A consistent global approach for the morphometric characterization of subaqueous landslides

Abstract Landslides are common in aquatic settings worldwide, from lakes and coastal environments to the deep sea. Fast-moving, large-volume landslides can potentially trigger destructive tsunamis. Landslides damage and disrupt global communication links and other critical marine infrastructure. Landslide deposits act as foci for localized, but important, deep-seafloor biological communities. Under burial, landslide deposits play an important role in a successful petroleum system. While the broad importance of understanding subaqueous landslide processes is evident, a number of important scientific questions have yet to receive the needed attention. Collecting quantitative data is a critical step to addressing questions surrounding subaqueous landslides. Quantitative metrics of subaqueous landslides are routinely recorded, but which ones, and how they are defined, depends on the end-user focus. Differences in focus can inhibit communication of knowledge between communities, and complicate comparative analysis. This study outlines an approach specifically for consistent measurement of subaqueous landslide morphometrics to be used in the design of a broader, global open-source, peer-curated database. Examples from different settings illustrate how the approach can be applied, as well as the difficulties encountered when analysing different landslides and data types. Standardizing data collection for subaqueous landslides should result in more accurate geohazard predictions and resource estimation.

The importance of subaqueous landslides for society, economy and ecology Terrestrial landslides are important agents for the transport of sediment and organic carbon (Korup et al. 2007;Hilton et al. 2008). They can dramatically modify landscapes and ecosystems (Keefer 1984;Swanson et al. 1988;Walker et al. 2009), and pose a hazard to critical infrastructure and human life (Petley 2012). High-resolution and regular satellite mapping, real-time monitoring, personal accounts, news reports, and even social media trends are used to record terrestrial landslide activity, thus providing valuable and temporally-constrained information that forms the basis of extensive landslide databases and catalogues (Malamud et al. 2004;Petley et al. 2005;Korup et al. 2007;Kirschbaum et al. 2010;Petley 2012;Klose et al. 2014;Pennington et al. 2015;Taylor et al. 2015). These databases can be interrogated to quantify preconditioning and triggering mechanisms, understand risk profiles for different regions, assess the extent and nature of ancient events, calibrate numerical models of slope stability, and inform forecasts of future landslide activity. Indeed, many countries now have operational real-time terrestrial landslide forecast systems in place (e.g. Chen & Lee 2004;Baum & Godt 2010).
Landslides that occur in subaqueous settings (ranging from lakes and coastal regions to the deep sea) are also societally, economically and ecologically important, yet our understanding of them is much less well developed than for their onshore equivalents . Subaqueous landslides can be many orders of magnitude larger than terrestrial landslides (Korup et al. 2007), transporting up to thousands of cubic kilometres of sediment (Moore et al. 1989(Moore et al. , 1994Watts & Masson 1995;Collot et al. 2001;Haflidason et al. 2004;Masson et al. 2006;Day et al. 2015) and large volumes of exhumed organic carbon (St-Onge & Hillaire-Marcel 2001;Smith et al. 2015;Azpiroz-Zabala et al. 2017). Submarine and sublacustrine landslides often generate long runout flows, which damage strategically important seafloor infrastructure including telecommunication cables, production platforms and hydrocarbon pipelines (Piper et al. 1999;Mosher et al. 2010b;Thomas et al. 2010;Carter et al. 2014;Forsberg et al. 2016;Pope et al. 2017). Tsunamis generated by subaqueous landslides threaten many coastal communities and have caused large numbers of fatalities (Tappin et al. 2001;Ward 2001;Harbitz et al. 2014). Low-lying Small Island Developing States, such as those in the South Pacific, are particularly at risk from locally-sourced tsunamis, but little is currently known about the scale, location and recurrence of tsunamigenic landslides in those areas (Goff & Terry 2016). Under burial, subaqueous landslide deposits are recognized as an important element of hydrocarbon systems: conditioning reservoir distribution (Armitage et al. 2009;Kneller et al. 2016), acting as seals (Cardona et al. 2016) and as potential reservoirs (Meckel 2011;Henry et al. 2017). Furthermore, heterogeneous buried landslides can compromise seal integrity and rearrange subsurface fluid plumbing systems (Gamboa et al. 2011;Riboulot et al. 2013;Maia et al. 2015). The extent of submarine landslide deposits informs the placement of international economic boundaries, as defined by the United Nations Convention on Law of the Sea (e.g. Mosher et al. 2016). The top surfaces of mass failure deposits and areas of evacuation scarring that result from subaqueous landslides are increasingly being recognized as important habitats for seafloor biological communities (Okey 1997;De Mol et al. 2007;Paull et al. 2010;Chaytor et al. 2016a;Huvenne et al. 2016;Savini et al. 2016). The direct impacts of subaqueous landslide activity may also disturb and modify seafloor ecology, and have been suggested as a mechanism for the dispersal of species between isolated islands, thus governing their local evolution (Caujapé-Castells et al. 2017). Subaqueous landslides are therefore relevant to a large number of disciplines, governments and industries, as clearly underlined in numerous papers in the predecessor volumes to this special publication (Solheim 2006;Lykousis et al. 2007;Mosher et al. 2010a;Yamada et al. 2012;Krastel et al. 2014;Lamarche et al. 2016).

