There is little appreciation of the level of extinction risk faced by one-sixth of the over 65,000 species assessed by the International Union for Conservation of Nature. Determining the status of these data-deficient (DD) species is essential to developing an accurate picture of global biodiversity and identifying potentially threatened DD species. To address this knowledge gap, we used predictive models incorporating species' life history, geography, and threat information to predict the conservation status of DD terrestrial mammals. We constructed the models with 7 machine learning (ML) tools trained on species of known status. The resultant models showed very high species classification accuracy (up to 92%) and ability to correctly identify centers of threatened species richness. Applying the best model to DD species, we predicted 313 of 493 DD species (64%) to be at risk of extinction, which increases the estimated proportion of threatened terrestrial mammals from 22% to 27%. Regions predicted to contain large numbers of threatened DD species are already conservation priorities, but species in these areas show considerably higher levels of risk than previously recognized. We conclude that unless directly targeted for monitoring, species classified as DD are likely to go extinct without notice. Taking into account information on DD species may therefore help alleviate data gaps in biodiversity indicators and conserve poorly known biodiversity.

34 Aim 35Evaluating the relative roles of biological traits and environmental factors that predispose 36 species to an elevated risk of extinction is of fundamental importance to macroecology. Results 54Range size was the most important predictor of extinction risk, reflecting the high frequency 55 of reptiles assessed under range-based IUCN criteria. Habitat specialists occupying accessible 56 ranges were at a greater risk of extinction: although these factors never contributed more than 57 10% to the variance in extinction risk, they showed significant interactions with range size. 58Predictive power of our global models ranged between 23 and 29%. The general overall 59 pattern remained the same among geographic, taxonomic and threat-specific data subsets. Main conclusions 62Proactive conservation requires shortcuts to identify species at high risk of extinction.

In response to growing demand for ecosystem-level risk assessment in biodiversity conservation, and rapid proliferation of locally tailored protocols, the IUCN recently endorsed new Red List criteria as a global standard for ecosystem risk assessment. Four qualities were sought in the design of the IUCN criteria: generality; precision; realism; and simplicity. Drawing from extensive global consultation, we explore trade-offs among these qualities when dealing with key challenges, including ecosystem classification, measuring ecosystem dynamics, degradation and collapse, and setting decision thresholds to delimit ordinal categories of threat. Experience from countries with national lists of threatened ecosystems demonstrates well-balanced trade-offs in current and potential applications of Red Lists of Ecosystems in legislation, policy, environmental management and education. The IUCN Red List of Ecosystems should be judged by whether it achieves conservation ends and improves natural resource management, whether its limitations are outweighed by its benefits, and whether it performs better than alternative methods. Future development of the Red List of Ecosystems will benefit from the history of the Red List of Threatened Species which was trialed and adjusted iteratively over 50 years from rudimentary beginnings. We anticipate the Red List of Ecosystems will promote policy focus on conservation outcomes in situ across whole landscapes and seascapes.