Humid tropical vertebrates are at lower risk of extinction and population decline in forests with higher structural integrity

Reducing deforestation underpins global biodiversity conservation efforts. However, this focus on retaining forest cover overlooks the multitude of anthropogenic pressures that can degrade forest quality and imperil biodiversity. We use remotely sensed indices of tropical rainforest structural condition and associated human pressures to quantify the relative importance of forest cover, structural condition and integrity (the cumulative effect of condition and pressures) on vertebrate species extinction risk and population trends across the global humid tropics. We found that tropical rainforests of high integrity (structurally intact and under low pressures) were associated with lower likelihood of species being threatened and having declining populations, compared with forest cover alone (without consideration of condition and pressures). Further, species were more likely to be threatened or have declining populations if their geographic ranges contained high proportions of degraded forest than if their ranges contained lower proportions of forest cover but of high quality. Our work suggests that biodiversity conservation policies to preserve forest integrity are now urgently required alongside ongoing efforts to halt deforestation in the hyperdiverse humid tropics. Not all forest cover is of equal quality. Here, the authors ask whether forest cover or forest structural complexity influences extinction risk in tropical rainforest vertebrates, finding that forest structural conditions are more important than cover alone in terms of buffering species against extinction and population declines.

Reducing deforestation underpins global biodiversity conservation efforts. However, this focus on retaining forest cover overlooks the multitude of anthropogenic pressures that can degrade forest quality and imperil biodiversity. We use remotely sensed indices of tropical rainforest structural condition and associated human pressures to quantify the relative importance of forest cover, structural condition and integrity (the cumulative effect of condition and pressures) on vertebrate species extinction risk and population trends across the global humid tropics. We found that tropical rainforests of high integrity (structurally intact and under low pressures) were associated with lower likelihood of species being threatened and having declining populations, compared with forest cover alone (without consideration of condition and pressures). Further, species were more likely to be threatened or have declining populations if their geographic ranges contained high proportions of degraded forest than if their ranges contained lower proportions of forest cover but of high quality. Our work suggests that biodiversity conservation policies to preserve forest integrity are now urgently required alongside ongoing efforts to halt deforestation in the hyperdiverse humid tropics.
Conservation efforts to date have largely failed to arrest the global biodiversity crisis 1 . Tropical rainforests, the most biodiverse terrestrial ecosystems on Earth 2 , face a particularly acute extinction crisis given the accelerating rates of rainforest conversion and degradation 3,4 . A potential solution to avert further losses of tropical rainforest species may be to preserve the remaining structurally intact native forests free from major human pressures (hereafter 'intact forests'), given their high biodiversity conservation value [5][6][7] . Less than half of the global tropical rainforest estate remains in its intact, native state 8,9 , and intact forests have long been sidelined in international environmental policy, with the maintenance and restoration of forest cover taking precedence over the preservation of forest intactness (or integrity) [10][11][12][13] . However, Article https://doi.org/10.1038/s41559-022-01915-8

Results
We found that high-integrity tropical rainforests were associated with significantly lower odds of species extinction risk and declining population trends across rainforest-obligate and rainforest-associated vertebrate groups, compared with forest cover alone ( Fig. 1; 95% confidence intervals (CIs) of odds ratios of standardized coefficients did not overlap 1; Supplementary Table 2, false discovery rate (FDR)-adjusted P < 0.05). For example, among rainforest-obligate mammals, greater high-integrity forest area within species ranges was associated with less than half the odds of being threatened (odds ratio: 0.40, 95% CI: 0.33-0.48) relative to greater area of forest cover alone (odds ratio: 1.32, 95% CI: 1.00-1.77). We also found that structurally intact forests tended to be associated with lower odds of species extinction risk and declining populations than forest cover alone. This pattern was stronger in some groups (for example, rainforest-obligate and rainforest-associated birds being threatened and amphibians having declining populations; Fig. 1) than in others (for example, rainforest-associated reptiles having declining populations). However, structurally intact forests tended to be associated with higher odds of species extinction risk and declining populations than were high-integrity forests, with this pattern again stronger in some groups (for example, rainforest-obligate and rainforest-associated mammals, reptiles and amphibians being threatened; Fig. 1) than in others (for example, rainforest-obligate birds having a declining population). The strength of high-integrity forests in being associated with low extinction risk and declining populations was similar for rainforest-obligate and rainforest-associated species (95% CIs overlapped each other; Fig. 1). We performed further tests of whether the observed association of high-integrity forests with low species extinction risk and declining population trends was an artefact of human pressures alone (that is, the FSII being dominated by the HFP) or the cumulative effect of both structural condition and human pressures. These analyses show that intact structural condition and low pressures (that is, forest integrity) tended to be cumulatively associated with lower odds of species extinction risk and declining population trends than either structural condition or human pressures considered individually (Extended Data Fig. 1 Table 3).