Value of a global consistent database of subaqueous landslides
Despite their importance, the study of subaqueous landslides is challenging due to their hard-to-reach nature: often in deep water and far from shore.
Step-increases in knowledge have been achieved over the past few decades, however. These are largely as a result of improvements in offshore surveying technologies (enhanced coverage, resolution and accuracy: Hughes Clarke 2018; Mountjoy & Micallef 2018), coupled with increased offshore resource exploration activities (Thomas et al. 2010), and recognition of the need to quantify the risk posed by subaqueous landslide hazards (Vanneste et al. 2014;Moore et al. 2018). Some of the major national and international programmes that catalysed this knowledge growth include GLORIA and STRATA-FORM (offshore USA), Seabed Slope Process in Deep Water Continental Margin (northwest Gulf of Mexico), STEAM and ENAM II (European Atlantic Margins), and COSTA (Mediterranean and NE Atlantic) (Nittrouer 1999;Locat & Lee 2002;Canals et al. 2004;Mienert 2004).
The IGCP-585, IGCP-511 and IGCP-640 projects helped to build an international community of subaqueous landslide researchers with diverse technical backgrounds who have documented a large number of subaqueous landslide studies from a range of physiographical, tectonic and sedimentary settings (see papers in Lykousis et al. 2007;Mosher et al. 2010a;Yamada et al. 2012;Krastel et al. 2014;Lamarche et al. 2016). This community of scientists recognizes the need for the compilation of a global subaqueous landslide database, to effectively integrate the wider community knowledge and tackle outstanding scientific questions. This is with a view to support the following activities: • Provide the basis for statistical analysis to robustly test hypotheses that are currently either only qualitatively addressed or supported by databases with relatively small sample sizes, such as exploring potential links between landslide frequency and sea level/climate change (Geist & Parsons 2006ten Brink et al. 2006;Clare et al. 2016b). • Identify and quantify the physical controls on landslide frequency-magnitude and triggering between different margin types, and in different settings (e.g. high to low sedimentation regimes, lakes compared to deep-sea, etc.). • Enable knowledge-gap analysis and to inform future strategies for a more complete data collection (e.g. identify potential blind spots, reconcile geographical, temporal and physiographical biases in the available data, and inform future selection of appropriate sampling and survey techniques). • Quantitatively compare landslide parameters across a range of scales (from experimental laboratory models, lacustrine and fjord slope failures, to prodigious continental slope collapses) to determine if any scaling relationships exist. For example, can we make informed inferences or extrapolations about the largest events on Earth from easier-to-access examples in lakes or fjords? Can we assess spatial extent through the examination of a failure deposit width or thickness (e.g.

Moscardelli & Wood 2016)?
Existing subaqueous landslide databases A number of subaqueous landslide databases already exist, but the manner in which parameters are measured, and hence the consistency between studies, varies between the discipline of the data-gatherer (e.g. lacustrine or marine, ancient or recent stratigraphy) and the end-user focus (e.g. tsunami modelling, seafloor hazard assessment, hydrocarbon exploration, benthic habitat mapping). Existing databases encompass: (i) the submarine landslide frequency (which is generally biased towards events in the last 40 kyr; Owen et al. 2007;Urlaub et al. 2013Urlaub et al. , 2014Brothers et al. 2013;Clare et al. 2014;Hunt et al. 2014); (ii) geotechnical properties (Day-Stirrat et al. 2013;Sawyer & DeVore 2015); (iii) damage to seafloor infrastructure (Pope et al. 2016(Pope et al. , 2017; and (iv) morphometrics (i.e. measurements that record the geospatial dimensions of a landslide: e.g. Moscardelli & Wood 2016). The latter is the most commonly recorded information as morphometrics are relevant to a wide range of applications, including seafloor geohazard assessments (runout distance, magnitude, spatial frequency), tsunami modelling (failure volumes and directionality), hydrocarbon exploration (extent of evacuation v. depositional zones) and benthic ecology (nature of scar and distribution of deposits

