) 133.0° to MY32 Ls 323.7°. The MAARSL results confirmed that RSL lengthening and fading occur concurrently on slopes with the same orientation and many times within some individual RSL. Slope angles of RSL and a slope slump show that some RSL start, stop, and have mean slope angles that are below the angle of repose. Our analysis shows that RSL are actively lengthening on at least one slope-facing direction in all HiRISE observations of the crater. We also found that NE-, N-, and NW-facing RSL in Garni crater lengthened during times of increasing shortwave insolation, while S- and SW-facing RSL lengthened during increasing and decreasing shortwave insolation. A (non-orthorectified) HiRISE image acquired shortly after the MY34 dust storm and shows RSL on every slope-facing direction, which is anomalous with respect to observations prior to the dust storm. We find that dust removal and deposition could explain the darkening and fading (respectively) of RSL, and could also explain the apparent lack of material being transported. Observations of RSL lengthening and fading occur concurrently on slopes could suggest overprinting of dry granular flows. Dry flows could also explain the significant lengthening activity of every slope-facing direction after a fresh layer of dust was deposited via the MY34 dust storm. Alternatively, briny shallow subsurface flows are consistent with observations of RSL on slopes below the angle of repose and those that exhibit concurrent lengthening and fading. However, the most significant problem with briny RSL flows is accessing a source of briny water and removing excess salt from the regolith.

1. Introduction

image coverage) are considered to be candidate sites. Stillman (2018) analyzed Mars Reconnaissance Orbiter (MRO) High Resolution Imaging Science Experiment (HiRISE; McEwen et al., 2007) data acquired up to July 15, 2017 and documented 98 confirmed RSL sites and 650 candidate RSL sites over seven geographic provinces of Mars with a latitude range of 42°N - 53°S. Each site is categorized qualitatively as excellent, very good, good, fair, or poor based on the size of individual RSL and the RSL areal density. Statistical analysis by McEwen (2018) has shown

Recurring slope lineae (RSL) are dark narrow features that have three characteristics: (1) incremental lengthening during the warm season, (2) fading during the cold season, and (3) annual recurrence (McEwen et al., 2011). RSL sites that possess all three characteristics are labeled as confirmed sites, while sites with similar features that have not demonstrated all three properties (typically due to insufficient

⁎

Corresponding author. E-mail address: [email protected] (D.E. Stillman).

https://doi.org/10.1016/j.icarus.2019.113420 Received 21 May 2019; Received in revised form 23 August 2019; Accepted 25 August 2019 Available online 30 August 2019 0019-1035/ © 2019 Published by Elsevier Inc.

Icarus 335 (2020) 113420

D.E. Stillman, et al.

2.1. Dry RSL formation mechanism The dry RSL formation mechanism hypothesis assumes that RSL are dry granular flows that are routinely triggered, allowing sand/dust to progress down steep slopes that exceed the angle of repose. Dundas et al. (2017) found that the vast majority of RSL occur on slopes > 28°, similar to the observed slopes of martian sand dunes (Atwood-Stone and McEwen, 2013). Atwood-Stone and McEwen (2013) estimated the dynamic angle of repose (the slope at which granular flow will come to rest) to be 30–34° at Herschel, Nili Patera, and Gale crater dunes. Similarly, Ermakov et al. (2019) found that the global distribution of martian slopes (using MOLA onboard Mars Global Surveyor; Zuber et al., 1992) has a distinct kink at a slope of 29.8°, which was interpreted to be the dynamic angle of repose. A more detailed survey of Bagnold Dunes' lee slopes measured the dynamic angle of repose to be ~29° where grainflows where present (Ewing et al., 2017). The static angle of repose (the shallowest angle at which granular flow will transpire) at Bagnold Dunes was measured were grainfall was present to be at ~33° (Ewing et al., 2017). Thus, we will measure the angle of RSL slopes to test the dry hypothesis. While incremental lengthening is assumed, the minimum HiRISE revisit period at Garni crater is 16 sols. Thus, a triggering mechanism that occurs with a cadence of < 16 sols would appear to be incremental. The triggering mechanism is also unknown; suggestions include wind (Schaefer et al., 2019; Vincendon et al., 2019), a reduction in the angle of repose due to a Knudsen pump (Schmidt et al., 2017), volume transition between hydrated and dehydrated mineral phases (Dundas et al., 2017), and dehydration leading to lower cohesion (Dundas et al., 2017). The darkening of RSL could be caused by vertical inverse grading, in which grains are sorted during the granular flow or by the ejection of finer grains and dust during flow (Schmidt et al., 2017). RSL fading could be caused by dust deposition (Schmidt et al., 2017). However, Schaefer et al. (2019) observed that relative albedo measurements at a southern mid-latitude RSL site (Tivat crater) suggest that the terrain surrounding RSL actually darken to the albedo of the RSL, thus causing an apparent fading of RSL. This suggests dust is removed from the RSL before being removed from the surrounding terrain. Dust is then re-deposited on the slopes just before RSL activity. Additionally, RSL typically occur in the same small-scale troughs every year, thus a dry process is needed to continuously reset the new dry flows. This can be done via atmospheric deposition if the flows are dust avalanches. However, if the flows are caused by sand grains, then aeolian processes are required to either blow sand back up these steep slopes (Dundas et al., 2017) or supply it from elsewhere.

Fig. 1. A regional view of central Calles Marineris, illustrating the location of Garni crater. Symbol shapes represent each RSL site's qualitative ranking, while the colour indicates the number of RSL characteristics. Background is the Thermal Emission Imaging System (THEMIS) daytime infrared mosaic, colorized with Mars Orbiter Laser Altimeter (MOLA) elevation (Hill et al., 2014) Java Mission planning and Analysis for Remote Sensing (JMARS) layer.

that RSL sites likely recur for at least 100 Mars Years (MYs), as no confirmed RSL site has ever ceased activity during the period of monitoring (3–5 MYs). Furthermore, RSL have never been shown to transport material, although slumps have formed at two RSL sites at the terminus of individual RSL and have transported material hundreds of meters downslope (e.g., Chojnacki et al., 2016; Ojha et al., 2017). Moreover, RSL slopes retain no craters down to the scale of HiRISE imagery. This suggests that the RSL formation mechanism co-occurs with frequent resurfacing and may influence the shape of the local landscape (McEwen, 2018). For this research, we perform a detailed analysis of a confirmed excellent RSL site within Garni crater (11.516°S, 290.308°E). This 2.4km-wide crater is on the floor (elevation −4910 m) of Melas Chasma (Fig. 1). Thus the RSL site exists within the Valles Marineris (VM) RSL province, which possesses the longest RSL and the greatest density of RSL sites (Stillman, 2018). The majority of VM RSL exist on the canyon walls, but this crater was chosen due to the presence of large RSL that exist on numerous slope orientations. Furthermore, we have much greater temporal coverage in HiRISE observations of Garni crater, compared to a similar confirmed RSL site (Avan crater, 3.3 km in diameter and 32 km to the north). In this paper, we use our detailed mapping of RSL in Garni crater to investigate the area, newly-darkened rate, fading rate, and slope angles of RSL and a slope slump to determine if such phenomena might be due to dry or wet processes. We then correlate the active lengthening season with shortwave solar insolation and dust opacity. We also compare our findings of the slope orientations within Garni crater to those of other RSL regions. Lastly, we discuss how our findings support and contradict dry and wet formation mechanisms and advocate that additional modeling of both formation mechanisms is needed.

2.2. Wet RSL formation mechanism The wet-dominated formation mechanism hypothesis is defined by brines that slowly percolate downhill through regolith consisting of sand or silty sand (Levy, 2012; Grimm et al., 2014; Stillman et al., 2016). These flows stay in the shallow regolith due to a shallow aquitard, likely caused by ice or salt hydrate that exists year-round in the regolith. Any water lost due to sublimation or evaporation must be replenished year after year. The source of the brine is likely a spring connected to a deeper pressurized aquifer (Stillman et al., 2014, 2016) via a permeable fracture (Watkins et al., 2014; Abotalib and Heggy, 2019). This brine would have a freezing point similar to the mean annual temperature so that it does not freeze within the fracture. The associated spring only flows during the warmest part of the year, with a near-surface ice dam forming during the colder months. When brine is discharged from the spring, it flows downhill and darkens the regolith. The length that an RSL can migrate is controlled by its loss (evaporation/sublimation) rate compared to its discharge rate (Grimm et al., 2014). The loss rate increases with temperature, melting point of the brine, and surface area of the RSL. If the RSL loss rate equals or exceeds the spring discharge rate, then RSL will stop lengthening or decrease in

2. Background Since the first discovery of RSL on Mars, both dry and wet formation and reset mechanisms have been discussed (McEwen et al., 2011). Yet no mechanism can sufficiently explain the behavior of all RSL observed to date. Below we briefly discuss the benefits and limitations of each mechanism (for a more detailed overview see Stillman, 2018).

2

Icarus 335 (2020) 113420

D.E. Stillman, et al.

of ~20 cm/pixel (Kirk et al., 2008). We performed computer-assisted analysis on the orthorectified HiRISE images and used the non-orthorectified images to investigate interannual variations.

size, respectively. Similarly, when temperatures drop, the time during which brine can stay liquid decreases until temperatures drop below the eutectic temperature. Additionally, colder temperatures can also lead to the refreezing of the ice dam. RSL then slowly sublimate (from the surface down) and fade. The wet-dominated flow mechanism hypothesis best explains the incremental lengthening and darkening of the surface when temperatures are warm, as well as how RSL darken and fade without moving a measurable amount of material over the last six MYs. Hydrated oxychlorine salts occur at the same location and seasonality as RSL at four sites (Ojha et al., 2015). However, Powell and Arvidson (2017) could not reproduce these spectral signatures and others suggest the hydration signal was due to noise inherent to CRISM (Leask et al., 2018; Vincendon et al., 2019). Wet-based hypotheses are confined to aquiferdriven mechanisms because they appear to require significant water budgets (Grimm et al., 2014; Stillman et al., 2016) that are greater than could be obtained by deliquescence of atmospheric water onto salts (e.g., Chojnacki et al., 2016; Stillman et al., 2017; Dundas et al., 2017) or cold trapping via subsurface water vapor transport. RSL sites at low elevation (such as Garni crater) could be formed via either confined or unconfined aquifers, but RSL found near the rim of Coprates Montes and Nectaris Montes remain a significant challenge (Chojnacki et al., 2016) for the wet-dominated mechanism to explain because remarkable pressurization of the aquifer is required. Additionally, since brine is required for RSL formation, it is unclear what happens to the large amount of salt that would be precipitated into the regolith each year. Such salt should form a caliche layer that would significantly reduce the permeability of future flows. No evidence of a caliche layer has been found using thermal data (Edwards and Piqueux, 2016), but caliche layers also fail to form in briny Antarctic wet tracks that may be the best terrestrial analogs to RSL (Levy, 2012).

