Streamlined subglacial bedform sensitivity to bed characteristics across the deglaciated Northern Hemisphere

Streamlined subglacial bedforms observed in deglaciated landscapes provide the opportunity to assess the sensitivity of glacier dynamics to bed characteristics across broader spatiotemporal scales than is possible for contemporary glacial systems. While many studies of streamlined subglacial bedforms rely on manual mapping and qualitative (i.e., visual) assessment, we semi‐automatically identify 11,628 sedimentary and bedrock bedforms, created during and following the Last Glacial Maximum across nine geologically and topographically diverse deglaciated sites in the Northern Hemisphere. Using this large dataset of landforms and associated morphometrics, we empirically test the importance of subglacial terrain on bedform morphology and ice‐flow behavior. A minimum bedform length–width ratio threshold provides a constraint on minimum morphometrics needed for streamlined bedforms to develop. Similarities in bedform metric distribution regardless of bed properties indicate that all bed types may support similar distributions of warm‐based ice flow conditions. Ice flow within valleys with easily erodible beds host the most elongate bedforms yet the widest range in bedform elongation and bedform surface relief. The presence of these highly elongate bedforms suggest high ice‐flow velocities occur within valley settings despite spatially heterogeneous landform‐generating processes. In contrast, lithified sedimentary beds within regions not constrained by topography on the scale of 1–102 km contain bedforms with high density and packing, low change in surface relief and low elongation, indicating spatially uniform and organized interactions at the ice–bed interface and consistency in ice‐flow velocity. Regardless of genesis, we find a sensitivity of bedform elongation (i.e., used to interpret ice‐flow speed or persistence) to topographic conditions on the scale of 1–102 km, while bedform density is sensitive to bed lithology. The findings presented in this study provide analogues for processes of subglacial erosion and deposition, ice–bed interactions and warm‐based ice flow within contemporary glacial systems.