Aims
An IGCP-640-funded workshop held in January 2017 set out to discuss improved integration between the disciplines for which subaqueous landslides have relevance, and to propose a uniform method for their measurement. A proposed long-term goal is the construction of a global comparative landslide database that will include morphometrics, as well as other parameters. Disciplines represented at the workshop included specialists in lacustrine and deep-water sedimentology, seafloor habitat mapping and ecology, marine geophysics, marine geochemistry, hydrocarbon exploration and production, subsurface fluid flow and storage, offshore and coastal geohazards, and volcanology. In this paper we tackle three overarching questions. First, what is the benefit of a global database of subaqueous landslides? We discuss how such a database can provide valuable and consistent data for scientific hypothesis testing (e.g. global to local scaling relationships), societally-relevant applications (e.g. hazard assessments), to determine systematic biases and identify data gaps that require filling.
Secondly, we ask what are the challenges and potential pitfalls in making morphometric measurements of subaqueous landslides using different data types, in different basins and in different ages of deposits having undergone different diagenetic changes? A global database should incorporate observations from the modern seafloor and lakes using hull-mounted and higher-resolution (e.g. autonomous underwater vehicle (AUV)) bathymetry, 2D and 3D seismic reflection data imaging both the seafloor and subsurface strata, and outcrop observations. But what are the implications of comparing measurements between these different data types? We aim to understand what can be reliably understood and interpreted from comparisons between morphometric studies.
Finally, we ask how do you measure and describe the morphometry of both modern and ancient subaqueous landslides in a consistent manner? No common method currently exists for the subaqueous landslide community. Here we present, and test, a method that can be widely adopted to enable consistent comparisons between workers, and thus assist in the development of a consistent ancient and modern global database. We identify a number of morphometric parameters to describe a subaqueous landslide and assess the repeatability of measurements made by different operators for the same landslide (Table 1).
How can a global database identify and address systematic biases and knowledge gaps?
We recognize that there are often a number of systematic biases in studies of subaqueous landslides. We now discuss why these biases exist and how a global database can be used to identify and address those biases, to ensure that future studies can be focused to fill outstanding data and knowledge gaps.

Scale bias
Many scientific studies have focused on large-scale landslides as they are easier to image in detail than small landslides that are close to the resolution limits of the imaging tools. These larger events are also often considered (e.g. Calves et al. 2015) to pose a greater danger to public safety (e.g. higher tsunamigenic potential) and are therefore the focus of attention. Furthermore, smaller landslides (≪1 km 3 ) may be imaged in some surveys, but are often not the foci of follow-up study as they may be less significant for Single event (mass-transport deposit) or multiple events (mass-transport complex). Multiple events should be linked to a parent ID Depth below seafloor (m) For landslides measured from subsurface data, this is the depth to the top of the landslide deposit. If calculated from seismic data, the two-way travel time (TWTT) should also be referenced. If mapped from seafloor data without seismic or core sample calibration this will not be possible to complete Depth below seafloor (TWTT in ms) For landslides measured from subsurface geophysical data, this is the depth in TWTT to the top of the landslide deposit

Measured landslide morphometrics
Latitude and longitude (WGS) Centre-point of the mapped feature. It is recognized that the entirety of a landslide may not be visible due to data coverage limitations; hence, this is primarily intended to locate the feature on a global database Water depth minimum (m) Minimum water depth for mapped landslide (only possible from multibeam data) Water depth maximum (m) Maximum water depth for mapped landslide (only possible from multibeam data) Total length, L t (m) Total mappable length of slide from the upslope limit of the headscarp to the downslope limit of the connected deposit (excludes outrunner blocks). This is measured along the axial course of the landslide if possible (e.g. from multibeam echosounder (MBES) data), otherwise this is a straight line (e.g. measured from 2D seismic data) and is an 'apparent' length measurement. Detail on the method should be listed as accompanying metadata Deposit length, L d (m) Total mappable length of the slide deposit (excludes outrunner blocks). This is measured along the axial course of the landslide if possible and, hence, is not necessarily a straight line (e.g. from MBES data); otherwise, this is a straight line (e.g. measured from 2D seismic data) and is an 'apparent' length measurement. Detail on the method should be listed as accompanying metadata Evacuated Length, L e (m) Length of the scar from the headscarp to the upslope limit of deposit measured along the axial course of the landslide. Should be equal to L t minus L d Length metadata For example, is this measured from a section and is it an apparent measurement (and thus may be an underestimate), or otherwise how was the distance calculated? Scar perimeter length, L s (m) Length