and Supplementary
Species tended to face higher probabilities of extinction risk and declining populations if their ranges contained high proportions of forest cover but forest of low structural condition and integrity (large extents of degraded forest) than if their ranges contained lower proportions of forest cover of high condition and integrity (Fig. 2). Evidence for this finding is the strong positive statistical interaction (95% CIs did not overlap zero and FDR-adjusted P < 0.05; Supplementary Tables 4 and 5) in 18 out of 32 models testing for two-way interactions between forest cover and condition and between forest cover and integrity on both response variables in each taxonomic group. A further nine such interactions were positive albeit statistically non-significant (95% CIs overlapped zero and FDR-adjusted P > 0.05). These patterns were also consistent for two-way interactions between forest condition and integrity on both response variables (Extended Data Fig. 2 and Supplementary Table 6). However, inconsistent with these general trends, rainforest-obligate birds, reptiles and amphibians had lower probabilities of declining populations when larger proportions of high-integrity forest cover remained within species ranges ( Fig. 2 and Extended Data Fig. 2), as indicated by negative two-way interactions between forest cover and condition, forest cover and integrity and forest condition and integrity in 6 out of 48 models (Supplementary Tables 4-6).
The buffering effect of forest integrity against species extinction risk was observed in every biogeographic realm ( Fig. 3 and Supplementary Table 7). After statistically controlling for the effects of forest cover (among species with average area of forest cover within their ranges), the probability of rainforest-obligate vertebrates being threatened decreased significantly with increasing forest integrity, compared with baseline estimates for each realm (without consideration of either forest cover or integrity; Fig. 3). Similarly, the probability not all forest cover is equal. The degree of anthropogenic degradation can diminish forest integrity 9,14 , which in turn may have severe adverse effects on biodiversity 6,15 . Recognizing this, the Convention on Biological Diversity (CBD) has included in their draft post-2020 Global Biodiversity Framework (GBF) a goal to enhance the integrity of native ecosystems 16 . The aspiration behind this goal is to achieve greater conservation success than occurred under the Aichi Biodiversity Targets 14 . Yet, there remains a lack of evidence on whether intact tropical rainforests have the potential to buffer species against extinction, when directly compared with forest cover.
Recent advances in remote sensing have facilitated the development of two fine-scale indices of tropical rainforest quality 8,9 , which now provide the ability to quantify the association between structurally intact forests under low human pressures and measures of biodiversity. The Structural Condition Index (SCI), a consistent 30 m resolution measure of forest condition across the global humid tropics, enables identification of taller, older, more structurally complex, closed-canopy rainforests (hereafter, 'structurally intact forests') 8,9 . Structurally intact forests may deteriorate with anthropogenic pressures (for example, settlements, roads, fire, selective logging and hunting), and the adverse impacts of such pressures on biodiversity may surpass those of deforestation alone 6 . To capture such pressures, the Forest Structural Integrity Index (FSII) 8,9 combines the SCI with the Human Footprint (HFP) 17 to distinguish rainforests of intact structural condition and minimal human modification (hereafter, 'high-integrity forests').
Here, we quantify the association between species extinction risk and population trend and the amount of high-integrity forest remaining within the geographic ranges of humid tropical vertebrates, relative to the amount of structurally intact forest and forest cover alone (without consideration of either structural condition or integrity). We used the IUCN Red List category of extinction risk and overall population trend 18 for 16,396 mammal, bird, reptile and amphibian species whose ranges overlap the tropical and subtropical moist broadleaf biome (also known as the tropical rainforest or humid tropical biome) 19 . We classified species as either rainforest-obligate (dependent on rainforests) or rainforest-associated (use rainforests as well as other habitat types) on the basis of the extent of range overlap with the tropical rainforest biome and association with tropical forest habitats 2 , expecting the potential effects of forest integrity to be stronger for rainforest-obligate species than for associated species. Within species humid tropical ranges (the sampling unit for this study), we used the SCI and FSII datasets to calculate the area (km 2 ) of structurally intact and high-integrity forests, relative to the area of structurally degraded and low-integrity forests. We also pooled all SCI values representing forest to calculate the total area of forest cover within species ranges, relative to non-forest area.