| INTRODUCTION
Understanding the conditions that control ice-sheet flow is particularly important for ice streams, conduits of fast-flowing ice at rates of 10 2 -10 3 m a À1 , due to their ability to efficiently drain and destabilize glacial catchments and greatly impact glacial contributions to sea level (Bamber & Aspinall, 2013;Rignot et al., 2019;Serrousi et al., 2017).
Patterns and rates of ice flow are commonly linked to known or perceived properties of the bed including topography and lithology (Clarke et al., 1977;Cuffey & Paterson, 2010;Whillians & van der Veen, 1997). Bed properties can have opposing effects and varying degrees of influence on ice-stream behavior (De Rydt et al., 2013;Falcini et al., 2018;Greenwood et al., 2021).
Valleys and troughs underlying both marine and terrestrial-based glacial systems have the potential to increase ice-flow velocities due to steering and thickening of ice, leading to increased pressure melting and overall meltwater abundance that enhance basal sliding and sediment deformation (Eyles et al., 2018;Hindmarsh, 2001). Similarly, ice flow is accelerated through strain heating of basal ice (McIntyre, 1985;Pohjola & Hedfors, 2003;Winsborrow et al., 2010) in areas of positive topographic relief (i.e., pinning points, ridges and banks) and regions of high bed roughness (i.e., spatial variation in surface elevation and slope; Falcini et al., 2018;Rippin et al., 2011;Siegert et al., 2005). In other circumstances, obstacles in the bed and deep valley regions surrounded by areas of high topographic relief enhance basal and lateral drag, leading to slower ice flow and potential grounding-line stabilization in marine-terminating systems (Falcini et al., 2018;Favier et al., 2016;Whillans & van der Veen, 1997).
Bed lithology also plays a fundamental role in ice-bed coupling, efficiency of meltwater transmission, and sedimentary processes such as deformation, erosion and deposition (Weertman, 1957). Permeable unlithified sedimentary beds allow for water infiltration and enhanced ice motion due to sediment deformation (Alley et al., 1986;Cuffey & Paterson, 2010;Tulaczyk et al., 2000), whereas more impermeable, "hard" beds facilitate sliding through liquid water being held between the ice and bed (Evans et al., 2006;Nienow et al., 2017). Rates of erosion and deposition in the subglacial environment are fundamentally impacted by bed lithology due to its control on meltwater transmission and, presumably, relative erodibility differences between different lithologies (Fowler, 2010;Greenwood & Clark, 2010;Ng, 1998).
Streamlined bedforms in deglaciated landscapes are used to interpret ice-flow behavior (e.g., Clark et al., 2009Clark et al., , 2018Dowling et al., 2015;Ely et al., 2016;Greenwood & Clark, 2010;Hättestrand et al., 2004;Spagnolo et al., 2012Spagnolo et al., , 2014Stokes et al., 2013) and used to understand ice-bed interactions applicable to contemporary glacial systems (e.g., Eyles et al., 2018;Greenwood et al., 2021). Qualitative (i.e., visually descriptive) and quantitative (i.e., morphometric and statistical) analyses of streamlined bedforms are arduous tasks as these bedforms have low, even sub-meter vertical relief and typically occur in "swarms" of tens to thousands of bedforms Ely et al., 2016;Hughes et al., 2010). While automated mapping methods more quickly detect streamlined bedforms than traditional manual mapping techniques, only a subset of these methods have been systematically applied across multiple sites (e.g., Cazenave et al., 2008;Clark et al., 2009;Saha et al., 2011;Spagnolo et al., 2017;Wang et al., 2017). Our study uses topographic positioning index (TPI; Tagil & Jenness, 2008;Weiss, 2001) to calculate "neighborhood" elevation and slope variations to semi-automatically identify subglacial streamlined bedforms from nine deglaciated landscapes in the Northern Hemisphere (Figure 1). This large, geographically diverse dataset of streamlined bedforms contains both bedrock and sedimentary bedforms, encompassing a range of erosional and depositional processes associated with ice flow of four former ice sheets. We aim to identify the sensitivity of warm-based ice-flow conditions and subglacial processes to variable bed conditions. Substrate influence on warm-based ice flow is inferred from bedform relationships with bed topography and lithology, independent of bedform genesis concepts. We classified topographic setting in the broadest sense as either "constrained" or "unconstrained" on spatial scales of 10 1 -10 2 km (Payne & Dongelmans, 1997). "Constrained" topography is defined as low elevation surrounded by more elevated regions on the horizontal scale of 1-10 2 km and vertical changes of at least 400 m from valley bottom to top, while "unconstrained" topography is defined as open, relatively uniform topography on the horizontal scale of 1-10 2 km with no more than 400 m of vertical elevation change across the site (Supporting Information Figure S1). Therefore, valleys, troughs and basins are considered constrained topography, while unconstrained topography lacks these features in the regional landscape.

| METHODOLOGY
Surficial bed lithology was generally classified as: "lithified sedimentary" including both clastic and carbonate bedrock; "crystalline"; "volcanic rock"; or, in one case, "mixed" bed including both crystalline and unlithified sedimentary beds. These distinctions were made due to the general variation in primary permeability, erodibility and likelihood of these lithologies to develop into a deformable subglacial bed.
Unlithified sedimentary beds are the most permeable, erodible and likely to support a deformable bed, whereas the crystalline bedrock sites are the least permeable, erodible and least likely to contribute to the formation of a deformed till layer compared to a sedimentary bed.
Only underlying bed composition was considered in determining site lithology; therefore, thin layers of till or soils overlying crystalline bed sites were not considered in bed lithology classification. While there are differences in properties between, for example, clastic and carbonate sedimentary rocks, the differences are subtle enough for the two categories to be classified together for the purpose of this regional and bedform-assemblage scaled work.
We mapped streamlined bedforms from the nine sites with a combination of manual identification and a novel semi-automated application of a TPI mapping tool, originally developed for the purpose of characterizing watersheds across regional landscapes (Weiss, 2001 Figure S2). Thresholding of TPI-mapped bedform metrics such as feature length, width, orientation and area attributes coupled with a manual assessment, conducted by visually removing incorrectly identified features and adding features missed by TPI, resulted in a more accurate dataset whose metrics were not influenced by morphometric threshold sorting (McKenzie et al., 2022; Figure S2).
For each mapped bedform, its long-axis length and orientation, width orthogonal to length, and the range in elevation along the landform's long axis (termed here "bedform surface relief") were calculated automatically in ArcGIS Pro using the "Minimum Bounding Geometry" and "Add Z Information" tools. Automatic calculation of streamlined bedform long-axis cardinal orientation is quantified in degrees, measured by the rotation of the bedform long axis from due north, and is used to infer direction of ice flow (Clark, 1997;Kleman et al., 2006;Kleman & Borgström, 1996). Parallel conformity, or the standard deviation of an entire streamlined bedform dataset's orientation values at a given site, was additionally calculated in Excel.
Bedform elongation ratio, calculated by dividing the bedform length by its width, and parallel conformity (i.e., the standard deviation of bedform orientation) were calculated in Excel. Defining bedforms solely by their shape rather than composition is a strength to this study, as it allows for the assessment of topographic and lithologic controls on ice streaming and sedimentologic and geomorphologic processes regardless of landform-generating processes that have been inferred for individuals or groups of landforms.