Preservational bias
We often make measurements based on surfaces preserved at seafloor or the lakebed, from seismic data or in outcrops; however, recent repeated surveys have shown that a dramatic reworking of landslide scars and deposits can occur very soon after deposition in some settings. For instance, the volume of a submarine landslide deposit in the head of Monterey Canyon, California was reduced by 80%, while the scar area increased by 40%, over the course of less than 2 years due to current reworking (Smith et al. 2007

Temporal bias
There is currently a strong bias in published databases and collations of subaqueous landslides to those that are less than c. 40 kyr old (i.e. the limits of radiocarbon dating: Brothers et al. 2013;Urlaub et al. 2014). Current sampling and dating methods limit the age controls we have on more ancient failure deposits. This temporal bias provides challenges when testing hypotheses such as the influence of sea level on failure frequency or linkages between climate and failure, as the spread of landslide occurrence does not span sufficient sea-level stands or climatic intervals (Pope et al. 2015). Future databases should integrate modern seafloor studies with studies of older landslides, which can be dated using other multiproxy methods (e.g. oxygen isotopes, coccolithophore biostratigraphy, magnetostratigraphy and tephrochronology: Hunt et al. 2014;Clare et al. 2015;Coussens et al. 2016) and imaged at depth using seismic data (e.g. Gamboa & Alves 2016).

Geographical and economic bias
Until recent years, compilations of submarine landslide morphometrics largely focused on the NE Atlantic, North American, Iberian and Mediterranean continental margins (Pope et al. 2015), where higher-resolution data were collected due to offshore What are the challenges and potential pitfalls for the morphometric characterization of subaqueous landslides?
We now outline the main issues encountered when attempting to measure the morphometry of subaqueous landslides.

Low data resolution relative to landslide scale
The accuracy of any morphometric landslide measurement is a function of the resolution of the data relative to the scale of the landslide (Fig. 1). In many cases, it may be possible to make reliable measurements of first-order morphometrics, such as total landslide length or scar width, using relatively coarse resolution (often hull-mounted) multibeam data (e.g. in Fig. 2b, a similar landslide outline could be mapped from 30 m binned data compared to that from 0.5 m bin size). However, it is still possible that many small landslides will be missed using such coarse-resolution data and more detailed measurements of evacuation or deposit length are often not feasible. It is unlikely that accurate measurements would be made of the landslides shown in Figure 2a or d using the 30 m bin size data alone. We must recognize, therefore, that landslide catalogues and databases are incomplete (Malamud et al. 2004;Urgeles & Camerlenghi 2013). Measurement of landslides from older legacy data, that are often very low resolution, is particularly prone to this problem. The growing trend for using autonomous underwater vehicles (AUVs: Wynn et al. 2014) and remotely operated vehicles (ROVs: Huvenne et al. 2016Huvenne et al. , 2018 to map the seafloor will enable us to tackle this issue and start populating the missing lower end of the scale. This is comparable to that encountered when mapping other seafloor features, such as bedforms, where new highresolution AUV data have enabled an update of a pre-existing classification system (Wynn & Stow 2002) to fill in some of the blanks (Symons et al. 2016). Length measurements of irregular features, such as scar perimeter, are often highly variable between operators, depending on how complex the feature is deemed to be by each individual and to what level of detail they define it. Limited time availability for measurement, coupled with a large number of landslides, can lead to reduced detail in mapping and, thus, resulting in smaller perimeter lengths compared to a more detailed analysis. Furthermore, the measured length of a complex feature will increase if data resolution is enhanced, due to the improved imaging of a greater morphological complexity. This issue is comparable to the coastline paradox of Mandelbrot (1967), wherein the coastline of Britain apparently lengthens as the resolution of measurement becomes finer.

