Data and code for: Can we trust vegetation resurveys?

Data and code for: Can we trust vegetation resurveys? This repository contains the data and R scripts used in the manuscript "Can we trust vegetation resurveys?" (Journal of Ecology, JEcol-2025-1831.R2). We compiled 20 permanent-plot monitoring time series from the ReSurveyEurope and LOTVS databases, spanning dry to wet European grasslands sampled at least 12 times (often annually) over 14–44 years. For each dataset, we extracted all possible pairs of surveys (a "two-visit resurvey") and compared trend estimates derived from these pairs to trends estimated from the full time series, using Bayesian multilevel models (brms/Stan). We evaluated trends in species richness and in occurrence- and abundance-based community turnover (Sørensen and Bray-Curtis dissimilarity, respectively). Data provenance and anonymization. The datasets provided here are anonymized derivatives (plot identifiers, site codes, and geographic information have been removed or recoded) sufficient to reproduce the analyses and figures in the manuscript. They are not the original vegetation survey data. The original, full monitoring time series belong to the ReSurveyEurope and LOTVS databases and must be requested directly from those resources; they are not redistributed here. See https://www.resurveyeurope.org and the LOTVS database (Sperandii et al. 2022) for access procedures. Contents Scripts A2_analysis.R: Main analysis. Splits each time series into a "small" (two-visit) dataset and a "large" (remaining-visits) dataset for every possible survey pair. Fits Bayesian multilevel models (Poisson/negative binomial for richness; Gaussian for turnover metrics) with site-level random effects using brms, run in parallel. Computes posterior differences between two-visit and trend model estimates, plus R-hat, effective sample size, and posterior predictive variance ratio diagnostics. Produces meta/two_visit_resurveys.csv. A3_postprocessing.R: Post-processing and figures. Filters/reconciles model family choices (Poisson vs. negative binomial), calculates the coefficient of efficiency and credible interval ranges, and generates Figure 1 (two-visit vs. trend model estimates) and Figure S1 (model diagnostics: R-hat and effective sample size).