Hard‐To‐Sample Species Are More Sensitive to Land‐Use Change: Implications for Global Biodiversity Metrics

ABSTRACT Aim Land‐use change drives biodiversity loss, but some species are more vulnerable than others. Indicators of global biodiversity must attempt to summarise these impacts representatively and meaningfully, to guide biodiversity recovery. Yet species that are hard to detect, and thus feature less in relevant databases, might possess traits that make them particularly sensitive to anthropogenic impacts. Using global data for plant, bird and spider species, we develop a statistical approach to analyse and correct for the impact of excluding hard‐to‐sample species from global biodiversity indicators. Location Worldwide. Time Period Abundance studies published in 1998–2020; species occurrence records available from 1600 to 2023. Major Taxa Studied Birds, vascular plants and spiders. Methods We first quantified the extent to which the recordability of a species mediates the relationship between site‐level abundance and broad land use type. We used the local abundance data in the Projecting Responses of Ecological Diversity in Changing Terrestrial Systems database (PREDICTS), for over 4000 plant, bird and spider species. As a proxy for species' recordability, we used its number of occurrence records in the Global Biodiversity Information Facility database (GBIF). We then extrapolated our fitted statistical model to all species with valid GBIF occurrence records (0.27 M species). Results Less recordable species tend to decline more as land‐use intensity increases, and problematically, they are underrepresented in PREDICTS. A more representative global indicator can be obtained by extrapolating our model to the hard‐to‐sample, and on average, more sensitive species unobserved in PREDICTS. Our extrapolated, aggregate estimates show a lower abundance of ‘the average species’ in anthropogenic land uses. For example, intensive agriculture only has 18% of the biodiversity level of primary vegetation, rather than the 47% estimated without extrapolation to the hard‐to‐sample species. Main Conclusions Given the bias encountered in PREDICTS and the considerable difference in abundance change estimations, we recommend that other existing indicators include an extrapolation solution based on ours to incorporate the available data as effectively as possible. Using occurrence data to predict species' sensitivity unlocks many possibilities to improve global biodiversity indicators by enhancing their overall coverage and accuracy, without demanding additional data on poorly known species.