Large landslide scales relative to the survey area
It is difficult to accurately define landslides whose extents are at the limits of the data resolution (Gamboa & Alves 2016). However, it is also clear through examining the distribution of landslide deposit sizes that there are many events that extend beyond the spatial limits of a survey or the lateral extent of outcropping strata (Moscardelli & Wood 2016). This latter issue is well illustrated by prodigiousscale landslides, such as the Sahara Slide (offshore NW Africa: Georgiopoulou et al. 2010), that are so large it is usually impractical to survey their full areal extent (Fig. 3e) (Li et al. 2017). Similarly, the full extent of landslides is often not imaged in seismic datasets where features are cropped at the limits of the survey area or whose thickness is close to the vertical resolution limits of the data (Alves & Cartwright 2009;Moscardelli & Wood 2016). In such scenarios, it is possible to make measurements of the partial scar or deposits, recognizing that measurements are likely to be underestimated. Where such measurements are recorded in a database, the limitations of the available data coverage relative to the  scale of the landslide should be acknowledged in accompanying metadata and must be considered in comparative analysis.

Differentiating evacuation from depositional zones
Assuming data are resolute enough and the entire landslide is imaged, the measurement of landslide length should be straightforward as it is defined by the major morphological features of a landslide (i.e. the distance from headscarp to toe: Fig. 4). Thus, to a first order, the scale of a landslide should be consistently recorded between operators. Inconsistencies may arise, however, when attempting to demarcate where an evacuation zone ends and the deposit begins, as a higher degree of interpretation is required. Some of this subjectivity can be removed where observations based on multibeam data can be calibrated with seismic data (e.g. Figs 2 and 5). Changes in acoustic character and breaks in the continuity of seismic reflections provide valuable information on defining limits of intact stratigraphy, zones of removed sediment and disruption of transported sediment (e.g. Alves & Cartwright 2009;Alves et al. 2014;Strupler et al. 2017). While this enables better demarcation of evacuation and depositional zones, any measurement of length that is based solely on coarsely-spaced 2D seismic data (or 2D outcrops for that matter) will be an 'apparent' measurement, and is thus likely to be an underestimate. Seismic lines are rarely acquired perfectly along the axis of runout (e.g. Fig. 2). Moscardelli & Wood (2016) recognized this shortcoming in their morphometric analysis of landslides and took a simplistic approach to measure length (straight line distance measured from headscarp to downslope limit of deposit). Thus, any comparison of measurements based on coarsely-spaced 2D seismic with those made from multibeam or 3D seismic data results in an estimate that may be misleading unless the line spacing is close enough. For this reason, it is preferable that measurements are integrated where complementary multibeam and seismic datasets are available.

How and where to measure slope gradient
The measurement of slope gradient is important given the sensitivity of slope stability analysis and volume calculations to slope gradients. This is also crucial for seismic-based studies of buried landslides, as the velocities considered for distinct overburden intervals will affect the measured slope angles. The location and the distance over which measurements of slope gradient are made will greatly influence the result. Thus, it is important that the location and length over which slope gradient is measured are well documented, otherwise comparisons between studies may be meaningless.  Table 1 showing (a) frontally-emergent and (b) frontally-confined landslide cases in cross-section, and (c) a plan view of the landslide.

