Landslide susceptibility modeling results as vector shapefile data offering a broad assessment of landslide hazards across five regions of Minnesota, USA. Data was created as part of an investigation to understand the link between a previously mapped landslide inventory, various environmental variables, and post-glacial landscape development through multivariate logistic regression analyses.

Landslide susceptibility maps are essential tools in infrastructure planning, hazard mitigation, and risk reduction. Susceptibility maps trained in one area have been found to be unreliable when applied to different areas (Woodard et al., 2023). This limitation leads to the need for a national map that is higher resolution and rigorous, but simple enough to be applied to diverse terrains and landslide types. The susceptibility maps presented here cover the conterminous United States (CONUS), Alaska (AK), Hawaii (HI), and Puerto Rico (PR) with a resolution of 90-m. Other United States (U.S.) territories were not considered due to insufficient landslide and digital elevation data. We also provide information on the proportion of susceptible terrain as well as the density (landslides per square kilometer) of documented landslides within susceptible terrain for each U.S. county. To generate the susceptibility maps we used 1/3 arc-second digital elevation models (DEMs) (U.S. Geological Survey, 2019) to calculate slope and 100-m relief, 613,724 unique landslides from our national landslide inventory compilation (Belair et al., 2022) to train the models and compute U.S. county aggregated susceptibility information, and high-performance computing resources to train the models (Falgout and Gordon, 2023). We present two slope-relief threshold models: (1) a linear regression model weighted by landslide density of each ecoregion (Wiken et al., 2011), and (2) a quantile nonlinear regression model fitted to the 10th quantile of the data. We (1) removed extraneous landslide data, (2) averaged 50 model runs, and then (3) down-sampled the maps from 10-m to 90-m resolution to account for uncertainty in the DEM and landslide position. The nonlinear model (n10) performs better under most topographic conditions and optimally balances our priorities of capturing observed landslides (98.9%) while minimizing area covered by susceptible terrain (44.6%). The weighted linear model (lw) captures slightly fewer landslides (98.8%) and has slightly less susceptible terrain (43.1%). The values of both maps represent the number of susceptible 10-m cells within each 90-m cell after down-sampling and can range from 0 to 81. While landslides are possible within any cells containing susceptible terrain, those with the highest concentration (or cell value) capture the majority of landslides, thus representing higher susceptibility areas. The susceptibility maps were then used to determine the total area of landslide susceptible terrain (square kilometers) for each U.S. county. The national landslide inventory compilation was used to determine the number of documented landslides within susceptible terrain for each county. This information was then used to calculate the proportion of susceptible terrain and the density of documented landslides within susceptible terrain for each county in the United States. This information is provided in tabular format, with columns corresponding to the information discussed above, and each row corresponding to a U.S. county. Further information about this analysis can be found in an interpretive publication (Mirus et al., 2024). This data release includes: (1) weighted linear susceptibility maps (lw_susc.zip), (2) quantile nonlinear susceptibility maps (n10_susc.zip), (3) landslide data used to develop the models (landslides.csv), (4) county aggregated susceptibility information (county_analysis.csv), (5) readme and analysis files, and (6) metadata. References Cited Belair, G. M., Jones, E. S., Slaughter, S. L., and Mirus, B. B., 2022, Landslide Inventories across the United States version 2: U.S. Geological Survey data release, https://doi.org/10.5066/P9FZUX6N Falgout, J. T., and Gordon, J., 2023, USGS Advanced Research Computing, USGS Yeti Supercomputer: U.S. Geological Survey, https://doi.org/10.5066/F7D798MJ Mirus, B. B., Belair, G. M., Wood, N. J., Jones, J. M., and Martinez, S. M., 2024, Parsimonious high-resolution