We used a generalized linear modelling framework that accounts for the phylogenetic non-independence of species 20 to test whether greater high-integrity forest area within species ranges is associated with lower odds of species: (1) being threatened and (2) having a declining population trend, relative to the area of structurally intact forests and forest cover alone. Further, we considered two-way statistical interactions between forest cover and each of the structural condition and integrity variables to test whether species are likely to be at higher risk of extinction and population decline if their ranges contain high proportions of forest cover but of low quality (large extents of degraded forest) than if their ranges contain lower proportions of forest cover of high quality. Finally, we examined variation in species extinction risk and population trend as a function of forest integrity across the four biogeographic realms within the tropical rainforest biome to test whether the potential buffering effect of high-integrity forests on biodiversity is consistent across major terrestrial regions. With this research, we aim to inform global environmental agreements such as the CBD post-2020 GBF on the potentially crucial need to preserve the integrity of remaining tropical rainforests worldwide.
Article https://doi.org/10.1038/s41559-022-01915-8 of rainforest-obligate vertebrates having declining populations also decreased significantly with increasing forest integrity in some realms, compared with baseline estimates. However, there was considerable variation in this trend among mammals and birds in the Indomalayan and Neotropical realms and reptiles in the Afrotropics, such that the probability of population decline with increasing forest integrity was not significantly different from baseline estimates (95% CIs overlapped; Fig. 3). Indomalayan vertebrates faced the highest overall risk (with the exception of amphibians in Australasia and reptiles in the Afrotropics), congruent with prior findings for non-vertebrates in this region 21 . These findings for rainforest-obligate vertebrates were mirrored in species associated with tropical rainforests (Extended Data Fig. 3).
Our conclusions were robust to a range of plausible error in mapping canopy cover and height with the SCI and FSII datasets, performed by adjusting the SCI classification boundaries for canopy cover and height up and down by 20% to simulate potential errors (Methods; Supplementary Table 8 and Extended Data Fig. 4) and thereafter propagating such errors to statistical models (Extended Data Fig. 5 and Supplementary Tables 9 and 10). Our findings also remained consistent when we pooled the area of moderate structural condition and integrity forests with high structural condition and integrity forests (Methods; Extended Data Fig. 6 and Supplementary Table 11), suggesting that forests of moderate structural condition and integrity can support biodiversity conservation. We conducted an alternative statistical analysis based on model selection with Akaike's Information Criterion (AIC) to further test the robustness of our conclusions. Here, we fit a candidate set of four univariate models (forest integrity, structural condition, human pressure and forest cover parameterized individually) to each response variable. This analytical approach also showed substantial support for forest integrity as the most important variable in predicting low likelihood of species extinction risk and declining population trends (lowest AIC score and highest model weight; Supplementary Table 12).
Our results were consistent across degrees of threat; critically endangered, endangered and vulnerable species showed positive effects of forest integrity (Extended Data Fig. 7 and Supplementary  Tables 13 and 14). We also performed sensitivity analyses to exclude species designated as threatened because of decline in habitat extent and/or quality (criterion A of the IUCN Red List 18 ) and because of restricted and fragmented geographic ranges (criterion B 18 ). The exclusion of species under criteria A and B avoids potential circularity between comparative analyses of extinction risk and the IUCN criteria used to assess extinction risk 22 . However, the association of high-integrity tropical rainforests with lower odds of species extinction risk (relative to forest cover alone) remained evident even after excluding species threatened under criteria A and B (Extended Data Fig. 8 and Supplementary . We note that we did not consider declining population trends in this analysis of potential circularity because the IUCN Red List criteria are not used for determining overall population trends. The humid tropical ranges of many species in our dataset overlap, leading to a potential lack of spatial independence when extracting and analysing forest structural condition and integrity data from the same regions across multiple species. Therefore, we tested model residuals for spatial autocorrelation as a function of distance between centroids of species humid tropical ranges using Moran's I and spline correlogram tests 23 . However, we found no evidence for spatial autocorrelation (Moran's I < 0.1 and P > 0.05; Supplementary Figs. 1-6).