3.2. Computer-assisted analysis with MAARSL Areas where RSL occur can be densely populated, with hundreds of individual RSL at a given site. To reduce the effort involved in manually mapping each RSL in each HiRISE image, we developed the Mapping and Automated Analysis of RSL (MAARSL) system. MAARSL analyzes a set of orthorectified HiRISE images and their associated DEM to detect candidate RSL, compute descriptive statistics, and characterize changes over time. MAARSL is an interactive tool that solicits feedback on its proposed candidates to filter and refine them with minimal human time investment. 3.2.1. RSL candidate detection HiRISE images vary greatly in their illumination and contrast. To achieve robust RSL detection despite changes in sun angle and image quality, MAARSL creates a per-image model Mi that predicts pixel intensity based on the position and orientation of the HiRISE instrument when image i was collected. The model is a least-squares linear fit that maps incident illumination to expected pixel intensity (Fig. 2); Mi(pj) has the form mi(I(pj)) + bi, where mi and bi are estimated separately for each image, and pj ranges over the pixels j in the image. In this work, we

2.3. Hybrid RSL formation mechanism Hybrid RSL formation mechanism hypotheses invoke wet-triggered granular flows as a way to explain incremental lengthening using less water. In these hypotheses, boiling of fresh water allows granular flows to be triggered. With a small supply of water, small amounts of sand pile up and then the piles fail, creating small granular flows of ~10 cm (Massé et al., 2016). At a relatively high water flow rate, sand flows levitate on water vapor (from boiling), enhancing any downslope sediment transport (Raack et al., 2017). These complex hybrid mechanisms suffer from the limitation of the wet and dry mechanisms, in that they need a source of liquid water that is recharged each year, dry material must be reset each year, and darkening and fading must be due to slight changes in surficial dust. 3. Methodology 3.1. Data Garni crater was targeted for this study because it possesses large RSL that exist on nearly every slope-facing direction. Thus, RSL are present in every HiRISE image, and we can map the seasonal changes on all slope-facing directions. Furthermore, Garni crater has 22 HiRISE orthorectified-images covering 1.5 Mars years (MY31 Ls 133.1° to MY32 Ls 323.7°) as well as a HiRISE-generated Digital Elevation Map (DEM) (available for download at the Planetary Data System Cartography and Imaging Sciences Node). There are an additional 14 images that have not been orthorectified via the SOCET SET toolchain (Kirk et al., 2008). However, these data allow us to qualitatively detect changes over an additional two Mars years (last image is MY34 Ls 321.9°). While many sites have been imaged for a longer period of time, few RSL sites have been imaged with the temporal frequency of Garni crater. The HiRISE orthorectified-images and DEM have a spatial sampling of 25 and 101 cm/pixel, respectively. SOCET SET DEMs have a vertical precision

Fig. 2. MAARSL illumination-intensity model. The model predicts the relative illumination using the HiRISE DEM (b) and compares it to the actual value in the HiRISE orthoimages (a). (c) Shows the linear least squares fit (blue dashed line) between illumination and intensity. Shadows are assumed to be at low relative illumination and RSL are assumed to be darker than the predicted intensity value at high illumination. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 3

Icarus 335 (2020) 113420

D.E. Stillman, et al.

number of examples that need to be individually reviewed. The left panel shows the currently-learned classification of RSL as “real” (blue) or “spurious” (red). The GUI iteratively selects the candidate that it is least confident about classifying into one of these categories and displays it in the middle panel. The user selects “real” or “spurious” (right panel), and the GUI re-trains an underlying classifier with the new judgment included. This updates the overall predictions for all candidates (left panel) and leads to the selection of a new least-confident candidate (middle panel). This process continues until the user is satisfied that all candidates are correctly classified, at which point all candidates marked “real” are saved, and the “spurious” ones are discarded. In our experience, we only needed to label < 30 candidates for a given image before the candidate RSL were classified correctly (Fig. 3). In addition, as the user moves on to the next image and its candidates, the model that was trained on the preceding image is used to initialize the system, so subsequent images require even fewer labels to properly classify the candidates. The underlying classifier for this system is a random forest (Breiman, 2001). A random forest is a collection of individual decision trees that vote on the final prediction. Each decision tree is trained on a different (randomly selected) subsample of the training data, so they learn slightly different models. It has been shown that the ensemble voting approach of the random forest is more robust than that of a single decision tree trained on all of the data. Random forests are quick to train and evaluate, making them suitable for the GUI's interactive setting. The random forest does not train directly on the input images. Instead, each RSL candidate is represented using a vector of computed features, including area (number of pixels), elevation, illumination, roughness, and the properties of an ellipse fit to the candidate's polygon (see Fig. 3, right panel). The interactive GUI allows fast filtering of spurious RSL candidates. However, it is also possible for MAARSL to fail to detect candidates, so many RSL do not make it to this stage of analysis. Therefore, we also used a final manual editing pass using Adobe Photoshop to modify the MAARSL-selected RSL picks and to map any RSL that were not detected by MAARSL. We then computed the precision neMAARSL/nMAARSL and recall neMAARSL/nTotal of the MAARSL RSL picks (Fig. 4), where nMAARSL is the total number of MAARSL-selected pixels, neMAARSL is the number of MAARSL-selected pixels that are included after manual editing, and nTotal is the total number of RSL pixels after manual editing. The precision and recall have a large variance, where precision and recall increase as RSL become more prevalent (especially on sand fans). MAARSL achieved an overall precision of 89% and recall of 23% for the first 21 images. Note that the last orthoimage ESP_040804_1685 was not analyzed with MAARSL because it was derived from a HiRISE image acquired with a 50 cm/pixel resolution, as opposed to 25 cm/pixel for

exclude pixels in the headwall region when computing Mi and bi, as the illumination estimates in the headwall are unreliable due to the rapid local changes in elevation and the mismatch in spatial resolution of the Garni DTM (1 m/pixel) vs. the orthoimages (0.25 m/pixel). Once computed, we apply the resulting Mi, bi on all pixels in the image to detect RSL. Illumination I(p j) is calculated using the method described in Horn (1981), which generates a shaded relief of the DEM with respect to the subsolar azimuth and incidence angle associated with each HiRISE image. The assumption behind this simple model is that, all other things being equal, areas with the same incident illumination should exhibit the same observed intensity. This model enables MAARSL to distinguish between areas that are dark due to the presence of shadows (lack of illumination) versus those that are dark due to surface processes such as RSL (independent of illumination). Creating a model for each image enables adaptation to overall image brightening or darkening. However, an important limitation of this model is that it does not factor in differences in surface composition that affect reflectance properties and therefore observed intensities. RSL candidates are detected by comparing the observed intensity O (p j) to that predicted by the illumination-intensity model. We expect areas with active RSL to be darker than predicted by the model, given the amount of illumination they receive. We compute the darkening d at each pixel pj as d = Z − (O(pj) − Mi(pj)),where Z = max O (pj ) − Mi (pj ) is the maximum difference between the modeli

predicted and observed intensity values across all images in the time series for pixel pj. This adjustment provides some accommodation of variations in surface composition; some areas (e.g., with larger/smaller grain size) appear consistently darker or lighter than other areas. Finally, MAARSL identifies regions within the image with darkening values that exceed a specified threshold (data number of 50 out of 255, so that shadowed regions are removed) and constructs an enclosing polygon for each connected region (RSL candidate). 3.2.2. RSL candidate interactive filtering Even after factoring in illumination and basic composition differences, the MAARSL threshold-based approach to RSL candidate detection yields several false detections when the threshold is chosen to be sufficiently sensitive to detect real RSL. We therefore employ a postprocessing step that enables fast manual review and filtering of the detections. Fig. 3 shows the graphical user interface (GUI) that provides the ability to review candidates. As noted above, a single image may contain hundreds of real RSL. Reviewing every candidate would be tedious and would not scale well to the desired goal of large-scale monitoring and analysis of new HiRISE images as they become available. Instead, the GUI employs active learning (Cohn et al., 1996) to minimize the

Fig. 3. Interactive graphical user interface for fast review and filtering of RSL candidates. Candidates are classified as real (blue) or spurious (red). Complete classification of all 1484 candidates was achieved for this image after labeling only 23 examples. The parameters in the “RSL info” panel are defined in supplementary research data file “image_stats_readme.pdf”, with the only difference being that the orientation parameter above is in radians. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 4. Precision and recall of MAARSL analysis. The unbolded image labels indicate images that had no MAARSL detection. This occurred because the RSL activity was small and typically in bedrock regions where the spatial resolution of the DTM is insufficient to resolve localized elevation changes necessary to accurately model illumination and shadows. Note that the total precision and recall values in the title are calculated using the total number of pixels, as opposed to finding the median of the images.

supplemental research data. Additionally, all the statistics for every RSL are also provided in the supplemental research data.

the others. Therefore, this image was only mapped manually due to the lack of resolution combined with the small RSL during MY32 Ls 323°.