Competing subaqueous landslide classification schemes
A large number of classification schemes exist for terrestrial and subaqueous landslides (e.g. Varnes 1958;Hampton et al. 1996;Mulder & Cochonat 1996;Locat & Lee 2002;Masson et al. 2006;Moscardelli & Wood 2008;Hungr et al. 2014). There is a high degree of subjectivity in the interpretation of failure mode or the nature of displacement, however. Furthermore, the complex and often transformative rheology of subaqueous mass movements along their course (e.g. Talling et al. 2007;Haughton et al. 2009;Richardson et al. 2011) makes a genetic classification challenging. On a more simple level, however, subaqueous landslides can be differentiated by: (i) the nature of the landslide front (i.e. the degree of frontal confinement); and (ii) the relationship of the landslide to its source area (i.e. attached or detached).
It is important to discriminate between landslides with different degrees of frontal confinement, as these are associated with different formative mechanisms, downslope propagation, internal kinematics and resultant deposits (Frey-Martínez et al. 2006). Frontal confinement is classified by Frey-Martínez et al. (2006) as either: (a) 'frontally-confined' landslides, where the landslide front abuts undisturbed sediments; or (b) 'frontally-emergent' landslides that ramp up from their original stratigraphic position to move across the lake or seafloor unconfined (Moernaut & De Batist 2011). Such a simple binary classification does not take into account natural complexity and only applies to translational failures which start on an intact slope profile; hence, we suggest that the following terms are also used: (c) 'frontally-confined with overrunning flow', where a debris flow or incipient failure may runout over the confined toe of a landslide; (d) 'frontally-unconfined' landslides, where there is no downslope buttressing, such as where the toe of a slope has been excavated by erosion or in the case of rotational failures (Lacoste et al. 2012); and (e) 'not identified', where the data do not enable the classification to be made. Moscardelli & Wood (2008) proposed a binary classification for landslide attachment that includes: (a) landslide deposits which are attached to their source area, which are typically regionally extensive features that occupy hundreds to thousands of square kilometres in area; and (b) landslide deposits that are detached from their scar, which are typically much smaller. Whether or not landslides are attached to their scar reveals information about the nature of the failure, if landslides were potentially tsunamigenic and has been suggested to provide an indication of a potential triggering mechanism (Moscardelli & Wood 2008). The use of both approaches ensures that at least one classification can be made even if only the source, or the front (terminal end), of a landslide is imaged and avoids the high degree of subjectivity in other more complicated genetic classification schemes.

Challenges in calculating landslide volumes
Numerous methods have been applied to the calculation of landslide volume from multibeam bathymetry data. The first is based on an estimation of the Fig. 5. Example of frontally-confined landslides in Lake Zurich (modified from Strupler et al. 2017). Volumes based on thickness measurements from multibeam data are a factor of 3 less than those calculated from seismic data. missing volume from a scar: calculated from the difference between the scar topography and an interpolated surface that connects the upper edges of the scar. This approach thus aims to reconstruct the prefailure topography (ten Brink et al. 2006;Chaytor et al. 2009;Katz et al. 2015;Chaytor et al. 2016a, b). The second method is based on the measured scar dimensions (McAdoo et al. 2000), wherein the landslide volume is modelled as a wedge geometry (volume = 1/2 × area × height). The lower plane of the wedge is derived from slope angles of the runout and/or scar, and the upper plane is based on the gradient of the unfailed slope immediately adjacent to the seafloor (assumed to be representative of the pre-failure slope). The third method is based on the measurements of the landslide deposit itself. This approach is often used when the scar is not preserved or surveyed (e.g. Masson et al. 2006;Alves & Cartwright 2009). In such a scenario, volume is determined as a function of landslide thickness and area (in the case of the lower measured value, this was estimated as volume = area × 2/3 maximum deposit thickness).
Ideally, additional data should supplement the calculation of landslide volume to calibrate the accuracy of measurements based on multibeam data alone. In Figure 5, we illustrate the value of complementary seismic data to calculate volumes of a frontally-confined lacustrine landslide in Lake Zurich (Strupler et al. 2017). First, we calculated volumes based on the multibeam bathymetry. A missing volume of 800 000 m 3 was derived from the scar height (5 m) and its areal extent (using the method of ten Brink et al. 2006). This value is comparable to the volume calculated from the deposit area and its height above the adjacent seafloor (3.5 m) mapped from bathymetry, which was calculated as 740 000 m 3 . High-resolution seismic profiles indicate that the thickness of the landslide (19 ms = 14 m) is actually much greater than the measured heights from multibeam bathymetry (3.5-5 m). The calculated volume was revised upwards by a factor of 3 to 2 200 000 m 3 . This is a fundamental issue, particularly when dealing with landslides that are buttressed at their downslope limit (i.e. 'frontally confined'), as the sediment does not run over the lakebed or seafloor: hence, its bathymetric expression is limited compared to the total thickness of sediments that are displaced. This underlines the importance of integrating seismic data (Alves & Cartwright 2009). 3D seismic data can provide more accurate landslide volume calculations if the deposit is fully covered by the survey and adequate time-depth conversions are made. Thus, landslide volume should be calculated based on the integration of multibeam and seismic data, where available. However, if only multibeam data are available, then the preferred volume estimates should be calculated based on scar morphometrics, following the approach of ten Brink et al. (2006).