Discussion
Reducing deforestation is a central pillar of global biodiversity conservation efforts and indeed represents a critical first step in averting species losses [10][11][12][13] . However, we demonstrate that high-integrity forests are associated with considerably lower risk of humid tropical vertebrate species extinctions and population declines, when directly compared with forest cover. We show that high-integrity forests are important not only for rainforest-obligate species but also for rainforest-associated species that may use these ecosystems as refugia or on a seasonal basis. Moreover, high-integrity forests were associated with lower odds of species extinction risk and declining population trends than either structural condition or human pressures considered individually, suggesting that intact structure alone may be insufficient to conserve rainforest biodiversity without also limiting human pressures within forests. Consequently, preserving the last-remaining structurally intact tropical rainforests and limiting human pressures within these ecosystems may prevent more species from becoming threatened and undergoing population declines over time. Furthermore, high-integrity forests may be more resilient to large-scale environmental perturbations (for example, climate change) than are degraded forests 7,24 . Thus, the buffering effect of high-integrity forests on biodiversity may potentially increase over time because forests that are already degraded will probably experience intensifying pressures exacerbated by climate change 25 . Overall, our work suggests that biodiversity conservation policies aimed at preserving forest structure and maintaining low human pressures

Rainforest obligate
Declining population

Rainforest associated
Forest integrity (FSII) Structural condition (SCI) Forest cover

Fig. 1 | The relative importance of forest integrity, structural condition and forest cover on the odds of mammals, birds, reptiles and amphibians being threatened and having declining population trends.
Forest integrity tended to be associated with a beneficial effect on biodiversity (lower odds of species being threatened and having declining population trends), relative to forest cover. For sample sizes, see Supplementary Table 1a. Point estimates represent median standardized odds ratios of species being threatened (circles) or having a declining population (squares), generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions (Supplementary Table 2). The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% CIs generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and rainforest-associated species for each response variable. Illustration credits: S. Traver, F. Sayol, B. Szabo and J. C. Arenas-Monroy.
Article https://doi.org/10.1038/s41559-022-01915-8 within structurally intact forests are now an urgent priority alongside current efforts to halt deforestation across the humid tropics. The higher odds of species extinction risk and declining populations associated with forest cover may be surprising given existing knowledge on the benefits of forest cover on biodiversity, typically when forest cover is considered in standalone analyses or compared with land-uses often inimical to biodiversity such as agriculture and development 22,26 . However, our work is an assessment of the importance of forest cover relative to remotely sensed measures of forest structural condition and integrity (variables that represent the quality of forest cover) on species extinction risk and population trends across the global humid tropics. We report odds ratios of standardized partial regression coefficients, which represent unbiased estimates of the effects of forest cover alone on each response variable, relative to structurally intact and high-integrity forests (controlling statistically for the effects of forest condition and integrity) 27,28 . Thus, our results probably reflect how various forms of structural degradation (for example, selective logging) and human pressures (for example, hunting) within forest cover may adversely affect biodiversity, compared with structurally intact forests with low levels of such pressures. We note that when forest cover was considered in univariate models for each response variable (not analysed relative to forest condition and integrity), it tended to be associated with low odds of species extinction risk and declining populations, as would be expected (Supplementary Table 18). Similarly, when human pressure was considered in univariate models, it was always associated with high odds of species extinction risk and declining populations as expected (Supplementary Table 18).
Large, well-connected forest landscapes are essential for biodiversity conservation, especially in an era of climate change 21,22,29 . We show a higher likelihood of extinction risk and declining populations when large extents of forest cover within species ranges were degraded, emphasizing the importance of minimizing human disturbances in remaining intact tropical rainforest landscapes 22 . In contrast, we found a lower likelihood of extinction and declining populations when species ranges contained lower proportions of forest cover but forest of high integrity, adding to the growing evidence that remnant high-integrity forests can play an important supporting role for biodiversity by providing refugia or habitat for numerous species [30][31][32] . Nevertheless, remnant high-integrity forests face a higher likelihood of loss compared with larger forested extents because of the severe land-use pressures around them and improved access for resource extraction 33 . Moreover, sensitivity to isolation in remnant forests may be a likely explanation for the higher probability of declining populations even in high-integrity remnant forests for rainforest-obligate birds, reptiles and amphibians 34 , potentially signalling the presence of an extinction debt for these vertebrate groups in fragmented landscapes 22 . Thus, proactively prioritizing the protection of remnant high-integrity forests from loss while simultaneously setting targets for restoring degraded forests are both important to limit the loss of already threatened and declining species.