3.3. Estimates of insolation and dust

3.2.3. RSL characterization via geostatistics Each verified RSL is described by a georeferenced enclosing polygon (perimeter). Verified dark RSL pixels were then categorized into individual RSL by grouping continuous pixels, where continuous RSL included any of the eight pixels that surrounded a dark RSL pixel. We performed a geostatistical analysis on these polygons to determine the number of RSL, the RSL areas, and the slope orientations of the RSL. Slope orientation was calculated using a python image processing package (scikit-image), where slope orientation is the angle of the major axis of an ellipse that best fits the individual RSL polygon. Slope orientation values range from 0° to 360°, representing the azimuth that slopes are facing, and are ordered clockwise from 0° (N-facing) to 90° (E-facing) to 180° (S-facing) and so on (Fig. S1). We also assess RSL changes over time. Each image in the series is first analyzed independently to find RSL, and results from consecutive images are then compared to identify areas in which RSL have newly formed, persisted, or faded. The number of pixels that have newly darkened, persisted, or faded are binned into 24 circular sectors that are 15° each, starting at an orientation of 352.5° (Fig. S1). We also compute the newly-darkened and faded rate by dividing the newly-darkened area or faded area, respectively, by the number of sols between the two images. To better understand the errors in our area statistics, we add or subtract a pixel (assuming our MAARSL mapping is correct to ± 1 pixel) to the entire perimeter of each individual RSL to determine our upper or lower error limits, respectively. Additional geostatistical properties are found by using the georeferenced perimeter polygons and DEM, such as: RSL starting and stopping elevation and slope, and minimum, mean and maximum slope. We generate DEM-based slopes at 1 m per pixel by using the first-order finite-difference method described in Horn (1981). Note that the starting and stopping slopes are found at the maximum and minimum elevation of each individual RSL. The mean slope was found by averaging all the slopes within the RSL polygon. RSL areas for each image, each RSL, and each orientation as well as area errors are provided in the

We used a simplified offline version of the Mars Atmospheric Regional Modeling System (MRAMS: Michaels and Rafkin, 2008) radiative transfer and orbital parameterizations to calculate the amount of shortwave radiation (the primary radiant energy from the sun in W/ m2) potentially incident on various Garni crater ground surface slope angles and orientations over a complete Mars year. This is a theoretical maximum value of the amount of incident shortwave insolation because we assume no atmospheric scattering or absorption, in order to remove the effects of any interannual variability in atmospheric dust loading (e.g., local and regional dust storms). We then computed a time-integrated amount of that incident shortwave insolation over each sol EMax for each slope angle and orientation. The general dust opacity cycle at Garni crater for MY24–33 was estimated using the dust opacity (specifically the 9 μm-band column absorption optical depth) climatologies of Montabone et al. (2015; 3 × 3 degree spatial resolution) that were created with TES (Thermal Emission Spectrometer onboard Mars Global Surveyor; Christensen et al., 2001), THEMIS (onboard 2001 Mars Odyssey; Christensen et al., 2004), and MCS (Mars Climate Sounder onboard MRO; McCleese et al., 2007) observations. 4. Results 4.1. Insolation As Garni crater is at 11.5°S latitude, EMax for a flat surface reaches its maximum value at Ls 247° (Fig. 5), just before perihelion (Ls 251°) when the subsolar point is in the southern hemisphere and moving further south each sol. The slope orientation allows for an enhancement of shortwave insolation when the sloped surface is facing toward the subsolar point, and vice versa. Thus, N-facing slopes gain additional shortwave energy when the direct rays of the sun are to the north of the 5

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 5. Shortwave insolation metrics (assuming no atmosphere) for Garni crater as a function of slope angle and orientation. EMax (time-integrated shortwave insolation per sol metric) for (a) N-facing and (b) S-facing slopes over a Mars year. Peak instantaneous shortwave insolation per sol for (c) N-facing and (d) S-facing slopes. Diurnal variation of instantaneous shortwave insolation during the sol with the maximum yearly EMax value for (e) N-facing and (f) S-facing slopes. For Nfacing slopes, EMax and peak instantaneous insolation are similar in shape, but are out-of-phase with regard to the insolation that is received by flat surfaces. Slopes similar to those that RSL are found upon (30° and 32.5°) reach their maximum EMax value slightly before equinox (Ls 180°). For S-facing slopes, EMax and peak instantaneous insolation are much different in shape, with the steepest slopes (≥30°) receiving less peak instantaneous insolation than a flat surface, but receiving a greater amount of EMax compared to a flat surface. These S-facing sloped surfaces receive their maximum EMax between perihelion (Ls 251°) and southern summer solstice (Ls 270°). The differences between EMax and peak instantaneous insolation are best displayed in f, where slopes ≥30° attain lower peak instantaneous insolation values, but are able to receive more energy in the morning and evening compared to the flat surface.

crater (Fig. 5a). This creates a small-amplitude dual-peak annual EMax curve for N-facing slopes, where the maxima change their seasonality and magnitude as a function of slope angle, but are generally near the equinoxes (e.g., Ls 0° and 180°). For S-facing slopes, a sharp high-amplitude EMax curve is predicted, which varies with slope angle and peaks slightly after perihelion (Fig. 5b). Additionally, steep slope angles produce the maximum instantaneous insolation enhancements during optimal geometries, but they generally receive less energy over the entire year than shallower slopes. Note that even at a tropical latitude of 11.5°S, a 32.5° N-facing slope's EMax only varies between 12.4 and 17.0 MJ/m2, while that of a 32.5° S-facing slope's varies by a factor of ~5 more between 3.6 and 23.0 MJ/m2. We plot the peak instantaneous shortwave insolation received each sol for N- (Fig. 5c) and S-facing slopes (Fig. 5d). The EMax and peak instantaneous insolation values for N-facing slopes behave in a similar fashion. However, the EMax and peak instantaneous insolation values for S-facing slopes are different. Variations in S-facing slopes are magnified because Garni is in the southern hemisphere and because perihelion occurs just before southern solstice, resulting in a much larger amount of insolation (vs. aphelion) due to Mars' significant orbital eccentricity. Fig. 5d demonstrates that steeply

Fig. 6. EMax (time-integrated shortwave insolation per sol metric) as a function of slope-facing direction and seasonality, assuming a slope angle of 32.5° at Garni crater.

6

Icarus 335 (2020) 113420

D.E. Stillman, et al.

sloped surfaces actually have lower peak instantaneous insolation values than a flat surface, but EMax (Fig. 5b) can be larger because S-facing steep slopes receive significantly more insolation in the morning and evening (Fig. 5f). The morning and evening insolation effect is minimized for N-facing slopes (Fig. 5e). Given a constant slope angle of 32.5° (typical of RSL; Dundas et al., 2017), Fig. 6 demonstrates how EMax varies seasonally with slope-facing direction. Any correlations with RSL lengthening and fading will be made using the EMax values rather than peak instantaneous insolation values because (1) S-facing steep slope peak instantaneous insolation values are not greater than that of a flat surface, which is potentially inconsistent with RSL occurring only on slopes with relatively high surface temperatures and (2) EMax is a better measure of temperature at shallow but significant depth (on the order of centimeters) compared to the peak instantaneous insolation which is a better proxy for maximum surface temperature. Using EMax would also be a better match for wet RSL models that involve diurnal remelting of brines or any dry models which may rely on a thicker heated layer to trigger their formation (Stillman, 2018). Note that we decided to compare the RSL statistical analysis to insolation instead of temperature, since temperature is also very sensitive to surface and subsurface thermal properties.

Fig. 7. (a) Orbitally-derived dust opacity above Garni crater for Mars Years 24–33. Note that MY 25 (blue) and 28 (cyan) have global-encircling dust storms. The recent MY34 dust storm (not shown) most closely resembles the MY25 storm. Further note that from Ls 0–130°, the dust opacity shows minimal interannual variability. (b) Maximum, median, and minimum dust opacities were calculated from the MY24–33 data. See supplemental file named “Garni_RSL_area.csv” to determine the MY that each HiRISE image was acquired in. For a more detailed plot of MY31–33 see Fig. S2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4.2. Dust The dust cycle is another important variable that can affect RSL activity by (1) reducing the amount of shortwave insolation that reaches the surface, (2) increasing the amount of downward longwave (e.g., infrared) radiation reaching the surface, (3) changing the albedo of the surface, and (4) possibly providing a recharge mechanism for dust that could then avalanche down the slope. Note that atmospheric dust opacity cannot simply be converted into downward longwave radiation because poorly-constrained assumptions would have to be made about the vertical atmospheric temperature and dust profiles and dust particle size distribution. Global-encircling dust storms were detected in MY25 (an equinox storm similar to the MY34 storm) and MY28 (a solstice storm) as well as more localized dust events. (Note that equivalent MY34 values are not available yet.) The maximum, median, and minimum of these data were then calculated (Fig. 7), and illustrate the large interannual variation in dust opacity from Ls 125–360° and the minimal interannual variability of dust opacity for the remainder of the year. Overall, if RSL and the dust cycle were highly correlated, significant interannual variations in RSL would be expected between Ls 125–360°, with little such variation for the remainder of the year. The MY34 globe-encircling dust storm considerably affected the RSL in Garni crater. The precursor storms started at Ls ~181° (Guzewich et al., 2019). At Ls ~193° the storm was considered globe-encircling (Malin et al., 2018a, 2018b). Dust started settling out at Ls ~205° (Malin et al., 2018c, 2018d), but did not reach typical levels until Ls ~250° (Malin et al., 2018e). The first HiRISE image of Garni crater after the dust storm was acquired at Ls 255.7° and has not been orthorectified, so only a qualitative analysis was done (Figs. 8 & S3). We found that every slope-facing direction had considerable RSL present, whereas given the season, only SW-, S-, and SE-facing slopes should have had RSL. Additionally, the RSL after the MY34 dust storm had numerous tributaries (Fig. 8). MAARSL mapped some RSL tributaries in earlier observations, but their density was never as great as it was after the dust storm. Note that we cannot rule out the possibility that these recent RSL with tributaries were merely more easily observable because of the overall lighter albedo of the surface (due to dust fall) after the dust storm. While the MY34 dust storm triggered RSL on all slope-facing directions, the total areal extent of the RSL did not appear to be vastly greater than our MAARSL mapped data. Additionally, the post-storm RSL do not appear to extend any further downward into the crater.

4.3. Quantitative mapping results 4.3.1. Area We mapped 2910 RSL in 22 orthoimages, with the image collected at MY31 Ls 133° having the largest number of RSL (276). Instead of concentrating on the number of RSL (as RSL can vary greatly in size), we focus on the darkened area of RSL as a function of seasonality and slope orientation. Fig. 9 displays all of the RSL (and the single slope slump) that were mapped quantitatively. Our analysis of the MAARSL results reveals amazing detail about how RSL lengthen and fade. RSL in Garni crater do not just lengthen or fade, but instead a single RSL can have a portion of itself fade, while another portion lengthens (Fig. 10b). This indicates that the processes that erase RSL must be occurring at the same time as RSL formation/ lengthening. Additionally, the same RSL can be active multiple times in the same Mars year. Fig. 10e shows a long, lengthening RSL (slightly to the west of center in the image) at MY32 Ls 77°. In Fig. 10f, the RSL is mostly a persistent RSL at Ls 89°, which then fades back uphill at Ls 145° (Fig. 10g). However, this RSL is subsequently rejuvenated and experiences rapid lengthening as observed in the Ls 178° image (Fig. 10i), further demonstrating how complex these dynamic features are. To further analyze the data for the RSL in each HiRISE image, we calculated the total darkened area, newly darkened area, persistent darkened area, and faded area as a function of slope orientation (Fig. 11). While RSL are found on all slope orientations, Fig. 11 demonstrates that NW-, N-, SW-, and S- are dominant. Since RSL start in bedrock, we also mapped the areal extent of bedrock as a function of slope orientation (Fig. 12a). We then compared the total amount of darkened area over a MY to the area of bedrock as a function of slope orientation. Fig. 12b shows little correlation between the normalized area of bedrock and RSL area. The total darkened area, newly-darkened rate, and fading rate are plotted on top of EMax in Fig. 13 to highlight that these parameters are qualitatively correlated with insolation. Additionally, Fig. 13c shows that RSL fading occurs throughout the RSL lengthening season, and 7

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 8. HiRISE image (ESP_056997_1685) of SW-facing (downhill to the bottom left on the images) Garni crater RSL at Ls 255.7° immediately after the MY34 dust storm. (a) Densely-packed RSL with numerous tributaries emanating from bedrock. Arrows point to low-albedo features that are likely dust devil tracks, which are typically not seen in Garni crater. (b) Zoomed-in portion of (a), highlighting the tributaries.