Modification of landslide morphology under burial
Modern multibeam bathymetry and high-frequency sub-bottom profiling data enable high-resolution mapping of modern landslides (i.e. those that can be imaged at seafloor); however, additional challenges are faced when measuring older landslides imaged in lower-frequency seismic data, besides just resolution issues. Under burial, lithification and compaction processes can change the original morphology of landslide deposits. Mapping of landslides from seismic data is typically based on changes in the morphology, as well as the seismic character within the landslide that is a function of both lithology and internal deformation (Ogiesoba & Hammes 2012;Alves et al. 2014). Thus, there must be a recognition that any comparison of recent landslide deposits with those that may have undergone significant post-depositional modification is not necessarily like-for-like. Despite this, there is considerable value in comparing recent landslides with the range of events that have happened over a longer timescale in Earth history. Such a comparison may lead to the development of correction factors to enable more effective integration between modern and ancient studies.

Further complications caused by natural complexity
Many subaqueous landslides are highly morphologically and structurally complex. Such complexity increases the number of interpretative decisions that must be made by the operator when measuring morphometry. Many landslides do not fail as one single event; instead, occurring in stages over both short and long timescales (e.g. Cassidy et al. 2014: Mastbergen et al. 2016. In such cases, the scar may be highly irregular, stepped or feature smaller incipient failures along the headscarp, complicating the measurement of headscarp height and scar dimensions (e.g. Georgiopoulou et al. 2013;Katz et al. 2015) (Fig. 3e). Areas that are highly prone to landslides may feature aggregated or cross-cutting evacuation scars and deposits from multiple different failure events. For instance, the Traenadjupet Slide overlies and cuts into the older Nyk Slide, offshore Norway (Lindberg et al. 2004). Figure 3d shows the case of the Tuaheni landslide complex, where multiple landslides intersect each other, and may have caused reworking of both deposits and parts of the scar (Mountjoy et al. 2014).
The large-scale Laurentian Fan landslide presented by Normandeau et al. (2018) is an example of a complex failure that also shows localized variation in its frontal confinement; in places, the front of the failure abuts the stratigraphy, while in others it ramps up and becomes emergent. It is thus difficult to classify into just one category. Landslide fronts can become frontally emergent at several locations, such as the 900 km 3 Traenadjupet Slide, offshore Norway (Laberg & Vorren 2000). In that case, multiple lobes formed at the different emergence points, thus providing several options for measuring total landslide length. The interaction of landslides with the underlying stratigraphy, particularly where erosion, ploughing or stepped frontal ramps occur, can further complicate the measurement of thickness and, in turn, the associated calculation of volume from deposits (e.g. Richardson et al. 2011;Puzrin 2016).
How can the morphometry of subaqueous landslides be measured consistently?
A standardized approach does not yet exist for consistent morphometric characterization of subaqueous landslides. Here, we present a method for measuring key subaqueous landslide morphometrics that can be applied to seafloor, subsurface and outcrop data in their full range of settings. The morphometric parameters chosen are deemed to be relevant to a broad suite of disciplines. We provide instructions on how to measure each parameter (Table 1; Fig. 4). Given variations in data limitations and extent of study area, it may not be possible to measure all of these parameters in all cases; however, our intention is to provide a comprehensive list to enhance the utility of a global database and to ensure that measurements are made consistent.

