Deliverable D3.1 Best practices for Detection Attribution Modelling

Detecting and attributing biodiversity changes is a multifaceted and demanding task. The first key challenge is gathering data on biodiversity metrics and the likely drivers that is sufficiently structured and aligned in space and time, and wide enough to cover the dynamical range of the target latent processes at play, enabling statistical inference. Demonstrating that a measure of biodiversity has significantly changed relative to a reference state — a reference which is often difficult to define due to a lack of past data — constitutes a second challenge. A third key challenge is designing an identification strategy that can isolate the contribution of multiple potential causal factors with statistical confidence.

The review comprising the deliverable D3.1 addresses these three key challenges in a coherent framework, meeting the task expectations. It is entitled "Advancing Causal Inference in Ecology: Pathways for Biodiversity Change Detection and Attribution" (Schrodt et al., Methods in Ecology and Evolution, under revision). This work was achieved in collaboration with the IMPACTS synthesis group of the French Foundation for Biodiversity Research (FRB). This text provides conceptual and practical guidance on taking advantage of existing causal methods to detect and attribute changes in biodiversity. There is an emphasis on how remote sensing data can mitigate pressing issues related to confounding factors that occur across scales.

By paying attention to the described challenges and relying on the suggested methods and workflow, the review introduces a solid basis to root biodiversity change studies in causal principles for better detection and attribution. The proposed manuscript is indeed highly interdisciplinary in its attempt to bring biodiversity studies closer to the science of attribution through causal inference from observational data. While this deliverable is fully autonomous, it is complemented by two perspective articles that are also under revision and a method decision tool that is under development. They cover related aspects of detection and attribution.

As deliverable D3.1 format is a scientific manuscript, it is provided in its most recent version in Annex 1 below.