2 m baseline slopes at the spatial start or end of the RSL, respectively. Thus the starting and stopping RSL angles have larger errors than the mean slope values. The maximum amount of slope error over a 2 m baseline could be 9° if neighboring pixels had the maximum given error (20 cm) in opposite directions. However, errors appear to be much < 9° as the RSL start and stop angles have a standard deviation ~2.2°, compared to the standard deviation of the mean RSL slopes of 1.5°. The mean of the starting RSL angle of 32.6° is very similar to the 33° static angle of repose (Ewing et al., 2017). Additionally, the mean of the stopping RSL angle is 30.7° is similar to the 29° (Ewing et al., 2017) and 29.8° (Ermakov et al., 2019) dynamic angle of repose. To further investigate the mean slopes that were < 28°, we plotted these as a function of RSL length (Fig. 14b). The only RSL that have mean slope values < 28° were < 200 m long. Additionally, we binned the mean slope data over 50 m of length to calculate the average and standard deviation of each bin (Fig. 14b black solid line and green boxes). This showed that the mean slope does not appreciably change with length and that within a 95.5% confidence interval mean RSL slopes are not below 28°. Thus, RSL with a mean slope of < 28° are statistical outliers. While the errors of the stopping slopes are larger, we also plotted these against RSL length (Fig. 14c) and found a small correlation of smaller stopping slope with increasing RSL length. This was expected as RSL generally have a steeper starting slope than stopping slope. However, the data do not support that RSL with lengths > 350 m have stopping slopes that are < 28°. We investigated individual RSL that had mean slopes below 28°, and Fig. 15 displays one such example. These plots show that even over longer baselines that the low slope angles can persist. No obvious DEM artifacts are visible in many of these areas, however, the low-angle slopes are typically on sand sheets that have few features that can be used as tie points for SOCET SET DEM creation. Thus, it is difficult to ascertain whether these low slopes are due to DEM artifacts or are actually low-slope RSL. Overall, Fig. 14 shows that RSL start on steeper slopes that are near the static angle of repose of 32.6 ± 2.3°, have a mean slope of 31.3 ± 1.5° that does not vary with length, and end on gentler slopes of 30.7 ± 2.1° that are near the dynamic angle of repose. While some RSL have slopes < 28°, we cannot rule out the dry mechanism hypothesis because these low-slope RSL are statistical outliers. To determine if these low-slope RSL exist, a method is needed that produces a high-resolution DEM with quantifiable topographic errors.

Fig. 9. Garni crater with all RSL detections (red) and the slope slump (cyan). Note that the slope slump is shown with translucent shading so that RSL that overlap the slump can be seen. The background HiRISE image is ESP_040804_1685, acquired MY32 Ls 323.7°. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

quickly stops after RSL stop lengthening. At periods of low EMax, RSL are dormant. Over the 1.5 MYs of orthoimage data, significant interannual variations suggest that MY31 had many more RSL compared to MY32 (Figs. 10, 11 & 13).

4.3.2. Slopes The slopes that RSL occur on were then analyzed to determine if their slopes were above the angle of repose, which would favor dry processes (Dundas et al., 2017). The mean of the angle at which RSL start and stop and the mean slope value of the RSL were all > 28°, which is the minimum value of measured slip faces of martian dunes (Atwood-Stone and McEwen, 2013). However, 271 and 73 out of the 2910 RSL (or 9.3% and 2.5%) have stopping and mean slope angle value < 28°, respectively. The mean angle of the RSL slopes is calculated by finding the mean of all the slopes (using a 2 m baseline) within each of the mapped RSL. The RSL starting and stopping angles are the

4.3.3. Slope slump In the HiRISE image acquired MY32 Ls 76.7°, a large slope slump is present on the N-facing slope of Garni crater (Chojnacki et al., 2016; Stillman et al., 2017; Ojha et al., 2017; Stillman, 2018). We mapped this feature (Fig. 16) and found that it spanned an area of 3633 m2, was 8

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 10. Example of complex lengthening and fading of NE-facing RSL from nine HiRISE orthorectified images, where the center of the images is at a latitude and longitude of 11.5332°S, 290.3017°E. (a) This is the first HiRISE image of Garni crater, thus all the RSL are mapped as newly darkened. (b) RSL activity is significant with a large amount of newly-darkened area (lengthening) and a lesser but significant area is faded. (c) NE-facing RSL lengthening has ceased and fading now dominates. (d) All NE-facing RSL have faded. (e) While RSL activity started near Ls 315°, NE-facing RSL remain small until the large increase in daily insolation started near Ls 90°. (f) New individual NE-facing RSL lengthen, while the previous RSL remain static. Note that most of the fading in this image is due to slight pixel mismatches. (g) NE-facing RSL continue to lengthen, but mostly because of the presence of new individual RSL (as opposed to incremental lengthening of the older RSL, since many of those are incrementally fading). (i) Daily insolation peaks near the Ls of this image, thus NE-facing RSL are at their maximum and show significant incremental lengthening and fading. (j) By Ls 253° insolation nears its minimum and all RSL have faded.

Ls 145.4°). Its fading rate can be constrained to be > 31 m2/sol, which is the average amount for N-facing RSL during this season (Fig. 13). Moreover, there was no significant dust activity (over Garni crater) in MY32 when the slump faded. The starting and stopping slope angles of the slump were 30.1° and 28.6°, respectively, with a mean of 27.5 ± 2.7°. Thus, the mean slope value is (within error) broadly consistent with the lowest measured martian sand dune lee slope bin of 28–29° (Atwood-Stone and McEwen, 2013). Additionally, many areas in the lower portions of the slope slump have values less than the mean, with a minimum slope angle of 13.1° (Fig. 16b). While some of these low angles may be due to errors in the DEM, the mean slope does suggest that this slump is near or below the dynamic angle of repose. If it is below the angle of repose, it would suggest that this flow would

204 m long, and formed in < 27 sols. This constrained its newly-darkened area rate and lengthening rate to be > 135 m2/sol and 7.6 m/sol, respectively. None of these values specifically rule it out as an RSL. However, the morphology of this feature is different, as it does not start near bedrock and appears to initiate along a slope failure ~9 m in length. Additionally, similar nearby older features suggest that these flows move material (Chojnacki et al., 2016). Note that while few slope slumps have been detected, incremental lengthening has never been detected on a slope slump. Therefore, it is likely that this feature formed in less than a sol. Furthermore, this slope slump darkened more area of the N-facing slopes than all the combined N-facing RSL present at the time this feature occurred. While the slump continued to appear dark 27 sols later (Ls 88.9°), it then completely faded away in ≤118 sols (by 9

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 12. (a) Bedrock area as a function of slope-facing orientation and (b) total darkened area normalized by bedrock area. Note that, bedrock areas are provided in a supplemental file named “Garni_bedrock_area.csv”.

5.1. Observations and correlations of north-facing slopes We use quantitative mapping to determine when N-facing RSL (Fig. 17) start and stop lengthening. Assuming that lengthening occurs at the same season in both MY31 and 32 (this allows us to compare MY31 and 32 images that have a closer Ls spacing), we can constrain the start of lengthening to be between Ls 314.1°–323.7°. However, nonorthorectified HiRISE images acquired at the end of MY33 and the beginning of MY34 have been qualitatively mapped and show a significantly different start time (Fig. S4). These observations did not detect RSL in MY33 at Ls 331.3° or in MY34 at Ls 19.3°, with the first RSL detected at MY34 Ls 56.5°. Our quantitative mapping suggests that N-facing RSL stop lengthening between Ls 160.7°–191.5°. When RSL are lengthening, the newlydarkened area rate is always greater than the fading rate. The MY32 data between Ls 145.4°–177.7° show that the largest newly-darkened rate value is about an order of magnitude greater than the fading rate, while the MY31 data between Ls 160.7° and 191.5° show that the fading rate is higher than the newly-darkened rate. Thus, we speculate that the RSL likely stopped between Ls 177.7° and 191.5°. The dust opacity for N-facing RSL varies significantly at the times when RSL start and stop, but once N-facing RSL start lengthening the interannual variations are small and the magnitude of dust opacity is low (Figs. 7, 18 & S2). Additionally, EMax (Fig. 18) is much lower when RSL start lengthening (mean EMax is 13.7 ± 0.2 MJ/m2 for nNE-, N-, and nNW-facing over Ls 314.1–323.7°) compared to when RSL stop lengthening (mean EMax is 16.8 ± 0.1 MJ/m2 for nNE-, N-, and nNWfacing over Ls 160.7–191.5°). Similarly, RSL in Chryse and Acidalia Planitiae (CAP) also start lengthening at much lower insolation values compared to when they stop lengthening (Stillman et al., 2016). We conclude that RSL lengthening only occurs when the insolation (or by proxy, surface temperature) is increasing, and that a higher absolute value of insolation (or by proxy, surface temperature) leads to a greater newly-darkened area rate. While RSL are lengthening, the newly-darkened area rates are greater than the fading rate (Fig. 17). Once RSL stop lengthening, fading rates reach their maximum values (120–590 m2/sol). Newlydarkened area rates vary from 4 to 330 m2/sol and generally increase with EMax. Note that the newly-darkened area rate for MY32 Ls 314.1–326.3° appears to be artificially low because lengthening is

Fig. 11. Seasonal variations in (a) newly-darkened area, (b) persistently-dark area, (c) faded area, and (d) total darkened area, as a function of slope-facing orientation. The estimated error of the total darkened area is displayed in Fig. S4.

have had a high inertia, allowing the flow to continue even at low slope angles. 5. Analysis In this section, we compare our quantitative mapping results to dust opacity and EMax. While using dust opacity to determine downwelling infrared and the true shortwave insolation would be preferable, too many assumptions (such as the grain size of the dust and the concentration and altitude of the dust) would be needed to do so. Likewise, we do not attempt to constrain the surface temperature, as it very dependent on the thermal conductivity of the surface and subsurface, which could vary by orders of magnitude between a dry and wet RSL mechanism. Thus, we attempt to compare our quantitative results to parameters that have fewer assumptions built in. 10