Statistical analyses including a linear Pearson correlation and
Shapiro-Wilk test were conducted in "R" to compare bedform metric distributions. Qualitative metrics, namely site topography and lithology, were assigned a fixed numerical value that was used in the Pearson correlation, where topographically unconstrained regions were assigned a value of 0 and topographically constrained were assigned a value of 1. Sedimentary bedrock was assigned a value of 0, crystalline bedrock a 1, volcanic bedrock a 2 and mixed lithology a 3.

| RESULTS
In the following subsections, we demonstrate our use of TPI to identify streamlined subglacial bedforms, the trends and correlations of bedform morphology across all sites, and the relationship between spatial orientation and distribution of bedforms. Lastly, we describe the relationship between bedform characteristics and bed topography and lithology.

| Streamlined subglacial bedform identification
Across the nine sites, TPI identified 7,635 bedforms, while 3,993 bedforms were manually mapped (i.e., added or adjusted from TPI mapping), resulting in a total dataset of 11,628 sedimentary and bedrock bedforms (Figure 2). Of the nine study sites, the Puget Lowland confidence in the ability of these dual methods to capture a "full" dataset of streamlined subglacial bedforms (Supporting Information Figure S3).

| Bedform morphology
The streamlined bedforms range in length from 94 to 15,388 m (mean 1,052 m; median 754 m) and in width from 19 to 2,323 m (mean 219 m; median 157 m). As determined by a linear Pearson correlation, bedform length and width have a positive correlation coefficient ( Figure 3). This statistical analysis was fitting for this data as it met the assumption that all groups were statistically significant in their difference from one another (p < 0.05). Puget Lowland (Site A) bedforms span the greatest range in width and length of all sites, while Chautauqua (Site C) has the smallest ranges of length and width. Interestingly, low bedform lengths and widths are common for all sites, while bedforms with lengths greater than $2000 m and widths greater than $600 m are less common ( Figure 4A).
Consistency in peak elongation ratios for all sites is also observed, with a median elongation ratio of 5:1 ( Figure 5; Table 2 Table 2).

| Bedform morphometric correlations with bed lithology and topography
The Puget Lowland (Site A), a topographically constrained mixed lithology site, has the greatest number of bedforms per area (i.e., density) and the highest bedform area per area (i.e., packing;  Table 2).
The topographically constrained volcanic bed of Bárðardalur (Site G) contains low-density and packing bedform swarms and has the most elongate bedforms in the dataset (Figure 8; Table 2). After Bárðardalur (Site G), the sites with the most elongate bedforms are also all topographically constrained sites (Table 2)  Across site bed settings, a "V" denotes a topographically constrained site, while "O" denotes a topographically unconstrained site. Parallel conformity is a measurement of the standard deviation of mean orientation.
with bedform width indicating that large widths are likely on sedimentary bedrock sites while bedforms with smaller widths are more common in mixed lithology sites. Conversely, bedform surface relief is highly correlated with surface lithology, where sedimentary bedrock sites have lower bedform surface relief and mixed lithology sites have greater bedform surface relief (Figure 3). Regardless of these differences, all sites have similar morphometric distributions and show considerable overlap with common means and medians, rather than distinct populations, and only differ in the degree of skewness.

| DISCUSSION
A discussion of the performance of TPI in mapping subglacial streamlined bedforms is presented in the first subsection, followed by discussions of streamlined bedform morphology and the relationship between bedform morphology and bed settings.

