s41586-024-07236-z.pdf

Type: Document | Status: ready

792  |  Nature  |  Vol 628  |  25 April 2024 Article The improved prediction in the correlated effect model is a consequence of handling temporal, spatial and phylogenetic non-independence. For the Living Planet data, recognizing temporal correlation between sequential abundance values (ρ = 0.52) introduces nonlinearity (residual variability in linear trends) in temporal trends, and more closely represents realistic population dynamics (Fig. 3, popu- lation level). Comparably, temporal non-independence in the Living Planet data was higher than the average across the ten datasets (ρ = 0.31, s.d. = 0.42, range: −0.65 to 0.99). Accounting for this temporal structure in population trends can also influence trend direction and uncertainty of species-level and site-level trends, relative to the random slope model (Fig. 3, site level). At the global level, the presence of temporal, spatial (proportion of variance captured by spatial correlation term = 0.30) and phylogenetic (proportion of variance captured by phylogenetic correlation term = 0.34) structures elevated the uncertainty around the overall trend (Extended Data Fig. 2), ultimately leading to more robust inference. The presence of the observed spatial and phylogenetic structures also has the added benefit of allowing us to make predictions beyond the species and location data (Fig. 4), offering important insight into species and locations most likely to exhibit declines and recoveries. Our ability to predict a given species trend is dependent on the spe- cies being contained and accurately described in a phylogeny. Efforts continue to expand the breadth and quality of phylogenetic informa- tion28, and across our 10 datasets, we were able to obtain phylogenetic information for 80% of species. Implications for biodiversity science The abundance datasets we analyse are influential in policy, tracking progress towards biodiversity targets at national and international scales16,17,29, and so it is vital that any inference gained is both valid and reliable. Our work shows that when biodiversity change datasets are analysed without accounting for correlative non-independence among phylogeny, time and space, there is a substantial risk that trend uncertainty could be underestimated, trend direction misestimated and policy misinformed. Further, once uncertainty is appropriately attributed in the correlated effect model, we failed to detect a single significant trend in collective abundance across the 10 datasets under 95% credible intervals. This pervasive pattern points to a highly uncer- tain status in collective abundance trends; that is, it is unclear how biodiversity is changing across vast spatial and taxonomic scales once uncertainty is appropriately estimated. The random intercept model, used by 43% of studies, underestimated trend uncertainty 26-fold. The random slope model, used by 50% of studies, performs better but still underestimates uncertainty 3.4-fold. This underestimated uncertainty has a substantial impact on trend inference, for which 9 datasets have significant trends at 50% credible intervals in the random intercept model compared to 7 datasets in the random slope model, and just 4 datasets in the correlated effect model. At 95% credible intervals, we found 8 significant collective trends in the random intercept model, 4 significant trends in the random slope model and zero significant trends in the correlated effect model. This raises questions about the robustness of existing estimates of abun- dance change in the literature. Past estimates of abundance change have pointed to declines15,21, no net change18,22,23 and recovery24. This high variability across studies and datasets could be well founded given their different spatial, temporal and taxonomic scales, but it is also paradoxical given the expectation that biodiversity has declined under intense and widespread global change1. In the correlated effect model, we partially resolve this vari- ability between datasets, as our results generalize under the common feature of substantial uncertainty. However, the absence of signifi- cant trends in the correlated effect model also further emphasizes the paradox of failing to detect biodiversity loss despite rapid global change. Ours is not the first study to fail to detect declining abundances. For instance, previous work has shown that most abundance trends exhibit no net change23, and that the magnitude of decline reduces after accounting for extreme values18 and random abundance fluctua- tions19. Other work suggests the current data collection infrastructure is too small and biased to detect a trend reliably30. Similarly, analyses of BioTIME suggest that declines are unlikely because environmental change generates winners as well as losers2,22, whose opposing popula- tion trends may cancel each other out at global scales. Increasing evidence of decline Increasing evidence of growth CI threshold (%) 80 Latitude (º N) Longitude (º W) 80 60 60 40 40 20 20 a 40 30 20 130 120 110 100 90 80 70 60 45 35 25 55 60 50 b Fig. 4 | Abundance change varies over phylogenetic and spatial extents. Evidence of abundance change at different significance thresholds (for example, at an 80% CI threshold, dark red indicates evidence of declines whereas dark blue indicates evidence of increases). a, For the phylogeny, the species-level trends were derived by summing across hierarchical taxonomic random effects and phylogenetic correlation terms. Asymptotic species-level confidence thresholds were derived using uncertainty in phylogenetic predictions at multiple z-scores. To improve visualization, phylogenetic branch lengths are log transformed.
b, For space, we take taxonomic and phylogenetic information from a for one iconic and abundant North American species, the American robin Turdus migratorius, and combine this with hierarchical and correlative spatial terms
to make population-level predictions across terrestrial space. Asymptotic confidence thresholds were derived at the population scale (for example, species in a given site) using multiple z-scores. These predictions are drawn from the correlated effect model and BioTIME data (Supplementary Table 1).

