Advancing causal inference in ecology: Pathways for biodiversity change detection and attribution

作者：Franziska Schrodt, Miriam Beck, Joaquim Estopinan, Diana E. Bowler, Colin Fontaine, Pierre Gaüzère, Romain Goury, Matthias Grenié, Inês S. Martins, Naia Morueta‐Holme, Luca Santini, Mickaël Hedde, Gabrielle Martin, Emmanuelle Porcher, Sara Si‐Moussi, Marianne Tzivanopoulos, Grant Vernham, Cyrille Violle, Wilfried Thuiller · 发表于：Methods in Ecology and Evolution · 年份：2025 · DOI：10.1111/2041-210x.70131 · 被引用次数：25 · 研究领域：Advanced Causal Inference Techniques、Bayesian Modeling and Causal Inference、Species Distribution and Climate Change

Abstract Understanding the causes of biodiversity change is essential for addressing environmental challenges. While causal attribution has advanced in other fields, ecologists remain cautious about causal claims or misinterpret predictive models as causal. With growing spatio‐temporal data, computational power and cross‐disciplinary collaboration, discussions on improving attribution methods in ecology are gaining momentum. However, practical guidance remains limited for non‐experts. Here, we identify the challenges and decisions involved in detecting and attributing biodiversity change and provide an overview of suitable methods based on available data and specific research questions. The first challenge we address pertains to biodiversity and driver data. Unlike controlled experimental data in other disciplines, ecological data often stem from monitoring programs or field samplings with varying degrees of rigour, which complicates the analysis due to sampling biases, interacting drivers, measurement error or spatio‐temporal variations. We specifically outline how data structure (e.g. structured vs. opportunistic data) and data coverage along the spatial and temporal scale impact detection and attribution. The second challenge involves the ability to detect directional change in the system of interest, which is associated with numerous hurdles. We provide an overview of the most relevant approaches to deal with sampling variability, gaps and biases in the data, non‐linearit...