| Success of semi-automatic mapping streamlined bedforms in deglaciated landscapes
Previous morphometric studies of streamlined subglacial bedforms have utilized Fourier spectra data (e.g., Spagnolo et al., 2017), manual identification (e.g., Principato et al., 2016), object-oriented automatic identification (e.g., Saha et al., 2011), contour-tree mapping (Wang et al., 2017) and other methods systematically utilized to identify bedforms across singular and multiple geographic locations (e.g., Clark et al., 2009Clark et al., , 2018Ely et al., 2016;Greenwood & Clark, 2010;Spagnolo et al., 2014;Stokes et al., 2013). While TPI was originally developed to classify landscapes and delineate watersheds (Tagil & Jenness, 2008;Weiss, 2001), its ability to characterize negative and positive relief features through slope variations is conceptually applicable to many landscapes. In the context of glacial landscapes, the distinct elongate morphologies and occurrence of numerous bedforms in close proximity make streamlined subglacial bedforms well suited for identification with TPI. In general, highly elongate bedforms with low bedform surface relief are more difficult to map due to small and narrow slope differentiations. Due to the difficulty of the TPI method in identifying these sizes of bedforms, many of the manually mapped bedforms were visually low-relief and highly elongate ( Figure S3).
Additionally, two sites with the greatest number of manually mapped bedforms occurred in northern Norway (Site H) and northern Sweden (Site I; Table 2), where preservation of non-glacial or minimally glacially modified landforms is significant across the entire region (Ebert et al., 2012;Hall et al., 2013;Kleman & Stroeven, 1997). These mapping challenges may be a result of residual pre-LGM landform assemblages and bedrock morphologies from pre-Quaternary formation and weathering processes (e.g., Ebert et al., 2012;Hall et al., 2013;Kleman & Stroeven, 1997). Landscape features that were manually removed include non-glacial positive relief features such as modern riverbanks and isolated bedrock highs, identified by their location, size, orientation or lack of any elongation. While there were some shortcomings in the bedform TPI identification method developed for this work, the hundreds of hours saved by semi-automatically identifying  Table 2). The hard beds in Sweden and Norway, less easily eroded, preserve a landscape legacy of older and persistent processes and therefore have many large, high relief features such as inselbergs (e.g., Ebert et al., 2012;Hall et al., 2013). High-relief features dominate the landscape, even in areas impacted by glacial processes. Therefore, glacial features left on the surface are often very subtle and low in relief, creating difficulties for TPI to identify these small relief bedforms amidst the more pronounced, high-relief features of these landscapes, making manual identification more suitable for these types of bedforms. TPI, therefore, does not perform as well on crystalline bedrock sites as it does on sedimentary bedrock sites where glacial erosion has a larger impact on surface relief. It is also important to note that the crystalline bedrock sites (Sites F, H, I) were deep in the ice sheet interior, where cold-based ice is thought to have been prevalent for long periods over multiple glaciations. Therefore, not only was the crystalline bedrock difficult to erode, but the glaciological environment did not support streamlined bedform processes (Margold et al., 2015;Stokes & Clark, 2001). For a similar reason, the large proportion of manually mapped bedforms in Bárðardalur (Site G) could, in part, be due to a variety of bedform composition and relief ranging from low-relief features comprised entirely of diamicton to high-relief features comprised of diamicton overlaying bedrock cores formed during rifting. More pronounced, high-relief features are more easily identified by TPI mapping, while low relief features are not as easily delineated by the tool.