Nature  |  Vol 628  |  25 April 2024  |  793 All things considered, it seems likely that collective abundance trends over varied taxa exhibiting varied responses to varied degrees of environmental change may not present as significant even with vast quantities of data. Perhaps a more considered approach is necessary, focused on describing which taxonomic groups and specific locations are declining and recovering, and why. The correlated effect model is particularly well placed for exploring this question, as the integration of space and phylogeny allows us to explore the particular locations and clades for which abundance trends shift from stable to decline. Although the high uncertainty around collective trends limits our gen- eral understanding of abundance changes, we observe an increase in accuracy of abundance forecasts and trend predictions under the correlated effect model, delivering a much-needed improvement in prediction at local scales (Fig. 4). The more complex representation of space, time and phylogeny is key to this improved prediction, for which, as an example, a priori information on evolutionary history can help predict which species are likely to decline and recover. Our new methods offer the hope of greater clarity in biodiversity trend estimation across different datasets and geographies, to inform and guide adaptive conservation policy responses. Despite failing to detect a decline in collective abundance across the ten datasets at 95% credible intervals, our results do not necessar- ily mean that wildlife abundances have not declined, or that the cur- rent estimates of trends are incorrect. It is possible that abundances may have increased on average, or perhaps declines have been far more extreme than we have previously imagined; simply, the uncer- tainty is too high to know. With this in mind, we re-emphasize calls to urgently expand data collection and improve trend detectability, but this will only help to estimate current and future biodiversity trends. Given the large uncertainty associated with our estimates of abundance change, it seems that past abundance patterns are lost and undetect- able at present. A shift towards causal frameworks of detection and attribution is probably necessary to estimate past biodiversity change14. In our study, we have solely focused on deriving the collective abun- dance trend given its potential political importance17, but the core statistical framework could be applied to other biodiversity data types (for example, occupancy data), adapted to other metrics (for example, species richness) and integrated into a global biodiversity observa- tion system14. For instance, the species trend coefficients could be extracted from the model and used alongside estimates of absolute species abundance to determine the absolute change in populations31. Weightings could be included to increase the influence of common species, allowing us to reconcile and test for differences between col- lective abundance trends, biomass decline and individuals lost. Precise and accurate estimates of abundance change in time and space also underpin a variety of policy-relevant facets of biodiversity, including ecosystem function through abundance-weighted functional diver- sity32,33, energy flux, and population stability and resilience34. The implications of our findings extend beyond biodiversity change. Spatial, temporal and phylogenetic data structures are common in ecological and evolutionary research. Under the presence of correlative non-independence, there is a potentially pervasive risk in the field that we have mis-specified our statistical models, violating data independ- ence assumptions, and producing unreliable inference. However, this also presents an opportunity, as the correlative effect model could be adapted for a wide array of settings, improving inference across ecology and evolution. The model could also streamline workflows by simultaneously capturing spatial, phylogenetic and temporal structures, avoiding the need to capture terms in multiple separate analyses. Our analytical advance offers new potential in predictive ecology, but given the severity of the potential implications of biodiversity loss35, it is vital that we continue to expand and improve these methods. We offer a general framework for addressing spatial, temporal and phy- logenetic non-independence, but further advances are necessary, considering underlying issues around time-series length36, bias and non-probability37, nonlinearity and varied responses to environmental change38, modern data collection philosophies39 and rigorous analysis approaches40. This combination of improved methods and data has the potential to reveal patterns of biodiversity change and disentangle the complex processes shaping our ecosystems. Online content Any methods, additional references, Nature Portfolio reporting summa- ries, source data, extended data, supplementary information, acknowl- edgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-024-07236-z. 1. Summary for Policymakers of the Global Assessment Report on Biodiversity and Ecosystem Services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES, 2019). 