Human influence is not limited to tropical rainforests but extends over much of Earth's land surface 17 . Future research should leverage global forest integrity 35 and ecosystem intactness 36 datasets to quantify the importance of ecosystem integrity for terrestrial biodiversity in all . Species tended to be at higher risk of being threatened and having declining populations when their ranges contained high proportions of forest cover but low structural condition and integrity (that is, large extents of degraded forest) than when their ranges contained lower proportions of forest cover but high condition and integrity. Median predicted probabilities were generated from 100 phylogenetic logistic regressions. See Supplementary Table 1a and Tables 4 and 5 for sample sizes and model estimates, respectively. Illustration credits: S. Traver, F. Sayol, B. Szabo and J. C. Arenas-Monroy.
Article https://doi.org/10.1038/s41559-022-01915-8 forest as well as non-forest biomes. In addition, datasets on potential forest structural complexity (the theoretical potential native vegetation in a region in the absence of human disturbance) 37 , when related to actual structural complexity (for example, the SCI data used here), can help to monitor the effectiveness of forest management and restoration efforts.
The SCI and FSII datasets are static in time (centred on 2013) 8 , while our biodiversity datasets are from 2019-20 (Methods), allowing for a time lag between rainforest degradation and species responses to potentially be reflected in IUCN assessments. Nevertheless, our analyses represent a space-for-time substitution. Albeit widely used in ecological studies given the paucity of long-term datasets, The bar plots show the baseline probabilities in each realm estimated without consideration of either forest cover or integrity. The adjacent line plots show the probability of being threatened and having a declining population with increasing forest integrity after statistically controlling for the effects of forest cover (among species with average area of forest cover within their ranges). Data points (1, threatened/declining; 0, not threatened/not declining) are vertically and horizontally jittered to reduce overlap. The bars and lines represent median predicted probabilities from 100 phylogenetic logistic regressions. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Error bars and the shaded areas of the lines represent median 95% CIs generated with 2,000 parametric bootstraps in each regression. These results were mirrored in rainforestassociated vertebrates (Extended Data Fig. 3). See Supplementary Table 1b and  Table 7  space-for-time substitution may underestimate the effects of tropical forest disturbances such as selective logging on biodiversity 38 . Indeed, preliminary analysis of the effect of change in SCI from 2012 to 2018 on species extinction risk and population trends suggests that degradation of tropical rainforest structural condition may pose a similar, and sometimes greater, threat to biodiversity than the outright loss of forest cover (Methods; Extended Data Fig. 9 and Supplementary  Table 19). Therefore, investigating the effects of change in forest structural condition and integrity on shifts in species extinction risk and population trends over longer timeframes is an important future research direction 39,40 .
Invertebrates and vascular plants comprise the greatest share of tropical biodiversity in terms of species diversity and biomass 41,42 . Comparable range map and habitat preference data for non-vertebrate groups remain unavailable, but future work with alternative datasets and statistical approaches can help to quantify the importance of forest integrity for such diverse yet understudied taxonomic groups 21 . Further, investigating links between remotely sensed indices of forest structural condition and integrity and species traits may offer insights into the potential role of intact forests as a buffer for functional species groups particularly susceptible to environmental change 43,44 .
Our findings demonstrate a clear need for the targeted preservation of the last-remaining high-integrity forests across the global humid tropics. A unique opportunity to advance biodiversity conservation is at hand, given that 86% of high-integrity tropical rainforests remain unprotected 9 . Focusing environmental policies and management actions on preserving their integrity alongside ongoing efforts to halt deforestation will probably improve conservation outcomes by preventing more tropical rainforest species from becoming threatened or undergoing population declines over time 16 . On the basis of our findings, we argue that the single most important policy action nations can take to prevent catastrophic biodiversity loss in tropical rainforests is to commit to a target of "net gain in area, connectivity and integrity" of these hyperdiverse ecosystems 16 . Proactive targets to preserve and restore forest integrity are an urgent priority to 'bend the curve' on species loss 1 and ensure that nations stand a chance "to put biodiversity on a path to recovery for benefit of planet and people" by 2030 16,45,46 .