| Bedform properties across the deglaciated Northern Hemisphere
While sub-meter relief bedforms like bedrock striations are not resolved in the dataset presented here, meter to kilometer-scale bedforms like drumlins and flutes are well resolved. Bedforms across the datasets have significant overlap and positive correlation between width and length (Figures 3 and 4A,B), indicating a shared genetic mechanism between bedforms regardless of their composition or whether they formed through erosional or depositional processes. Additionally, both length and width data are positively skewed, with more small bedforms than large ( Figure 4A,B). This process of bedform elongation is commonly referenced in the field and has led to the use of elongation as a proxy for ice streaming velocity and length of time that a particular flow regime has been operating (e.g., Benediktsson et al., 2016;Livingstone et al., 2016;Stokes et al., 2013;Stokes & Clark, 2002).
Although our nine study sites were glaciated by different ice sheets on different continents, resulting subglacial bedforms were found to be morphologically similar (e.g., Stokes & Clark, 2002;Clark et al., 2009;Saha et al., 2011;Spagnolo et al., 2014;Principato et al., 2016). The similarities in quantitative bedform morphometrics across the study sites suggest similar genetic relationships between all streamlined subglacial bedforms. In this study, we find a minimum length to width ratio (i.e., elongation; e.g., Clark et al., 2009;Ely et al., 2016) of 1.12:1, indicating that barely elongate bedforms are (1) resolved in the dataset and (2) occur for the full size range of bedforms ( Figure 4A,B) potentially due to the highly dynamic and rapidly evolving bedform development beneath warm-based ice (Stokes et al., 2013). In other terms, we observe that a large bedform can be equally as stunted as a small one, suggesting that growth of bedforms does not necessarily add length or strip width preferentially and there is not one single style of bedform development. The unimodal distribution of elongation ratios with positive skewness seen in this work, indicating a continuum of bedform types, has also been found in other morphological bedform assessments (Figure 5; e.g., Clark et al., 2009;Saha et al., 2011;Stokes et al., 2013;Spagnolo et al., 2014;Ely et al., 2016;Principato et al., 2016).
We find that less elongate bedforms are interspersed with more elongate features and not present at the margins of bedform swarms ( Figure 9). This result is not unique and has been hypothesized to be a result of the rapidly changing subglacial dynamics responsible for bedform formation and development (Rattas & Piotrowski, 2003;Stokes et al., 2013). for ice streaming where fast-flowing ice is bounded by slower moving ice (Margold et al., 2018). The spatial organization of elongation at M'Clintock Channel (Site D) contains classic assemblage characteristics associated with fast-flow corridors (e.g., Spagnolo et al., 2014;Stokes et al., 2013;Stokes & Clark, 2001, 2002. While particularly notable at topographically constrained and easily eroded bed substrates, all sites showcase a trend of more elongate bedforms corresponding to lower bedform surface relief ( Figure 4C (Greenwood & Clark, 2010) and topography (Falcini et al., 2018;Favier et al., 2016;Whillans & van der Veen, 1997) or could be indirectly related through substrate control on glaciologic processes. However, this distinction was not explicitly assessed for individual bedforms within this study.

