Landslide susceptibility probability tables and sample points from Reconstructability Analysis, Wilson-Trask-Nestucca and Siletz-Yaquina watersheds, Oregon Coast Range, USA

Sample points and conditional landslide probability tables for the Wilson-Trask-Nestucca and Siletz-Yaquina (WTNSY) watersheds of the Oregon Coast Range, USA. The probability tables were produced with Reconstructability Analysis (RA), an information-theoretic modelling method that works directly on categorical data. The data accompany Percy et al., “Information-theoretic modeling of landslide susceptibility using mostly categorical spatial data”, submitted to Natural Hazards and Earth System Sciences. The study area was stratified into volcanic and sedimentary lithologies and modelled separately. Each row of a probability table is one combination of predictor states (for example elevation bin, terrane group, landform, soil taxonomic subgroup and particle size class) with the model probability of landslide presence, P(LS=1), the predicted class, and the number of training samples (n) from which the probability was estimated. Landslide presence comes from the Statewide Landslide Information Database for Oregon (SLIDO-4.5, Oregon Department of Geology and Mineral Industries); predictor variables come from DOGAMI geologic mapping, the USDA-NRCS Soil Survey Geographic Database (SSURGO), NLCD 2021 land cover and terrain data. Categories below 1% of a stratum's sample count were collapsed into a single category labelled “(other/rare)”. All files are in WTNSY_landslide_RA_data_v1.0.zip, which unpacks to a single folder (README.txt is also provided separately): • volcanic_probability_table.csv and sedimentary_probability_table.csv: flat, human-readable tables covering the full state space (144 volcanic and 512 sedimentary predictor-state combinations). • landslide_probability_tables.sqlite: the same tables in normalized form (a value table for each predictor and a prediction table for each stratum, linked by integer keys), with views volcanic_readable and sedimentary_readable that rebuild readable tables for browsing; these views leave out the (other/rare) categories, so use the CSV files for the complete state space. • samples_volcanic.csv and samples_sedimentary.csv: the sample points (6,948 volcanic, 12,559 sedimentary) with landslide presence, WGS84 coordinates, and their terrain, soil, geology and land-cover attributes. • README.txt and README_samples.txt: column descriptions, schema notes and usage examples for SQLite, Python, R and DuckDB. The models were built with PyOccam v0.9.5 (https://doi.org/10.5281/zenodo.20352230).