Testing a standardized approach
In order to test our approach for measuring landslide morphometrics, we analysed data from the Valdes Slide, offshore Chile (Fig. 3a) (Völker et al. 2012). A relatively simple case study was chosen for this applications test to first understand the limitations of the method in a close-to-ideal scenario. The Valdes Slide is considered to be a relatively simple landslide as it does not feature multiple lobes, the scar is well imaged and it is of a scale such that most morphometrics can be measured clearly. Each operator's analysis was performed in isolation to try to reduce interpretational bias. Software packages used for the analysis varied between operators, and included ESRI ArcGIS, Global Mapper, Teledyne CARIS, Fledermaus and Open Source QGIS. Operators based their analysis of the bathymetry on a number of different attribute tools, including contour, hillshaded illumination, slope angle and aspect tools (e.g. Fig. 1), as well as 3D visualization. Results from each of the individual operators were then collated and compared to understand the variance in outputs (Table 2; Fig. 6).
Consistency in measurement of first-order parameters. Parameters that locate the Valdes Slide (latitude, longitude and water depth) showed very good  Table 2). Measurements of total length measured along the landslide axis (L t ) and the height drop (H z : defined here as the difference between the minimum and maximum water depth) were comparable between operators (c. 12% RMMV). The headscarp height (H s ) and evacuated height (H e ) also yielded comparable values (8-12% RMMV: Table 2). Landslide length (runout), height drop and headscarp height are important first-order parameters in quantifying the scale of a landslide. It is therefore reassuring that the measured values are similar between operators and provide a degree of confidence for comparing other well-defined landslides using these first-order metrics. Thus, a global database should provide useful and comparable measurements of landslide location and scale.
Variance arising from increasing operator decision making. As anticipated, evacuated length (L e ) and depositional length (L d ) yielded more disparate results (44 and 36% RMMV, respectively: Table 2). This is attributed to the fact that the operator needs to make an interpretative judgement based on the analysis of bathymetry data as to where evacuation ends and deposition starts. This subjectivity could be reduced by integrating supplementary datasets such as sub-bottom profiles; however, in situations where further data are not available, it is important that the potential error is made clear in any metadata accompanying these measurements. Measurements of scar width (W s ) and deposit width (W d ) provided RMMV of 29 and 45%, respectively (Table 2). An even wider spread of values (57% RMMV) was determined for scar perimeter length (L s ). The variance in these parameters is attributed to the fact that these measurements are based on a higher degree of operator decision mapping, which introduces a large degree of subjectivity to the analysis. We suggest a spline should be fitted to the measured perimeter length to ensure consistency in measurement to account for different levels of data resolution. The least consistently measured parameters were slope angles (S, S s , S t : 44-62% RMMV). This relates to the distance over which slopes were measured and variations in the specific locations where those measurements were taken.
Only two operators attempted to calculate volume for the Valdes Slide, and provided highly variable values of 0.3 and 1.3 km 3 . The highest measured value (1.3 km 3 ) was based on an estimate of the missing volume from the scar: calculated from the difference between the scar topography and an interpolated surface that connects the upper edges of the scar (i.e. aiming to reconstruct the pre-failure topography, following the approach of ten Brink et al. 2006). The lower measured value (0.3 km 3 ) was based on the landslide deposit itself.

Importance of metadata to record uncertainty
An Open Source version of the morphometric parameter inventory is hosted through a Google Fusion database. This web-based access enables the wider community to contribute morphometric data to a growing global database. In light of the challenges associated with data resolution and operator decision making, a free text metadata field accompanies the entry for each of the measured metrics to record comments on the uncertainties, errors and operator decision making involved in the data collection, analysis and measurement.

Conclusions
No common method exists for describing the morphometry of subaqueous landslides. This hinders Fig. 6. Mean values (symbols) and total range (whiskers) from a morphometric analysis of the Valdes Slide (Fig. 3a) performed by the authors based on data in Table 2. the effective integration of results from different research groups, disciplines and based on disparate data types. In this paper we presented and tested an approach that can be adopted to enable consistent global comparisons, and so form the basis for the compilation of a global database to integrate studies ranging from modern to ancient timescales and lacustrine to marine settings. We identified a number of challenges.
The first challenge is that a number of biases exist in data collection and analysis, spanning spatial, preservational, temporal, geographical and economic issues. These and other biases can be better recognized and addressed by a global database of subaqueous landslides. Future data collection should aim to address these issues, such as the limited data availability in margins surrounding developing countries. In the absence of margin-specific data, a consistent global database of subaqueous landslides can have a powerful role, however, by enabling the inference of information (e.g. landslide likelihood) from analogous, better-studied margins.
Second, we highlighted how the accuracy and number of parameters that can be mapped is a function of landslide scale relative to the data resolution and extents. Small landslides are difficult to map accurately (if at all) from low-resolution data, whereas large landslides may not be fully imaged by high-resolution datasets with limited extents. A global database should allow for the testing of scaling relationships on a local and global scale to provide guidance in both situations.
Finally, we presented and tested a method to enable the consistent measurement of subaqueous landslides. We found that as the degree of decision making by the operator increased, so did the uncertainty in the measured parameter. Basic parameters that describe the overall landslide scale (e.g. width, length) were most consistently measured. Parameters that required increased operator judgement (e.g. pre-failed slope, scar perimeter length) resulted in a wider range of results. We introduced a standardized method of measuring morphometry, and emphasized the importance of accompanying metadata to explain any decisions made in the measurement process to inform future comparative analysis. We feel this method of documenting subaqueous landslides will provide substantial benefits to both the research and applied community so that a consistent global landslide database can be developed.