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 13. (a) Total darkened area, (b) newly-darkened area per sol, and (c) faded area per sol are colour shaded at the Ls of each orthoimage as a function of slope orientation, where the absence of colour indicates a value below the minimum value. To enhance detail of varying RSL activity, the colorscale is logarithmic. The background grayscale shading is the time-integrated shortwave insolation per sol metric (EMax; see Fig. 6).

believed to have started during this interval. Additionally, the newlydarkened area rates and fading rates between MY32 Ls 88.9–145.4° are abnormally small (Fig. 17) and affected by a long observation cadence. We postulate that the newly-darkened area rates and fading rates are actually much higher within this interval, but many newly-darkened areas had faded before they were imaged, thereby lowering both the measured newly-darkened area rates and fading rates. This underscores the importance of re-imaging RSL sites during their active period to increase the precision of quantitative analysis. In Fig. 19, we use the coefficient of determination r2 to calculate the correlation of EMax with the total dark area, newly-darkened rate, and fading rate. Note that for the r2 and best fit lines, we only used data obtained when the insolation was increasing, as the r2 and best fit lines are much less correlated if all data were used. For N-facing slopes, we find good r2 and good fits for the total dark area and newly-darkened area rate (Fig. 19ab). The fading rate (Fig. 19c) is not as well correlated, but the less-correlated data points could be explained by the large interval between measurements, resulting in a low apparent fading rate (as well as a small fading rate at the beginning of the RSL lengthening season, when fewer persistent RSL are available to be faded). Overall, we conclude that N-facing RSL are correlated with the magnitude of EMax as long as it is increasing.

as long as it is increasing. NE- and NW-facing RSL should receive very nearly the same amount of insolation during every sol of the year (Figs. S6 & S8). However, NEfacing slopes are expected to reach slightly lower peak temperature as they receive their maximum insolation before noon, while NW-facing slopes receive their maximum insolation after noon when the downwelling infrared radiation is greater. However, regardless of latitude, RSL observations confirm that W-facing slopes are much more likely to possess RSL compared to E-facing slopes (Stillman, 2018). Garni crater is no exception to this rule, with NW-facing RSL possessing the greatest mapped areal extent of RSL compared to any other slope orientation. 5.3. Observations and correlations of south-facing slopes Our quantitative mapping shows that S-facing RSL start lengthening between Ls 145.4–160.7° with a mean EMax when RSL start lengthening of 8.6 ± 0.7 MJ/m2 (sSE-, S-, and sSW-facing). This time period occurs before major dust events (Figs. 21 & S2), which may explain why there appears to be little interannual variability in the few HiRISE images available for this season. Determining when S-facing RSL stop lengthening is more complicated. Both the qualitative observations (Fig. S4) and quantitative mapping show significant interannual and annual variations in timing. Such variations are likely real and due to significant local dust opacity variability because the S-facing RSL lengthening season occurs during the dusty time of year. Unlike N-facing RSL, where newly-darkened area rates are greater than fading area rates, S-facing mapping shows that newly–darkened area rates and fading area rates alternate. This leads to three detected instances where the total dark area decreases, but then later increases. There is no large variance in the dust opacity data (Figs. 7 & S2) in MY31 that might be causing the decrease between Ls 218.1–266.7° or that could trigger the greatest newly-darkened rate measured between 266.7 and 280.8°. We postulate that other

5.2. Observations and correlations of northeast- and northwest-facing slopes The behavior of the NE- (Figs. S5–S6), N- (Figs. 17–19), and NW(Figs. S7–S8) facing RSL are remarkably similar, with NE-facing RSL lengthening for a slightly longer period of time. RSL on these slope orientations start lengthening at Ls 314.1–323.7°, lengthen through a relative minimum in EMax near Mars aphelion, and then stop lengthening when EMax reaches its maximum value. Additionally, the total dark area, newly-darkened area rate, and fading area rate for NE-, N-, and NW-facing orientations are correlated with the magnitude of EMax 11

Icarus 335 (2020) 113420

D.E. Stillman, et al.

to occur between Ls 350.0–1.4°, which has a mean EMax of 10.7 ± 0.4 MJ/m2 (sSE-, S-, and sSW-facing). Thus, the start- and stoplengthening values of EMax are similar, with the stopping value being ~20% larger. In Fig. 22, we use the coefficient of determination r2 to calculate the correlation of EMax to the total dark area, newly-darkened area rate, and fading area rate. We correlated and fit per-sol values separately using (1) sols with increasing EMax only and (2) the full set of data. We found that the total dark area was correlated with all the EMax data, while newly–darkened area rates and fading area rates were less correlated. Overall, the amount of solar radiation certainly affects the total darkened area. However, variations in dust opacity appear to cause large changes in newly-darkened area rates and fading area rates. 5.4. Observations and correlations of southwest-facing slopes SW-facing RSL start lengthening when EMax begins increasing from its minimum at northern summer solstice (Ls 90°; Fig. S9 & S10). SWfacing RSL start at the lowest EMax value compared to any of the other slope orientations. The newly-darkened area rate then continues to increase with EMax until Ls 220°, and the newly-darkened area rate is always greater than the fading area rate. After Ls 220°, the fading area rate and newly-darkened area rate alternate as to which is larger until lengthening stops near Ls 5°. EMax peaks near perihelion (Ls ~251°), so SW-facing RSL are able to lengthen for a considerable time while EMax is decreasing. Thus, this behavior is similar to that observed for S-facing RSL. 5.5. Observations and correlations of west- and east-facing slopes W- (Figs. S11–S12) and E- (Figs. S13–S14) facing RSL lengthen for a considerable amount of time, but not when EMax is at its maximum yearly value. RSL on both slope-facing directions also lengthen through the absolute minimum of EMax. The correlation of seasonality with EMax of E- and W-facing slopes is different from other slope-facing directions. It is also anomalous that W-facing slopes do not dominate the darkened area, as W-facing slopes qualitatively appear to dominate at most RSL sites. W-facing slopes in Garni crater typically have few RSL, but in MY31 Ls 160.7° the RSL total darkened area was an order of magnitude greater than it was in any other image. Furthermore, the MY32 mapped darkened area was about an order of magnitude lower than typical for Ls 145° and 178°, but about an order of magnitude greater at Ls 324°. While W-facing slopes have large interannual variations, RSL on E-facing slopes appear to behave very similarly between MY31 and 32.

Fig. 14. (a) Histogram of starting and ending slopes and the mean slope angle of all 2910 RSL. Note that 2.6%, 2.5% and 9.3% of starting, mean, and ending slopes, respectively, are below 28°, which is lower than the minimum lee slope mapped by Atwood-Stone and McEwen (2013). The mean of the starting RSL angle is 32.6° with a standard deviation of 2.3°, and is thus similar to the 33° dynamic angle of repose (abbreviated to AoR in the legend) found by Ewing et al. (2017). Additionally, the mean of the stopping RSL angle is 30.7° with a standard deviation of 2.1° and is thus similar to the 29° and 29.8° static angle of repose found by Ewing et al. (2017) and Ermakov et al. (2019), respectively. (b) Mean slope angle versus RSL length shows that many RSL slopes have a mean slope > 28°. Additionally, the average and the standard deviations (solid black line and green boxes) were calculated by binning RSL every 50 m of length. This shows that the mean slope does not appreciably change with RSL length and that even when subtracting two standard deviations (95.5% confidence interval), mean RSL slopes never get below 28°. Thus, RSL with a mean slope of < 28° are statistical outliers. (c) Stopping slope angle versus RSL length scatterplot shows a slight trend (best fit is the black solid line) toward lower slopes with increasing RSL length. However, this is not robust, with an r2 value of 0.07 and a − 0.32 Spearman rank correlation coefficient. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

5.6. Observations and correlations of southeast-facing slopes SE-facing RSL possess the smallest areal extent of any slope orientation (Figs. 13 & S15). Additionally, the eSE-facing orientation (between ESE and E) is the only 15° circular sector of Garni crater where no RSL were observed. The SE-facing orientation does have a significant amount of bedrock outcrops, but few RSL form there. SEfacing RSL form around Ls 180° and continue to lengthen until the EMax reaches its maximum value around Ls 270° (Figs. S15–S16). The quantitative and qualitative observations show little interannual variation for SE-facing slopes (Figs. S4 & S15). 6. Discussion 6.1. Observations in regard to RSL mechanisms In the list below, we label our observations that support or contradict RSL formation and recharge mechanisms (discussed earlier in sections 2.1 and 2.2). 1) Interannual variations were common in the areal extent of RSL on many slope orientations between MY31 and 32 (Figs. 11, 13, 17, S5, S7,

undetected events could cause this fading and growth, such as early morning hazes or small localized dust events. Overall, our quantitative geostatistical analysis detects the last lengthening of the active season 12

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 15. Example of a low mean slope RSL. (a) Stretched HiRISE image of RSL #23 in ESP_028501_1685 (MY31 Ls 160.7°) in green outline (see supplemental file “RSL-stats-ESP_028501_1685.csv”), where box colour is the slope angle in every subfigure. Note that the RSL has a slope orientation of 2.1° east of north and thus is considered an N-facing RSL. (b) Variation in slope angle. (c) Elevation versus traverse distance. (d) Variations in slope angle versus traverse distance. Note, both (c) and (d) are colour coded by slope angle. The mapped slope along the majority of this RSL is ≤28° with a slope mean of ~25.8°. Thus, if the slopes are correct this RSL could not be triggered by a dry flow mechanism. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

6.2. Observations for and against the dry RSL hypothesis

S9 & S11), and the non-orthorectified data in MY33 and 34 show interannual variations with regard to when RSL start. 2) After the MY34 dust storm, RSL formed on every intercardinal and cardinal slope-facing direction (Figs. 8 & S4). Our quantitative data (Table 1) suggests that RSL should have only been present on three of the eight slope orientations. Additionally, numerous tributaries are detected (Fig. 8). 3) Our quantitative results demonstrated that RSL lengthening and fading occur on the same slopes and even on the same RSL concurrently (Figs. 10, 11, 13, 17, 19, 20, 22, S5, S7, S9, S11, S13 & S15). 4) The correlation of RSL darkened area to EMax (Figs. 18, 19, 21, 22, S6, S8, S10, S12, S14 & S16) and the threshold value of EMax needed to start RSL lengthening (Table 1) differ depending on the slope-facing direction. The correlation of RSL lengthening and EMax shows that NE-, N-, and NW-facing RSL are correlated with increasing EMax. Interestingly, northern mid–latitude RSL in Chryse and Acidalia Planitiae (CAP) also lengthen during increasing EMax (Stillman et al., 2016). CAP RSL and NE-, N-, and NW-facing Garni crater RSL also have similar seasonalities and are responding to similar insolation and dust opacities. However, S- and SW-facing RSL are correlated with both increasing and decreasing EMax. Interestingly, RSL in the southern mid-latitudes (SML) have similar seasonalities and also lengthen during increasing and decreasing EMax (Stillman and Grimm, 2018). 5) Slope slump fading rate is consistent with the fading rates of RSL. 6) Slope angle results for RSL (Figs. 14 & 15) show that some RSL have starting, mean, and stopping slopes that are < 28°. However, statistical analysis shows that these low-slope RSL are outliers. The unknown uncertainties in the DEM preclude detailed analysis to determine if these outliers may truly exist or are just artifacts of the DEM.