Geographic range maps
We conducted our analyses across the full extent of the tropical and subtropical moist broadleaf forest biome, which encompasses the present-day distribution of tropical rainforests around the Equator and primarily between the Tropics of Cancer and Capricorn 19 . These forests largely span the latitudes between 23.5° N and 23.5° S but extend into the subtropics in some areas (Extended Data Fig. 4). Despite covering a mere 14% of Earth's terrestrial area 19 , these forests are home to over half of the world's vertebrate species 2 , such that the continued loss and degradation of these imperilled ecosystems is likely to result in a disproportionately high number of extinctions. We followed the protocols in ref. 2 to obtain the latest established geographic range maps for all species of mammals 18 , birds 47 , reptiles 48 and amphibians 18,49 . The original datasets contained range maps for 5,566 mammals, 11,125 birds, 10,064 reptiles and 6,684 amphibians and include ranges for species that are extinct as well as range extents based on uncertain data.
We filtered all geographic range map datasets with three successive IUCN Red List of Threatened Species spatial attributes to remove extinct species and records on the basis of uncertain data. First, we retained only species known to be "extant", while discarding polygons representing parts of a species range where it was reported to be "possibly extant", "possibly extinct", "extinct" and "presence uncertain". Second, we filtered this list of extant species to retain only those that are "native" and "reintroduced" while discarding polygons representing parts of a species range where it was reported to be "introduced", "vagrant", "origin uncertain" and "assisted colonization". Third, we filtered the list of species from the second step above to retain only "resident" and "non-breeding" parts of the range for mammals (the only ones remaining for mammals after the first two filters above). For birds, we retained "resident", "breeding", "non-breeding" and "passage" parts of the range while discarding "seasonal occurrence uncertain". For amphibians, we retained "resident" parts of the range, which was the only one remaining after the first two filters above. The final list of amphibians from the IUCN Red List after this third filter included 6,607 species. However, this list of amphibians from the IUCN does not comprise all known species. Therefore, we included range maps for 659 additional amphibian species from ref. 49 , after cross-verification to omit synonyms and extinct species. Because we obtained the reptile database from a source other than the IUCN Red List 48 , we were unable to perform the same suite of filters on reptiles. However, our analyses showed that ten species from this list are now regarded as extinct. Therefore, we discarded these ten species. After performing these filters, our list of species for subsequent analyses included 5,529 mammals, 10,935 birds, 10,054 reptiles and 7,264 amphibians, for a total of 33,782 species of extant terrestrial vertebrates worldwide.
We projected all geographic range maps to the World Mollweide projection before analyses and used Python code implemented with the ArcPy module in ArcGIS Pro 2.5.0 to perform a union of the range map of each species with the map of the tropical rainforest biome. This procedure allowed us to distinguish parts of the global range of species that overlap the tropical rainforest biome, should there be such overlap for a given species. We did not set a lower bound for range overlap with tropical rainforests because such a threshold would be arbitrary and also exclude species that marginally occur in tropical rainforests but for which these ecosystems nevertheless represent important habitats (for example, some species of wintering migratory birds) 2 . Thereafter, we used species-level attributes from the IUCN Red List of Threatened Species to obtain data on the major habitats in which each species occurs to limit some forms of commission or false-positive errors that may occur with range maps. Specifically, these errors include species whose ranges may overlap with the tropical rainforest biome but do not actually use the forests within that biome 2 . For species having range overlap with the tropical rainforest biome, we retained only species reported to occur in tropical forest habitat types listed in the IUCN Habitats Classification Scheme 50 . We merged this list of species reported to occur in tropical forest habitats with the list of species whose ranges overlap the tropical rainforest biome to retain 3,327 mammals, 7,704 birds, 3,828 reptiles and 5,298 amphibians, for a total of 20,157 species 2 . We discarded additional species from this dataset on the basis of matching species names with those in the respective phylogenetic trees (for the final list of species in this study, see section on Statistical analyses). We note that the habitat associations of ~30% of reptile species whose ranges overlap tropical forests remain unknown because reptiles are one of the most understudied terrestrial vertebrate groups 2,51 . We took several steps to limit potential geographic bias from this issue. Specifically, we matched reptile species in geographic range maps with the best-available reptile phylogenetic trees as with the other taxonomic groups, analysed each taxonomic group independently and explicitly estimated variation in species extinction risk across biogeographic realms (see section on Statistical analyses).

