Data and code for Global vegetation trend decomposition depends on ecosystem type and CO2 specification

This archive contains the R code, dataset provenance, input-file fingerprints, summary data, validation outputs and reference figures supporting the study “Global vegetation trend decomposition depends on ecosystem type and CO2 specification”. The analysis covers 2000–2020 at a 0.05-degree working grid and distinguishes five temporally stable vegetation classes. Ecosystem-stratified generalised additive models include site random intercepts and AR(1) errors. The primary model includes a spherical spatial smooth and CO2; a matched 2 × 2 design tests sensitivity to these two terms. Trend inference uses Newey–West standard errors and global Benjamini–Hochberg correction. Global and ecosystem summaries use actual pixel area, with spatial-block bootstrap intervals and spatial-block cross-validation. The archive includes pooled out-of-fold level, trend and sign diagnostics, area-representative held-out trend shares, five CO2 spatial-support scenarios, and machine-readable CSV versions of all 30 parts of Supplementary Tables 1–9. The environment–CO2 baseline and unexplained-residual shares are descriptive co-directional trend components. Their association with composite human pressure is exploratory and does not identify causal effects of human activity. Raw third-party datasets, high-resolution derived rasters, site-level source and prediction tables, and model objects are excluded where redistribution rights are unclear. Users must acquire source data from the official providers and follow the documented reconstruction workflow. The included Natural Earth coastline is public domain.