Landslide susceptibility maps are essential tools in infrastructure planning, hazard mitigation, and risk reduction. Susceptibility maps trained in one area have been found to be unreliable when applied to different areas (Woodard et al., 2023). This limitation leads to the need for a national map that is higher resolution and rigorous, but simple enough to be applied to diverse terrains and landslide types. The susceptibility maps presented here cover the conterminous United States (CONUS), Alaska (AK), Hawaii (HI), and Puerto Rico (PR) with a resolution of 90-m. Other United States (U.S.) territories were not considered due to insufficient landslide and digital elevation data. We also provide information on the proportion of susceptible terrain as well as the density (landslides per square kilometer) of documented landslides within susceptible terrain for each U.S. county. To generate the susceptibility maps we used 1/3 arc-second digital elevation models (DEMs) (U.S. Geological Survey, 2019) to calculate slope and 100-m relief, 613,724 unique landslides from our national landslide inventory compilation (Belair et al., 2022) to train the models and compute U.S. county aggregated susceptibility information, and high-performance computing resources to train the models (Falgout and Gordon, 2023). We present two slope-relief threshold models: (1) a linear regression model weighted by landslide density of each ecoregion (Wiken et al., 2011), and (2) a quantile nonlinear regression model fitted to the 10th quantile of the data. We (1) removed extraneous landslide data, (2) averaged 50 model runs, and then (3) down-sampled the maps from 10-m to 90-m resolution to account for uncertainty in the DEM and landslide position. The nonlinear model (n10) performs better under most topographic conditions and optimally balances our priorities of capturing observed landslides (98.9%) while minimizing area covered by susceptible terrain (44.6%). The weighted linear model (lw) captures slightly fewer landslides (98.8%) and has slightly less susceptible terrain (43.1%). The values of both maps represent the number of susceptible 10-m cells within each 90-m cell after down-sampling and can range from 0 to 81. While landslides are possible within any cells containing susceptible terrain, those with the highest concentration (or cell value) capture the majority of landslides, thus representing higher susceptibility areas. The susceptibility maps were then used to determine the total area of landslide susceptible terrain (square kilometers) for each U.S. county. The national landslide inventory compilation was used to determine the number of documented landslides within susceptible terrain for each county. This information was then used to calculate the proportion of susceptible terrain and the density of documented landslides within susceptible terrain for each county in the United States. This information is provided in tabular format, with columns corresponding to the information discussed above, and each row corresponding to a U.S. county. Further information about this analysis can be found in an interpretive publication (Mirus et al., 2024). This data release includes: (1) weighted linear susceptibility maps (lw_susc.zip), (2) quantile nonlinear susceptibility maps (n10_susc.zip), (3) landslide data used to develop the models (landslides.csv), (4) county aggregated susceptibility information (county_analysis.csv), (5) readme and analysis files, and (6) metadata. References Cited Belair, G. M., Jones, E. S., Slaughter, S. L., and Mirus, B. B., 2022, Landslide Inventories across the United States version 2: U.S. Geological Survey data release, https://doi.org/10.5066/P9FZUX6N Falgout, J. T., and Gordon, J., 2023, USGS Advanced Research Computing, USGS Yeti Supercomputer: U.S. Geological Survey, https://doi.org/10.5066/F7D798MJ Mirus, B. B., Belair, G. M., Wood, N. J., Jones, J. M., and Martinez, S. M., 2024, Parsimonious high-resolution

Hurricane Maria induced about 70,000 landslides throughout Puerto Rico, USA (Hughes and others, 2019, https://doi.org/10.5066/P9BVMD74). Data in this project pertain to two areas situated in Puerto Rico’s rugged Cordillera Central range. Combined, these areas account for more than half of the hurricane-induced landslides. One of these areas encloses two neighboring municipalities, Lares Municipio, and Utuado Municipio, and the second area encloses Naranjito Municipio. These data include one-meter (1-m) resolution raster grids derived from post-hurricane light detection and ranging (lidar) digital elevation models (DEM) available at https://apps.nationalmap.gov/lidar-explorer/#/. The elevation data as well as slope and flow accumulation grids derived from them were the primary inputs for soil-depth models and slope-stability models. We used outputs from these models to map susceptibility to landslide initiation and evaluate future landslide impacts from storms like Hurricane Maria for these three municipalities. The data accompany an interpretive paper that is currently under review. The area covering Lares and Utuado is divided into four overlapping tiles. A fifth tile covers Naranjito. Digital elevation model (DEM) tiles extend far enough into neighboring tiles and municipalities to allow assessment of flow accumulation as well as landslide runout and debris-flow inundation in watersheds that straddle boundaries. The raster grids are grouped into zip archives according to their associated tiles and each zip archive contains a complete set of nine 1-m resolution input and output grids of the following data types and naming conventions, in which the "*" identifies the tile location as well as selected model options: *pm.tif - 1m- lidar-derived digital elevation model (DEM), with elevation in meters nar1m*floacc.tif or larutu*floacc.tif - D8 flow accumulation raster, used as input for soil depth estimates *_Zn.tif - input-parameter-zone codes (unitless) to define geologic terrane, used as input for soil depth and slope-stability estimates RG_slope_*.tif - slope of the ground surface (degrees) computed using REGOLITH 1.0 (https://doi.org/10.5066/P9U2RDWJ) and used as input for slope-stability analysis RG_NASD_smo_*.tif - soil depth (meters) estimates computed using REGOLITH 1.0 (https://doi.org/10.5066/P9U2RDWJ) and used as input for slope-stability analysis TRfs_min_*.tif- 1d factor of safety (unitless) computed using TRIGRS 2.1 (https://doi.org/10.5066/F7M044QS) TRp_at_fs_min_*.tif- 1d pressure head (meters) computed using TRIGRS 2.1 (https://doi.org/10.5066/F7M044QS) and used as input for 3D slope-stability analysis SL3_fs3d_*.tif - quasi-three-dimensional (3D) factor of safety (unitless) for trial surface centered at each grid cell computed using Slabs3D 1.0 (https://doi.org/10.5066/P9G4I8IU) SL3_fs3dmn_*.tif - minimum 3D factor of safety (unitless) for any trial surface intersecting each grid cell computed using Slabs3D 1.0 (https://doi.org/10.5066/P9G4I8IU) Each tile has six asociated metadata files. Parameter inputs to and outputs from the programs REGOLITH 1.0, TRIGRS 2.1, and Slabs3D 1.0 are grouped into corresponding metadata files for each tile. The metadata files have the following naming convention, in which the "*" is a code that identifies the tile ("larutu1," or "a1," "larutu2," or "a2," "larutu3," or "a3," and "larutu4," or "a4," for the Lares and Utuado tiles and "nar" for the Naranjito tile): *pm.xml - 1m-lidar-derived DEM *floacc.xml - D8 flow accumulation raster *_Zn.xml - input-parameter-zone raster RG_*.xml - REGOLITH 1.0 parameter input file, slope output raster and soil-depth output raster TR_*.xml - TRIGRS 2.1 parameter input file, pressure head output raster and local factor of safety output raster SL3_*.xml - Slabs3D 1.0 parameter input file, 3D factor of safety output rasters Parameter input files accompany this data release, with the following naming convention, in which the "*" is a code that