2. Dornelas, M. et al. BioTIME: a database of biodiversity time series for the Anthropocene. Glob. Ecol. Biogeogr. 27, 760–786 (2018). 3. Sauer, J. R. et al. The first 50 years of the North American Breeding Bird Survey. Condor 119, 576–593 (2017). 4. Pilotto, F. et al. Meta-analysis of multidecadal biodiversity trends in Europe. Nat. Commun. 11, 3486 (2020). 5. Comte, L. et al. RivFishTIME: a global database of fish time-series to study global change ecology in riverine systems. Glob. Ecol. Biogeogr. 30, 38–50 (2021). 6. Freshwater Fish Counts (UK Government, 2021, accessed 1 May 2023); https:// environment.data.gov.uk/dataset/ce2618db-d507-4671-bafe-840b930d2297. 7. Johnson, T. F., Cruz, P., Isaac, N. J. B., Paviolo, A. & González-Suárez, M. CaPTrends:
a database of large carnivoran population trends from around the world. Glob. Ecol. Biogeogr. 31, 2475–2482 (2022). 8. Living Planet Index Database (LPI, 2023, accessed 1 May 2023); https://www. livingplanetindex.org/data_portal. 9. Maureaud, A.A., Palacios-Abrantes, J., Kitchel, Z. et al. FISHGLOB_data: an integrated dataset of fish biodiversity sampled with scientific bottom-trawl surveys. Sci. Data 11, 24 (2024). 10. Quimbayo, J. P. et al. TimeFISH: Long-term assessment of reef fish assemblages in a transition zone in the Southwestern Atlantic. Ecology 104, e3966 (2023). 11. Jandt, U. et al. More losses than gains during one century of plant biodiversity change in Germany. Nature 611, 512–518 (2022). 12. Ripple, W. J. et al. World scientists’ warning to humanity: a second notice. BioScience 67, 1026–1028 (2017). 13. Pereira, H. M. et al. Essential biodiversity variables. Science 339, 277–278 (2013). 14. Gonzalez, A., Chase, J. M. & O’Connor, M. I. A framework for the detection and attribution of biodiversity change. Philos. Trans. R. Soc. B 378, 20220182 (2023). 15. Living Planet Report 2020 - Bending the Curve of Biodiversity Loss (WWF, 2020). 16. Ledger, S. E. H. et al. Past, present, and future of the Living Planet Index. npj Biodivers. 2, 12 (2023). 17. Geldmann, J., Byaruhanga, A., Gregory, R., Visconti, P. & Xu, H. Prioritize wild species abundance indicators. Science 380, 591–592 (2023). 18. Leung, B. et al. Clustered versus catastrophic global vertebrate declines. Nature 588, 267–271 (2020). 19. Buschke, F. T., Hagan, J. G., Santini, L. & Coetzee, B. W. T. Random population fluctuations bias the Living Planet Index. Nat. Ecol. Evol. 5, 1145–1152 (2021). 20. Loreau, M. et al. Do not downplay biodiversity loss. Nature 601, E27–E28 (2022). 21. van Klink, R. et al. Meta-analysis reveals declines in terrestrial but increases in freshwater insect abundances. Science 368, 417–420 (2020). 22. Dornelas, M. et al. A balance of winners and losers in the Anthropocene. Ecol. Lett. 22, 847–854 (2019). 23. Daskalova, G. N., Myers-Smith, I. H. & Godlee, J. L. Rare and common vertebrates span a wide spectrum of population trends. Nat. Commun. 11, 4394 (2020). 24. Haase, P. et al. The recovery of European freshwater biodiversity has come to a halt. Nature 620, 582–588 (2023). 25. Dornelas, M. et al. Quantifying temporal change in biodiversity: challenges and opportunities. Proc. R. Soc. B 280, 20121931 (2013). 26. Browning, E., Freeman, R., Boughey, K. L., Isaac, N. J. B. & Jones, K. E. Accounting for spatial autocorrelation and environment are important to derive robust bat population trends from citizen science data. Ecol. Indic. 136, 108719 (2022). 27. Dinnage, R., Skeels, A. & Cardillo, M. Spatiophylogenetic modelling of extinction risk reveals evolutionary distinctiveness and brief flowering period as threats in a hotspot plant genus. Proc. R. Soc. B 287, 20192817 (2020). 28. Parks, D. H. et al. Recovery of nearly 8,000 metagenome-assembled genomes substantially expands the tree of life. Nat. Microbiol. 2, 1533–1542 (2017). 29. Status of Priority Species: Relative Abundance (Department for Environment, Food & Rural Affairs, 2023). 30. Valdez, J. W. et al. The undetectability of global biodiversity trends using local species richness. Ecography 2023, e06604 (2023). 31. Rosenberg, K. V. et al. Decline of the North American avifauna. Science 366, 120–124 (2019). 32. Sol, D. et al. The worldwide impact of urbanisation on avian functional diversity. Ecol. Lett. 23, 962–972 (2020).