| Bedform morphology in relation to bed setting
Spatially stable and persistent ice streaming conceptually contributes to spatial homogeneity in erosion and deposition processes, leading to the formation of consistently orientated and shaped bedforms. Deviations to bedform orientation occur from both temporal and spatial variations, where bedforms can be preserved from multiple glaciations, such as in northern Sweden (Site I), or across constrained topography, such as in the Puget Lowland (Site A), Bárðardalur (Site G) and northern Norway (Site H). Regions with highly elongate bedforms correspond to qualitatively greater flow orientation organization (Figures 6 and 8; Table 2), which is potentially attributed to multi-generational or persistent bedform formation and development. Multiple glaciations in the same orientation produce more uniformly oriented features (i.e., higher parallel conformity). In the case of spatially influenced orientation, topographic constrains on ice-flow develops bedform clusters with orientation that reflects the orientation of the valley in which it is constrained, such as in Bárðardalur (Site G).
Findings from previous work on bedform presence and morphologies identify glaciological drivers as fundamental controls on bedform morphology via subglacial processes, while others identify direct substrate control on resulting bedform morphologies due to inherent characteristics in the bed (e.g., Greenwood & Clark, 2010;King et al., 2009;Rattas & Piotrowski, 2003;Stokes et al., 2013).
Qualitatively soft, more permeable and easily eroded beds allow for greater production and transport of sediment to the ice margin (Clark, 1993;Rattas & Piotrowski, 2003;King et al., 2009;Livingstone et al., 2016). The abundance of deformable till forming from the permeable environments of both lithified and unlithified sedimentary beds explains the presence of high-density bedforms in the regions with unlithified sediment (King et al., 2009;Stokes et al., 2013). Conversely, "hard," crystalline beds are generally more resistant to erosion (Eyles & Doughty, 2016;Krabbendam et al., 2016) and more likely to have "fixed" bedforms resistant to sediment production and transport (Zoet et al., 2021). The greatest number of bedforms per area, present on a mixed (unlithified sedimentary bed system with crystalline bedrock) bed in the Puget Lowland (Site A), likely occur due to high availability of unlithified sediments and meltwater presence from strain heating. Strain heating occurs as ice flows over bedrock highs, collectively allowing for greater bed erosion across the permeable unlithified sediment, material deposition and subsequent ice streaming (McIntyre, 1985;Pohjola & Hedfors, 2003;Winsborrow et al., 2010). In addition to the mixed bed site, lithified sedimentary beds were also densely populated with streamlined bedform features (Table 2). Crystalline and volcanic beds in both constrained and unconstrained topographic settings have the lowest bedform densities, suggesting that bed lithology, rather than topography, is a more dominant control on streamlined bedform density. However, it remains unclear and difficult to assess whether bed substrate controls either sedimentary processes regardless of flow regime or whether substrate controls streaming conditions. In either case, a distinct bedform signature across a particular bed substrate is produced.
The unimodal and positively skewed distribution of bedform elongation indicates that similar distributions of elongation occur at a multitude of sites regardless of bed topography and lithology or climatological and glaciological factors (Table 1; Figure 4A,B; e.g., Clark et al., 2009;Ely et al., 2016;Principato et al., 2016;Saha et al., 2011;Spagnolo et al., 2014;Stokes et al., 2013). The minimum elongation threshold and similarity in relative elongation distribution across sites highlight a similarity of ice-bed interactions across "soft" and "hard" beds in both topographically confined and unconfined settings, suggesting a self-organization of ice-bed processes regardless of site characteristics. The concept of streamlined bedforms developing as a self-organizing phenomenon is not novel in the field of glacial geomorphology and has been suggested to occur independently from local bed lithologic and topographic conditions (Clark, 2010;Spagnolo et al., 2017;Clark et al., 2018). This independent organization is well represented by the unit-less elongation ratio and allows for comparison of warm-based ice-flow velocities or persistence between individual sites. Differences in elongation distributions across sites of variable topography or bed lithology would suggest an influence of substrate on bedform formation and development or substrate influence of glaciologic processes at the bed. However, from the similarities in bedform elongation distribution we see here and in previous work, we suggest that regions of warm-based ice flow, leading to the development of streamlined subglacial bedforms, exhibit potential for equivalent relative ice-flow velocity distribution or persistence of iceflow pathways regardless of bed character.
While all sites have similar elongation distribution trends, topographically constrained sites produce bedforms with the highest mean and median elongation ratios with the most elongate bedforms of the overall dataset ( Figure 6B; Tables 1 and 2). Topographic constraint on ice flow results in topographic funneling and increased ice speed (Hindmarsh, 2001;Wellner et al., 2001;Hall & Glasser, 2003;Ottesen et al., 2008;Roberts et al., 2010;Eyles et al., 2018). While bedform elongation is enhanced in regions that are topographically constrained, bedform elongation is not contingent on bedrock substrate ( Figure 6B; Tables 1 and 2), which we interpret to reflect a higher sensitivity of warm-based ice-flow velocity and persistence to bed topography than bed substrate (Greenwood et al., 2021;Halberstadt et al., 2016;Ignéczi et al., 2018;Serrousi et al., 2017;Stokes & Clark, 2003;Winsborrow et al., 2010). Additionally supporting this argument, we find the topographically unconstrained and lithified sedimentary bed sites in Chautauqua (Site C) and northwestern Pennsylvania (Site B) have the least elongate bedforms, despite the potential for deformable substrates on these more easily erodible and mobile beds, indicating that distinct bedform morphologies are developed through topography. More specifically, it is the effect of topography on glacial conditions that allows for distinct bedform elongation to develop.
Topographically constrained regions also produce large variations in individual bedform surface relief (Figure 6), indicating less consistency in erosion and deposition distribution across the surface. This effect is particularly apparent at topographically constrained sites with easily eroded beds such as those in the Puget Lowland (Site A) and Bárðardalur (Site G; Figure 6). Large values of bedform surface relief in the Puget Lowland (Site A) may also be explained by the presence of isolated crystalline bedrock highs or the presence of bedforms on steep slopes (Spagnolo et al., 2012). However, due to the role of glacial ice in shaping these crystalline bedrock highs, and the significance of other metrics that correspond to large variations in bedform surface elevation, these features with large surface elevation changes were included in the final dataset.
The streamlined bedforms present in the Bárðardalur valley (Site G) are lower in elevation than those found along the valley edges (Supporting Information Figure S1) and were therefore topographically controlled during ice thinning of the Icelandic Ice Sheet (Benediktsson et al., 2022).
However, it is unlikely that the bedforms with the highest surface relief in Bárðardalur (Site G) are entirely composed of diamicton like the bedforms with smaller surface relief. We suggest the high relief bedforms in Bárðardalur (Site G) are comprised of diamicton overlying bedrock cores from pre-existing topography; however, further investigation of these features is needed to confirm this interpretation.
Topographically unconstrained sites with lithified sedimentary bed conditions create bedforms with the most uniform elongation and surface relief (Figure 6), indicating these regions are most suitable for persistent, warm-based ice flow across the bed with well-developed processes of erosion and deposition in the subglacial environment (e.g., Benedikttson et al., 2016;Rattas & Piotrowski, 2003;Stokes et al., 2013). Lithified sedimentary sites that are topographically unconstrained have some of the greatest bedform densities (Table 2), highest orientation uniformity (Table 2; Figures 7 and 8), and smallest bedform relief and elongation, as previously mentioned (Figure 6), further suggesting these settings are favorable for persistent ice flow.