The primary argument for a dry mechanism is that RSL never extend to the base of a slope because RSL cannot flow when the scree slope becomes less than the dynamic angle of repose (Dundas et al., 2017). Our observation #6 shows that 9.3% of the ending RSL slope angle values at Garni crater are below those measured for martian sand dunes (Atwood-Stone and McEwen, 2013), which is consistent with the observations of Teblot et al. (2019) at other RSL sites. This appears to argue against dry flows, but it is unclear what the accuracy of the slopes based on the HiRISE-derived DEM is. Dundas et al. (2017) and Schaefer et al. (2019) used a low-pass filter over a 20-m baseline to smooth similar HiRISE-derived DEMs (Kirk et al., 2008). They found that most RSL never go below the angle of repose. However, even our mean slope values (which averages the 2 m baseline slopes over the entire mapped RSL) show 2.5% of RSL have slopes of < 28°. Fig. 15 shows that some RSL traverse decameters over low-angle slopes. While these low-slope RSL may be statistical outliers, this does not indicate that they are fictitious. Thus, a DEM with known errors and uncertainties is needed to determine if the slopes of these few RSL are indeed erroneous or not. Lastly, while our slope values are noisy with a 2 m baseline, we argue that important high frequency variability in slope is lost when they are smoothed with a 20 m baseline. A possible issue with the dry mechanism is that RSL would not fade quickly enough in regions with an apparently low abundance of dust. For instance, slope streaks take decades to fade away (Schorghofer et al., 2007; Bergonio et al., 2013). However, if we assume that the slump that formed in Garni crater was a dry flow, then its fading (observation #5) proves that dust fall in Garni crater is sufficient to fade dry RSL-like features in RSL-like timescales. The concurrent fading and lengthening of RSL (observation #3) is also consistent with the fading 13

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Fig. 16. (a) MAARSL mapping of a slope slump (transparent cyan) and all mapped RSL (red) to demonstrate that RSL had previously existed at the top of the slope slump. Background HiRISE image is ESP_040804_1685, acquired MY32 Ls 323.7°. (b) Slope slump figure overlaid on contoured slope angle values, demonstrating significant slope variability in the lower half of the slump. The mean slope of the slope slump is 27.5 ± 2.7°. Associated geostatistical values are given in the supplemental material as “SlopeSlump_geostats.xlsx”. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 17. MAARSL statistics for N-facing slopes, with the top graph demonstrating the RSL lengthening season based on MY31 and MY32 MAARSL data. N-facing RSL simultaneously have newly-darkened and fading areas and the amount of both increase until RSL stop lengthening. The newly-darkened area rates and fading rates appear to be comparable between the two MYs. Furthermore, the newly-darkened area rates appear to rise throughout the season as the shortwave insolation increases. Likewise, the fading rates also increase with insolation, but are always less than the newly-darkened area rate until the N-facing RSL stop lengthening.

14

Icarus 335 (2020) 113420

D.E. Stillman, et al.

change paths due to slight variations in deposition and incomplete removal of previous granular flows. This would leave locations that are no longer overprinted to fade, while the overprinted parts would continue to be relatively free of dust and dark. Similarly, the availability of dust could cause interannual variability in observed RSL (observation #1). The concurrent fading and lengthening of RSL (observation #3) is inconsistent with the explanation of dry dusty fading suggested by Schaefer et al. (2019) at Tivat crater. In Tivat crater, RSL were argued to be darker than their surroundings because dust had preferentially blown off those surfaces first. These RSL then faded when the dust on the surrounding slopes was removed later. This hypothesis does not work when one portion of an RSL fades between images, while another portion of the same RSL darkens/lengthens (Fig. 10). If RSL are constantly overprinting, but not transporting any material at HiRISE scales over six MY, then the dry flows must be transporting very little material or it must be material that could be blown away. Thus, the yearly transport of an effective layer of dust tens of microns thick could satisfy the observed fading and darkening. An influx of dust via the MY34 dust storm would significantly refresh and possibly oversupply this shallow dust layer and lead to greater RSL activity (observation #2). Additionally, the widespread increase in the tributaries shows a well-distributed dust source (observation #2). Moreover, interannual variability and reduction in RSL at the beginning of MY32, possibly MY33, and at the beginning of MY34 may have been due to a lack of dust availability. The greatest unknown for the dry RSL mechanism is finding a way to repeatedly trigger dry flows. Additionally, if RSL are overprinting then they must trigger multiple times a year. Schmidt et al. (2017) described a Knudsen pump mechanism that may work, although the seasonality does not match newer results (Stillman, 2018). The Knudsen pump mechanism requires a large solar insolation that is quickly removed via a shadow. However, the conditions during or immediately after the MY34 dust storm do not suggest a large solar insolation for Nfacing slopes. Vincendon et al. (2019) and Schaefer et al. (2019) hypothesize that RSL occur when removable dust is available to be lifted by wind. Additional modeling of these hypotheses needs to explain how wind can be channeled into such narrow channels (Fig. 8), produce incremental lengthening over hundreds of sols, and create lengthening RSL during periods of both low and constant dust opacity (NE-, N-, and NW-facing RSL) and high and variable dust opacity (S- and SW-facing RSL). Regardless, a new dry model would need to fit our observation #4 that NE-, N-, and NW-facing RSL behave differently with respect to EMax than S- and SW-facing RSL. Additionally, the dry triggering model would also have to explain why RSL do not simply occur on all slopes (Ojha et al., 2014) that have dust coatings that are greater than the angle of repose. Overall, dry dusty flows can explain recharge and fading via atmospheric dust deposition, darkening by the removal of dust, a strong response after the MY34 dust storm, and the lack of topographic evidence of material being transported (as the dust is just blown away). However, no existing dry model can reproduce a triggering mechanism that allows such dust flows to increase in size and overprint the existing topography for a significant portion of a Mars year. Additionally, if the low slope angles measured are not DEM artifacts, then RSL cannot be dry unless the dry mechanisms can also impart the necessary momentum to continue to flow through these below-angle-of-repose slope portions.

Fig. 18. Insolation on 32.5° N-facing slopes and dust opacity variations, as a function of season. Additionally, the MAARSL-derived RSL lengthening seasonality is noted on the figure. Note that the seasonality of lengthening best corresponds to increasing EMax rather than the absolute value of EMax. Also, Nfacing RSL are not lengthening when atmospheric dust opacity is high.

Fig. 19. Correlation of EMax and (a) total darkened area, (b) newly-darkened rate, (c) fading rate for N-facing slopes. Note that the fits and r2 only take into account data points with increasing insolation, as that appears to be the parameter most correlated with RSL activity. Additionally, the fits and r2 start at the EMax when RSL start lengthening (13.7 ± 0.2 MJ/m2).

of the slump (observation #5). It further suggests that RSL do not incrementally lengthen at the bottom, because if so then RSL would fade from the top to the bottom. Instead, RSL appear to overprint previous flows with the subsequent flows getting longer. The sequence of RSL displayed in Fig. 10 could be the imaging of significant rapid fading of RSL and overprinting. In this figure dry overprinting is increasing in size up to Ls 160.7°, but decreasing at Ls 191.5°. Overprinting could

6.3. Observations for and against the wet RSL hypothesis If more accurate DEMs (with fewer or better understood artifacts) and additional mapping at other sites (e.g., Teblot et al., 2019) confirm that RSL routinely flow on slopes that are below the angle of repose (observation #6), then a wet RSL formation mechanism (which easily allows flows on lower-angle slopes) provides a way to explain those 15

Icarus 335 (2020) 113420

D.E. Stillman, et al.

Table 1 MAARSL-derived RSL seasonality (in Ls) versus slope-facing direction in Garni crater. Note that W-facing RSL lengthen two times per Mars year. The fourth and fifth columns show the mean EMax (time-integrated shortwave insolation per sol metric) for when RSL start and stop lengthening. There appears to be little correlation between the mean EMax and when RSL start and stop lengthening. NW-, N-, NE-, and SE-facing RSL lengthen with increasing EMax, and all but the last also continue to lengthen through a local minimum near Mars aphelion. S-, SW-, W-, and E-facing RSL lengthen with both increasing and decreasing EMax. Additionally, W- and Efacing RSL lengthen at their absolute minimum EMax value, but do not lengthen at the time of their absolute maximum EMax. Slope facing direction

RSL start lengthening (Ls °)

RSL stop lengthening (Ls °)

EMax when RSL start (MJ/m2)

EMax when RSL stop (MJ/m2)

Lengthening compared to trend and extrema of EMax

NW W1 W2 SW S SE E

314.1–323.7 64.6–76.7 266.7–270.8 88.9–133.1 160.7–177.7 177.7–191.5 270.8–280.8

160.7–191.5 235.4–245.8 1.4–9.5 1.4–9.5 350.0–1.4 270.8–280.8 191.5–218.1

15.1 ± 0.6 10.0 ± 1.0 18.8 ± 0.9 6.8 ± 1.2 8.6 ± 0.7 14.9 ± 0.8 18.7 ± 0.9

16.7 19.0 13.9 12.0 10.7 21.3 17.9

NE N

314.1–323.7 314.1–323.7

191.5–218.1 160.7–191.5

15.1 ± 0.6 13.7 ± 0.2

16.9 ± 0.4 16.8 ± 0.1

Increasing insolation & local min Increasing & decreasing insolation, and at absolute min, but not at max insolation Increasing & decreasing insolation Increasing & decreasing insolation Increasing insolation Increasing & decreasing insolation, and at absolute min, but not at max insolation Increasing insolation & relative min Increasing insolation & relative min

observations. If a better DEM shows that RSL do always flow on slopes > 28°, it may just be that these slopes formed via mass wasting have created the morphology that RSL then flow through and that RSL never reach below-angle-of-repose slopes because they are volumelimited. Additionally, concurrent fading and lengthening (observation #3) would be expected for wet flows and would occur when evaporation rates are high and the flow takes slightly different flow paths (possibly caused by loss of permeability due to precipitation of salt) that would funnel water to different parts of the RSL at different times. Interannual variation (observation #1) could be explained by variations in surface temperature that would be caused by variations in dust opacity. However, in our analysis, we could not explain all variability with the current dust opacity maps gridded at 3 × 3°. This suggests that higher-resolution data is needed before such variations could be used to rule out or confirm this hypothesis. Additionally, RSL occurred on every slope after the MY34 dust storm and even on many slopes that should not have had RSL in a “typical” year (observation #2). We know that the slopes that should not have had RSL are very sensitive to increasing EMax (like NE-, N-, and NW-facing slopes).