Landslides are damaging and deadly, and they occur in every U.S. state. However, our current ability to understand landslide hazards at the national scale is limited, in part because spatial data on landslide occurrence across the U.S. varies greatly in quality, accessibility, and extent. Landslide inventories are typically collected and maintained by different agencies and institutions, usually within specific jurisdictional boundaries, and often with varied objectives and information attributes or even in disparate formats. The purpose of this data release is to provide an openly accessible, centralized map of existing information on landslide occurrence across the entire U.S. The data release includes digital inventories created by both USGS and non-USGS authors. It provides an integrated database of all the landslides with a selection of uniform attributes, but also includes links to the original digital inventory files (whenever available). Given the wide range of landslide information sources in this data compilation, we also provide an attribute to assess the relative confidence in the characterization of the location and extent of each landslide. Further details about each landslide and more recent information (when it exists) can be accessed by clicking the “more information” attribute link to the original source information. This database will be updated intermittently and was most recently updated in March 2019. Please contact gs-haz_landslides_inventory@usgs.gov for more information on how to contribute additional inventories to this community effort.

This map shows the relative likelihood of deep landsliding based on regional estimates of rock strength and steepness of slopes. On the most basic level, weak rocks and steep slopes are more likely to generate landslides. This shows the distribution of one very important component of landslide hazard. It is intended to provide infrastructure owners, emergency planners and the public with a general overview of where landslides are more likely. The map does not include information on landslide triggering events, such as rainstorms or earthquake shaking, nor does it address susceptibility to shallow landslides such as debris flows. This map is not appropriate for evaluation of landslide potential at any specific site. For visualization: If gridcode is 8,9,10 than area is High Susceptibility for landslides

Landslides are damaging and deadly, and they occur in every U.S. state. However, our current ability to understand landslide hazards at the national scale is limited, in part because spatial data on landslide occurrence across the U.S. varies greatly in quality, accessibility, and extent. Landslide inventories are typically collected and maintained by different agencies and institutions, usually within specific jurisdictional boundaries, and often with varied objectives and information attributes or even in disparate formats. The purpose of this data release is to provide an openly accessible, centralized map of existing information on landslide occurrence across the entire U.S. The data release includes digital inventories created by both USGS and non-USGS authors. It provides an integrated database of all the landslides with a selection of uniform attributes, but also includes links to the original digital inventory files (whenever available). Given the wide range of landslide information sources in this data compilation, we also provide an attribute to assess the relative confidence in the characterization of the location and extent of each landslide. The confidence level reflects the resolution and quality of input data, as well as the method used for identification and mapping of each landslide. We include a classification attribute for polygons to differentiate between individually mapped landslides and features that are evidence of landsliding. Such evidence may include landslide complexes, Quaternary landslide deposits, alluvial fans, unstable slopes, landslide impacts, and other evidence indicative of landslide occurrence. Further details about each landslide and more recent information (when it exists) can be accessed by clicking the “InventoryURL” attribute link to the original source information. Relative to the initial data release (version 1), this update (version 2) includes more inventories, updated confidence rules, and a new classification attribute. This database will be updated intermittently using the version 2 doi, and was most recently updated in March 2022. Please contact gs-haz_landslides_inventory@usgs.gov for more information on how to contribute additional inventories to this community effort.