| CONCLUSIONS
The large, semi-automatically mapped dataset developed in this work provides key insight into topographic and bed lithology controls on ice flow properties that should be applied to understanding contemporary systems (King et al., 2009). The application of TPI developed in this study highlights its widespread ability to quickly map thousands of bedforms with little computational time, about half of the human error, and subjectivity with only a few shortcomings, including its difficulty in identifying low-relief, elongate subglacial bedforms across landscapes with prominent pre-glacial topography (Ebert et al., 2012;Hall et al., 2013). From the results of the dataset, we learn that landform signatures of warm-based ice flow have remarkable morphometric distribution similarities regardless of bed topography and lithology.
All regions of ice flow, measurable by the presence of streamlined bedforms, are capable of similar ice-flow velocity distributions regardless of bed characteristics. However, we find that topographically constrained sites have the most elongate bedforms of the entire dataset, indicating that topography has a first-order control on ice flow velocity and persistence, controlling streamlined bedform elongation through topographic funneling (Eyles et al., 2018;Hall & Glasser, 2003;Hindmarsh, 2001;Ottesen et al., 2008;Roberts et al., 2010;Wellner et al., 2001). In this case, it is interpreted that substrate properties control the glaciological environment to develop bedforms with distinct high elongation.
While bedform elongation is controlled by topography, sites with sedimentary beds contain the greatest number and area of bedforms per area, indicating that bedrock lithology has a more dominant control on density and packing than bed topography. Topographically unconstrained sedimentary beds support formation and development of bedforms with uniform elongation ratios, low bedform surface relief, uniform bedform orientation and high bedform density, indicating that these sites are most suitable for the development of persistent ice flow with well-organized subglacial erosive and depositional processes.
Due to the fundamental role of bed topography and substrate in determining ice dynamics (Clarke et al., 1977;Cuffey & Paterson, 2010;Greenwood et al., 2021;Whillians & van der Veen, 1997), assessment of streamlined bedform morphologies provides crucial information on bed-related controls to ice flow (Stokes & Clark, 2001, 2002King et al., 2009). As contemporary ice streams continue to retreat across environments with variable topography and bed lithology that are difficult to access and visualize, the use of preserved streamlined bedforms from paleo-subglacial environments is highly beneficial to constraining subglacial process sensitivities to variable bed conditions (Eyles et al., 2018;Greenwood et al., 2021;King et al., 2009;Stokes & Clark, 2001, 2002.

ACKNOWLEDGMENTS
We thank ESPL Editor Dr. Stuart Lane, Dr. Chris Clark, an anonymous reviewer, and A. P. Lepp for their comments and reviews which greatly improved the manuscript and clarity of our work. We acknowledge WADNR, USGS, and the Polar Geospatial Data Center data sources as well as A. Weiss, S. Tagil, and J. Jenness for making their data and code accessible. Much of the data analysis and interpretation presented in this study were conducted in Charlottesville, Virginia, on land that the Monacan Nation has protected and cultivated for thousands of years, and the authors acknowledge their ongoing stewardship of the lands. This work was funded by the Chamberlain Endowment and the H. G.
Goodell Endowment at the University of Virginia.