± ± ± ± ± ± ±

0.3 0.7 0.5 0.7 0.4 0.6 0.5

Fig. 21. Insolation on 32.5° S-facing slopes and dust opacity variations with season. Additionally, the MAARSL-derived RSL lengthening seasonality is noted on the figure. Note that the seasonality of lengthening is nearly symmetrical with EMax. Also, S-facing RSL lengthen when atmospheric dust opacity is high.

Fig. 20. MAARSL statistics for south-facing slopes, with the top graph demonstrating the RSL lengthening season based on MY31 and MY32 MAARSL data. Southfacing RSL continue to lengthen well after the maximum value of EMax has been reached. Furthermore, there is not a clear dominance between newly-darkened area rate and fading rate. Additionally, very high newly-darkened area rates and fading rates were also detected. 16

Icarus 335 (2020) 113420

D.E. Stillman, et al.

area, a large insolation value when RSL are long may cause these RSL to enter an equilibrium flow regime (Grimm et al., 2014; Stillman et al., 2016). To explain why the slope slump fading rate is consistent with the fading rates of RSL (observation #5), Stillman et al. (2017) speculated that slumps may remove dry overburden, exposing a relatively waterrich darker regolith layer that then desiccated over ∼30–170 sols. Interestingly, all of the observed slope slumps (~10 in Juventae Chasma and ~2 in Garni crater) occurred in the same season (Ls 0–120°; Ojha et al., 2017). Additionally, Ojha et al. (2017) reported that there may be a correlation between the formation of these features and the presence of atmospheric obscurations with H2O ice near times when the slumps form. However, if slope slumps are a purely dry process then this would suggest that dark albedo features within Garni crater could fade due to both dust deposition and evaporation. Overall, briny shallow subsurface flows are consistent with belowangle-of-repose slope angles and concurrent RSL lengthening and fading. Arguments can also be made to suggest why interannual variability exists, why correlations and threshold of EMax vary between different slope orientations, how slope slumps fade at the same rate as RSL, and why so many RSL exist after a dust storm. However, the most significant problem with briny RSL flows is accessing a source of briny water and removing excess salt from the regolith. 6.4. 6.3. Future modeling, data products, and observations needed Overall, better modeling, better derived data products, and additional observations are needed to help determine if RSL are formed via a dry or wet mechanism. In particular, dry models that can repeatedly trigger long flows of dust or very fine sediment are needed to bolster the dry mechanism. Better wet models are needed to determine the significant differences with respect to insolation of NE-, N-, and NW-facing and S- and SW-facing RSL. HiRISE-scale DEMs are needed that are more accurate, and have better estimates of errors in the slope topography. Furthermore, more orthoimages are needed and additional quantitative mapping needs to be performed at numerous sites. This quantitative mapping can then be used to further evaluate dry and wet models. In order to get the best quantitative mapping results, more HiRISE data is needed with fewer sols between images. While this is extremely difficult given the limited bandwidth and numerous interesting features on Mars, we would plead that HiRISE monitoring of a few excellent RSL sites be augmented to allow more frequent imaging (with a spacing of < 55 sols).

Fig. 22. Correlation of (a) total-darkened area, (b) newly-darkened rate, (c) fading rate and time-integrated shortwave insolation per sol for S-facing slopes. Note that the black line and annotations are for fits and r2 that only take into account data points with increasing EMax and > 8.6 MJ/m2, while the blue line and annotations are for fits that use EMax data that are > 8.6 MJ/m2 (which is the EMax when S-facing RSL start lengthening).F. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Therefore, we suggest that as the dust opacity increased, the downwelling infrared energy, and the shortwave insolation could have actually increased the total received energy at the surface to above the threshold value (~14 MJ/m2 per sol for NE-, N-, and NW-facing slopes). This might allow these RSL to start lengthening anomalously. Furthermore, during the dust storm it is possible for mean surface temperatures to increase due to the increase in downwelling infrared energy, so heat could penetrate deeper into the subsurface, possibly accessing additional sources for wet flows. Such additional sources could produce the enhanced density of tributaries that are detected (observation #2). Additionally, a new layer of dust on the surface would reduce the vapor flow out of the subsurface, thus reducing the evaporation rate of a water-rich RSL. The variability in the correlation of RSL darkened area to EMax and the threshold value of EMax needed to start RSL lengthening differs between slope-facing directions (observation #4), suggesting that NE-, N-, and NW-facing RSL are different from the S- and SW-facing RSL. These changes may occur because the seasonal total insolation (or by proxy, near surface temperature) is so different. S- and SW-facing RSL receive a shorter-duration, larger temperature pulse compared to those on NE-, N-, and NW-facing slopes. Thus, the depth of the ice dams that would need to be melted on these slopes would be different, leading to different correlations and threshold values. The salinity of the water could also be different, but this would be unlikely if wet RSL access an aquifer. Additionally, because RSL loss rate is proportional to surface

7. Conclusion We performed quantitative mapping of RSL in Garni crater, located in Melas Chasma of Valles Marineris (Fig. 1). To reduce the effort involved in manually mapping each RSL in each HiRISE image, we developed MAARSL to analyze a set of orthorectified HiRISE images along with a DEM to detect candidate RSL, compute descriptive statistics, characterize changes over time, and interactively filter candidates (Figs. 2 & 3). However, varying illumination conditions in different HiRISE images of the same region and the lower spatial resolution of HiRISE DEMs (1.01 m) pose challenges for automated detection, particularly for narrow RSL. Thus, to ensure every RSL was accounted for, many additional RSL were manually mapped. Using MAARSL combined with manual mapping, we identified 2910 RSL in 22 orthoimages that cover MY31 Ls 133.0° to MY32 Ls 323.7° (Fig. 9). RSL were mapped in every 15° circular sector of Garni crater with the exception of the eSE sector, in which no RSL form (Figs. 9 & 13). We quantified that RSL lengthening and fading occur concurrently on slopes with the same orientation and even within the same RSL (Fig. 10). Slope angles of RSL (Fig. 14) and of the slope slump (Fig. 16) show that many RSL start, end, and have significant portions at angles that are below the angle of repose, similar to the findings of Teblot et al. 17

Icarus 335 (2020) 113420

D.E. Stillman, et al.

References

(2019). Quantitative data was used on every slope-facing direction to determine when the RSL started and stopped lengthening (Table 1). Interannual variations were common in the areal extent of RSL on many slope orientations between MY31 and 32, and the non-orthorectified data in MY33 and 34 show interannual variations with regard to when RSL start. The RSL lengthening seasons were compared and correlated to dust opacity and the theoretical maximum (if Mars had no atmosphere) time-integrated shortwave insolation per sol (EMax). The correlation of RSL lengthening and EMax shows that NE-, N-, and NW-facing RSL are correlated with increasing EMax (Figs. 17–19, S5–S8), similar to CAP RSL. Meanwhile, S- and SW-facing RSL are correlated with both increasing and decreasing EMax (Figs. 20–22, S9–S10), similar to SML RSL. These differences, and also the threshold value of EMax needed to start RSL lengthening (Table 1) between slope-facing directions, need to be explained in any model of RSL formation. Shortly after the MY34 dust storm cleared, RSL are clearly visible on every intercardinal and cardinal slope-facing direction (Figs. 8 & S4). Our quantitative data (Table 1) suggests that RSL should have only formed on three of the eight slope orientations. Additionally, numerous tributaries were detected (Fig. 8). This confirms that dust (atmospheric and/or surface) strongly modulates RSL formation. A slope slump was also mapped and its fading rate (while poorly constrained) is consistent with RSL fading. If the slope slump is a purely dry feature then this suggests that atmospheric dust deposition could be causing RSL to fade. Our observations found that dry dusty flows can explain recharge and fading via atmospheric dust deposition, darkening by the removal of dust, a strong response after the MY34 dust storm, and the lack of topographic evidence of material being transported (as the dust is just blown away). However, no existing model can reproduce a triggering mechanism that allows such dust flows to increase in length and overprint the existing topography for a significant portion of a Mars year. The briny shallow subsurface flows implicated in a wet mechanism are consistent with below-angle-of-repose slopes and concurrent lengthening and fading. However, the most significant problem with briny RSL flows is accessing a source of briny water and removing excess salt from the regolith. More complex RSL formation models are needed to fit these complex observations, and a better understanding of DEMs is needed to determine if the scree slopes that RSL occur on are below or above the angle of repose. Lastly, while RSL sites have been imaged repeatedly by HiRISE, additional images are needed - preferably with a small number of sols between images – in order to continue mapping how RSL behave spatially, seasonally, and how they change interannually (e.g., after the significant MY34 dust storm).