Landslides are damaging and deadly, and they occur in every U.S. state. However, our current ability to understand landslide hazards at the national scale is limited, in part because spatial data on landslide occurrence across the U.S. varies greatly in quality, accessibility, and extent. Landslide inventories are typically collected and maintained by different agencies and institutions, usually within specific jurisdictional boundaries, and often with varied objectives and information attributes or even in disparate formats. The purpose of this data release is to provide an openly accessible, centralized map of existing information on landslide occurrence across the entire U.S. The data release includes digital inventories created by both USGS and non-USGS authors. It provides an integrated database of all the landslides with a selection of uniform attributes, but also includes links to the original digital inventory files (whenever available). Given the wide range of landslide information sources in this data compilation, we also provide an attribute to assess the relative confidence in the characterization of the location and extent of each landslide. The confidence level reflects the resolution and quality of input data, as well as the method used for identification and mapping of each landslide. We include a classification attribute for polygons to differentiate between individually mapped landslides and features that are evidence of landsliding. Such evidence may include landslide complexes, Quaternary landslide deposits, alluvial fans, unstable slopes, landslide impacts, and other evidence indicative of landslide occurrence. Further details about each landslide and more recent information (when it exists) can be accessed by clicking the “InventoryURL” attribute link to the original source information. Relative to the initial data release (version 1), this update (version 2) includes more inventories, updated confidence rules, and a new classification attribute. This database will be updated intermittently using the version 2 doi, and was most recently updated in March 2022. Please contact gs-haz_landslides_inventory@usgs.gov for more information on how to contribute additional inventories to this community effort.

1. Abstract Landslides are damaging and deadly, and they occur in every U.S. state. However, our current ability to understand landslide hazards at the national scale is limited, in part because spatial data on landslide occurrence across the U.S. varies greatly in quality, accessibility, and extent. Landslide inventories are typically collected and maintained by different agencies and institutions, usually within specific jurisdictional boundaries, and often with varied objectives and information attributes or even in disparate formats. The purpose of this data release is to provide an openly accessible, centralized map of existing information about landslide occurrence across the entire U.S. This data release is an update of previous versions 1 (Jones and others, 2019) and 2 (Belair and others, 2022). Changes relative to version 2 are summarized in us_ls_v3_changes.txt. It provides an integrated database of the landslides from these inventories (refer to US_Landslide_v3_gpkg) with a selection of uniform attributes, including links to the original digital inventory files (whenever available) (“Inv_URL”). The data release includes digital inventories created by both USGS and non-USGS authors. The original inventory is denoted by an abbreviation in the “Inventory” attribute. The full citation for each abbreviation can be found in us_ls_v3_references.csv. The date of the landslide event is included as a minimum and maximum (“Date_Min” and “Date_Max”) to accommodate events that happen within a range of dates. The date value is inherently difficult to interpret or discern due to the nature of landsliding, where some landslides move for long periods of time or move intermittently, and some areas can exhibit multiple landslide events. To preserve the constituent inventories as much as possible, we include all entries even if they are not considered landslides, such as “gullies” or “avalanche chutes.” We include a landslide type attribute when that information is available (“LS_Type”). The landslide classification system used in the original inventories is not always known or stated in the metadata, but many mapping entities use the schema from Cruden and Varnes (1996) or the updated schema from Hungr and others (2014). Given the wide range of landslide information sources in this data compilation, we provide an attribute to assess the relative confidence in the characterization of the location and extent of each landslide (entry) (“Confidence”). The confidence level reflects the resolution and quality of input data, as well as the method used for identification and mapping. This confidence does not reflect a formal accuracy assessment of field attributes. Relative to the previous data releases (version 1 and 2), this update (v3) includes more inventories, updated confidence rules, a new landslide type attribute, a new unique identifier (“USGS_ID”), new machine-readable date fields, and an ancillary database containing all fields from the original inventories (refer to US_Landslide_v3_ancillary). Please contact gs-haz_landslides_inventory@usgs.gov for more information on how to contribute additional inventories to this community effort. When possible, please cite the constituent inventories as well as this data release. This data release includes: (1) a landslide point file and polygon file in multiple forms (US_Landslide_v3_gpkg, US_Landslide_v3_shp, US_Landslide_v3_csv), (2) an ancillary database with original fields (US_Landslide_v3_ancillary), (3) a spreadsheet that summarizes the confidence rules, their justification, and any extra analyses (us_ls_v3_analyses.csv), (4) a summary file of the changes made between version 2 and version 3 (us_ls_v3_changes.txt), (5) a file containing the references of the constituent inventories (us_ls_v3_references.csv), (6) and a readme (README.txt). Disclaimer: Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. 2. Data