wvSlpFailureML: A dataset for slope failure occurrence predictive modeling using machine learning and LiDAR -derived topographic variables for the entirety of the state of West Virginia, USA.

This dataset consists of 575,670 point locations across the state of West Virginia. Those mapped to the “slpF” class (n = 116,413) were interpreted to be initiation locations of slope failures (e.g., landslides, debris flows, rock topples, etc.). Each slope failure location was collected as a point feature at the interpreted slope failure initiation location by an analyst then subsequently checked by another interpreter. Interpretation primarily relied on light detection and ranging (LiDAR)-derived hillshades and slopeshades and other ancillary geospatial data. All non-slope failure points are labeled as “not” (n=459,257) and were collected as random points at least 30 meters away from the mapped slope failure locations and occurring within the state extent. The “not” points serve as pseudo-absence samples. All topographic variables were derived from a 2 m spatial resolution digital terrain model (DTM) produced from LiDAR data. For topographic measures requiring a moving window, a circular window with a 7 cell radius was used. Physiographic regions are defined relative to Major Land Resource Areas (MLRAs): https://www.nrcs.usda.gov/resources/data-and-reports/major-land-resource-area-mlra. The wvSlpFailures.csv file provides only the attributes while the wvSlpFailures.shp file includes the spatial coordinates. We have included a Word document and PDF that further describes the data.