Abotalib, A.Z., Heggy, E., 2019. A deep groundwater origin for reoccurring slope lineae on Mars. Nat. Geosci. 12, 235–241. Atwood-Stone, C., McEwen, A.S., 2013. Avalanche slope angles in low-gravity environments from active Martian sand dunes. Geophys. Res. Lett. 40 (12), 2929–2934. https://doi.org/10.1002/grl.50586. Bergonio, J.R., Rottas, K.M., Schorghofer, N., 2013. Properties of martian slope 643 streak populations. Icarus 225 (1), 194–199. https://doi.org/10.1016/j.icarus.2013.03.023. Breiman, L., 2001. Random forests. Mach. Learn. 45 (1), 5–32. Chojnacki, M., McEwen, A., Dundas, C., Ojha, L., Urso, A., Sutton, S., 2016. Geologic context of recurring slope lineae in Melas and Coprates Chasmata, Mars. J. Geophys. Res. Planets 121, 1204–1231. https://doi.org/10.1002/2015JE004991. Christensen, P.R., et al., 2001. Mars global surveyor thermal emission spectrometer experiment: investigation description and surface science results. J. Geophys. Res. 106 (E10), 23,823–23,871. Christensen, P.R., Jakosky, B.M., Kieffer, H.H., Malin, M.C., McSween Jr., H.Y., Nealson, K., Mehall, G.L., Silverman, S.H., Ferry, S., Caplinger, M., Ravine, M., 2004. The Thermal Emission Imaging System (THEMIS) for the Mars 2001 odyssey mission. Space Sci. Rev. 110, 85–130. Cohn, D., Ghahramani, Z., Jordan, M.I., 1996. Active learning with statistical models. J. Artif. Intell. Res. 4 (1), 129–145. Dundas, C., et al., 2017. Granular flows at recurring slope lineae on Mars indicate limited role for liquid water. Nat. Geosci. 10, 903–907. Edwards, C.S., Piqueux, S., 2016. The water content of recurring slope lineae on Mars. Geophys. Res. Lett. 43. https://doi.org/10.1002/2016GL070179. Ermakov, A.I., Kreslavsky, M., Scully, J.E.C., Hughson, K.H.G., Park, R.S., 2019. Surface roughness and gravitational slope distributions of Vesta and Ceres. J. Geophys. Res. Planets 124, 14–30. https://doi.org/10.1029/2018JE005813. Ewing, R.C., et al., 2017. Sedimentary processes of the Bagnold Dunes: implications for the eolian rock record of Mars. J. Geophys. Res. Planets 122, 2544–2573. https://doi. org/10.1002/2017JE005324. Grimm, R.E., Harrison, K.P., Stillman, D.E., 2014. Water budgets of Martian recurring slope lineae. Icarus 233, 316–327. Guzewich, S.D., Lemmon, M., Smith, C.L., Martínez, G., de Vicente-Retortillo, Á., Newman, C.E., et al., 2019. Mars science laboratory observations of the 2018/Mars year 34 global dust storm. Geophys. Res. Lett. 46, 71–79. https://doi.org/10.1029/ 2018GL080839. Hill, J., Edwards, C.S., Christensen, P.R., 2014. Mapping the Martian surface with THEMIS infrared global mosaics. In: 8th International Conference on Mars, Pasadena, CA, Abs, pp. 1141. Horn, B.K.P., 1981. Hill shading and the reflectance map. Proc. IEEE 69 (1), 14–47. Kirk, R.L., Howington-Kraus, E., Rosiek, M.R., Anderson, J.A., Archinal, B.A., Becker, K.J., Cook, D.A., Galuszka, D.M., Geissler, P.E., Hare, T.M., Holmberg, I.M., Keszthelyi, L.P., Redding, B.L., Delamere, W.A., Gallagher, D., Chapel, J.D., Eliason, E.M., King, R., McEwen, A.S., 2008. Ultrahigh resolution topographic mapping of Mars with MRO HiRISE stereo images: meter-scale slopes of candidate Phoenix landing sites. J. Geophys. Res. 113, E00A24. Leask, E. K., Ehlmann, B. L., Dundar, M. M., Murchie, S. L., & Seelos, F. P., (2018). Challenges in the search for perchlorate and other hydrated minerals with 2.1-μm absorptions on Mars. Geophys. Res. Lett., 45, 12,180–12,189. doi:https://doi.org/10. 1029/2018GL080077. Levy, J., 2012. Hydrological characteristics of recurrent slope lineae on Mars: evidence for liquid flow through regolith and comparisons with Antarctic terrestrial analogs. Icarus 219, 1–4. https://doi.org/10.1016/j.icarus.2012.02.016. Malin, M.C., Cantor, B.A., Britton, A.W., 2018a. MRO MARCI weather report for the week of 11 June 2018–17 June 2018. In: Malin Space Science Systems Captioned Image Release, MSSS-535, Retrieved from. http://www.msss.com/msss_images/2018/06/ 20/. Malin, M.C., Cantor, B.A., Britton, A.W., 2018b. MRO MARCI Weather Report for the Week of 18 June 2018–24 June 2018, Malin Space Science Systems Captioned Image Release, MSSS-536. Retrieved from. http://www.msss.com/msss_images/2018/06/ 27/. Malin, M.C., Cantor, B.A., Britton, A.W., 2018c. MRO MARCI Weather Report for the Week of 2 July 2018–8 July 2018, Malin Space Science Systems Captioned Image Release, MSSS-538. Retrieved from. http://www.msss.com/msss_images/2018/07/ 11/. Malin, M.C., Cantor, B.A., Britton, A.W., 2018d. MRO MARCI Weather Report for the Week of 16 July 2018–22 July 2018, Malin Space Science Systems Captioned Image Release, MSSS-540. Retrieved from. http://www.msss.com/msss_images/2018/07/ 25/. Malin, M.C., Cantor, B.A., Britton, A.W., 2018e. MCO MARCI weather report for the week of 3 September 2018–9 September 2018, Malin space science systems captioned image release, MSSS-547. Retrieved from. http://www.msss.com/msss_images/ 2018/09/12/. Massé, M., et al., 2016. Transport process induced by metastable boiling water under Martian surface conditions. Nat. Geosci. 9, 425–428. McCleese, D.J., Schofield, J.T., Taylor, F.W., Calcutt, S.B., Foote, M.C., Kass, D.M., Leovy, C.B., Paige, D.A., Read, P.L., Zurek, R.W., 2007. Mars climate sounder: an investigation of thermal and water vapor structure, dust and condensate distributions in the atmosphere, and energy balance of the polar regions. J. Geophys. Res. 112, E05S06. https://doi.org/10.1029/2006JE002790. McEwen, A.S., 2018. Are recurring slope lineae habitable. In: From Habitability to Life on Mars, pp. 249–274. McEwen, A.S., et al., 2007. Mars reconnaissance orbiter's High Resolution Imaging

Funding Funding source is given in the first sentence of the acknowledgements NASA Mars Data Analysis Grant NNX16AJ47G Acknowledgements This work was performed under NASA Mars Data Analysis Grant NNX16AJ47G. We thank Mikhail Kreslavsky and Thomas Heyer for their thorough reviews that greatly improved the paper. We also thank Colin Dundas for useful discussions regarding the slopes of RSL and Matthew Wu for his contributions toward terrain analysis that helped inform the automatic RSL candidate detection method. Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.icarus.2019.113420. 18

Icarus 335 (2020) 113420

D.E. Stillman, et al.

streak activity on Mars. Icarus 191 (1), 132–140. Stillman, D.E., 2018. Unraveling the mysteries of recurring slope lineae. In: Dynamic Mars, pp. 51–85. Stillman, D.E., Grimm, R.E., 2018. Two pulses of seasonal activity in Martian southern mid-latitude recurring slope Lineae (RSL). Icarus 302, 126–133. Stillman, D.E., Michaels, T.I., Grimm, R.E., Harrison, K.P., 2014. New observations of martian southern mid-latitude recurring slope lineae (RSL) imply formation by freshwater subsurface flows. Icarus 233, 328–341. Stillman, D.E., Michaels, T.I., Grimm, R.E., Hanley, J., 2016. Observations and modeling of northern mid-latitude recurring slope lineae (RSL) suggest recharge by a presentday martian briny aquifer. Icarus 265, 125–138. https://doi.org/10.1016/j.icarus. 2015.10.007. Stillman, D.E., Michaels, T.I., Grimm, R.E., 2017. Characteristics of the numerous and widespread recurring slope lineae (RSL) in Valles Marineris, Mars. Icarus 285, 195–210. https://doi.org/10.1016/j.icarus.2016.10.025. Teblot, M., Schorghofer, N., Groudge, T., Levy, J., 2019. Slope, elevation and thermal inertia trends of recurring slope lineae: RSL initiation and termination points fall outside the angle of repose. In: 50th Lunar and Planet. Sci. Conf., March 18–22, the Woodlands, 1561.pdf. Vincendon, M., Pilorget, C., Carter, J., Stcherbinine, A., 2019. Observational evidence for a dry dust-wind origin of Mars seasonal dark flows. Icarus 325, 115–127. Watkins, et al., 2014. Structurally controlled subsurface fluid flow as a mechanism for the formation of recurring slope lineae. In: 45th Lunar and Planet. Sci. Conf. The Woodlands, pp. 2911 March 17-21. Zuber, M.T., Smith, D.E., Solomon, S.C., Muhleman, D.O., Head, J.W., Garvin, J.B., Abshire, J.B., Bufton, J.L., 1992. The Mars observer laser altimeter investigation. J. Geophys. Res. 97, 7781–7797.

Science Experiment (HiRISE). J. Geophys. Res. 112, E05S02. McEwen, A.S., et al., 2011. Seasonal flows on warm martian slopes. Science 333, 740–743. Michaels, T.I., Rafkin, S.C.R., 2008. Meteorological predictions for candidate 2007 Phoenix Mars Lander sites using the Mars Regional Atmospheric Modeling System (MRAMS). J. Geophys. Res. 113, E00A07. https://doi.org/10.1029/2007JE003013. Montabone, L., Forget, F., Millour, E., Wilson, R.J., Lewis, S.R., Cantor, B., Kass, D., Kleinböhl, A., Lemmon, M.T., Smith, M.D., Wolff, M.J., 2015. Eight-year climatology of dust optical depth on Mars. Icarus. https://doi.org/10.1016/j.icarus.2014.12.034. Ojha, L., et al., 2014. HiRISE observations of recurring slope lineae (RSL) during southern summer on Mars. Icarus 231, 365–376. Ojha, L., et al., 2015. Spectral evidence for hydrated salts in recurring slope lineae on Mars. Nat. Geosci. 8, 829–832. Ojha, L., Chojnacki, M., McDonald, G.D., Shumway, A., Wolff, M.J., Smith, M.D., Toigo, A., 2017. Seasonal slumps in Juventae Chasma, Mars. J. Geophys. Res. Planets 122, 2193–2214. https://doi.org/10.1002/2017JE005375. Powell, K.E., Arvidson, R.E., 2017. Geologic setting and mineralogy of Horowitz crater. In: 48th Lunar Planet. Sci. Conf., Mar 20–24, the Woodlands, TX, 1489. Raack, J., Conway, S.J., Herny, C., Balme, M.R., Carpy, S., Patel, M.R., 2017. Water induced sediment levitation enhances downslope transport on Mars. Nat. Commun. 8, 1151. Schaefer, E.I., McEwen, A.S., Sutton, S.S., 2019. A case study of recurring slope lineae (RSL) at Tivat crater: implications for RSL origins. Icarus 317, 621–648. Schmidt, F., Andrieu, F., Costard, F., Kocifaj, M., Meresescu, A.G., 2017. Formation of recurring slope lineae on Mars by rarefied gas-triggered granular flows. Nat. Geosci. 10, 270–273. Schorghofer, N., Aharonson, O., Gerstell, M.F., Tatsumi, L., 2007. Three decades of slope

19