Data and code for "Landslide susceptibility peaks at intermediate vegetation density: a global satellite analysis of critical NDVI thresholds"

Curated data and reproducible analysis code accompanying the manuscript "Landslide susceptibility peaks at intermediate vegetation density: a global satellite analysis of critical NDVI thresholds" (iScience, manuscript ISCIENCE-D-26-03863). The archive contains a curated global inventory of 18,028 rainfall-triggered landslides (2000–2024) with pre-failure MODIS/Sentinel-2 vegetation indices, multi-source land cover, climate zone, and SRTM-derived terrain; the R and Python scripts that reproduce the segmented-regression breakpoint detection, Random Forest validation, all sensitivity analyses, and every figure and table; and the supplementary figures and tables. The analysis demonstrates a unimodal, non-monotonic ("vegetation paradox") relationship in which landslide susceptibility peaks within a Critical Transition Zone at NDVI 0.769–0.868. Note: an apparent pre-failure NDVI "early-warning" decline reported in earlier drafts is shown here (rev_B_contamination.R) to be an artefact of the 16-day MODIS composite straddling the failure date; we make no early-warning claim, and the central non-monotonic result is unaffected.