Definition of tropical rainforest-obligate species
We defined dependency on tropical rainforests following the criteria established by ref. 2 . We considered a species to be rainforest-obligate if (1) 80-100% of its global range overlapped with the tropical rainforest biome and (2) it was near-exclusively reported from the tropical rainforest habitat types listed in the IUCN Habitats Classification Scheme 50 . We did not exclude wetlands, rocky and cave habitats from this second criterion, making the reasonable assumption that for species with >80% range overlap with the tropical rainforest biome and nearly exclusively Article https://doi.org/10.1038/s41559-022-01915-8 associated with rainforest habitats, these three other habitat types are likely to be within tropical rainforests (for example, bats that roost in caves within rainforest habitats).

Tropical rainforest structural condition and integrity indices
We used two indices of tropical rainforest quality in our analyses-the SCI and the FSII 8,9 . The SCI is a fine-scale (30 m resolution) raster derived from three datasets: global tree canopy cover in 2010 52 , time since forest loss (between 2000 and 2017) 52 and canopy height in 2012 53 . It identifies locations of taller, older, more structurally complex, closed-canopy rainforests across the global humid tropics. The reference year is 2013, with canopy cover from 2010, forest loss expressed as year of loss before 2018 and canopy height for 2012. The SCI ranges from 1 to 18, encompassing short, open-canopy recently disturbed forests to tall closed-canopy stands 8,9 . The lowest SCI value delineates stands <5 m tall, disturbed since 2012 or with canopy cover <25%. The highest SCI value represents tall, closed-canopy stands undisturbed since 2000. To ensure that our analysis deals with the structure of stands that meet the criteria of being forest and is not confounded with recent forest loss, we categorized the lowest SCI values of 1 as non-forest and completely removed these values from analyses of structural condition and integrity (see section on Predictor variables). The FSII is derived by overlaying the HFP, a 1 km resolution measure of the cumulative, in-situ pressures humans exert on natural areas across terrestrial Earth 17 , on the SCI. The HFP ranges from 0 to 50, representing a gradient of increasing human pressure 17 . The original 1993 HFP 54 was updated to 2009 55 and more recently to 2013 17 . The FSII ranges from 0.1 to 18 with the higher values representing rainforests high in structural complexity and low in human pressure. For comprehensive details on the SCI and FSII datasets, see refs. 8,9 . As with the range maps, we projected the SCI, FSII and HFP raster datasets to the World Mollweide projection before analyses. Given the differing resolutions of the SCI and FSII rasters (30 m and 1 km, respectively), we first made them comparable by resampling both to 1 km resolution (identical to the HFP) in ArcGIS 10.7. After resampling, the SCI raster comprised 1 km resolution pixels of values ranging from 1 to 18. We also converted the continuous pixel values of the FSII dataset to the nearest integer, such that the resampled FSII raster comprised 1 km resolution pixels of values ranging from 0 to 18. A relatively fine resolution such as used here facilitates efficient identification of forest cover and structurally intact and high-integrity forests within species ranges and is recommended when the objective is to distinguish the effects of broad habitat categories on biodiversity 56,57 .
We then used Python code implemented with the ArcPy module in ArcGIS Pro 2.5.0 to calculate the area (km 2 ) of each of the 18 values of the SCI, 19 values of the FSII and 51 values of the HFP rasters within the humid tropical range of each species. Following the criteria established by ref. 9 , we categorized and summed the area of SCI pixel values ranging from 2 to 5 (>25% canopy cover and >5 m canopy height) as low SCI or structurally degraded forest, values from 6 to 13 as moderate SCI forest and values from 14 to 18 (>75% canopy cover and >15 m canopy height) as high SCI or structurally intact forest. We followed a similar procedure to categorize and sum the area of FSII values from 1 to 5, 6 to 13 and 14 to 18 as low-, moderate-and high-integrity forest, respectively. These high-integrity forests represent rainforests of intact structural condition and low human pressures, specifically HFP values ≤4 (ref. 9 ). Similarly, we followed the criteria in refs. 8,9 to categorize and sum the area of HFP pixel values ranging from 0 to 3 as low human footprint or pressure, 4 to 15 as moderate pressure and 16 to 50 as high pressure regions within species ranges.

Simulating plausible error in structural condition and integrity indices
We expected errors of 10-30% as plausible in maps of canopy cover and height derived from multispectral satellite imagery [58][59][60] . Therefore, we adjusted SCI classification boundaries for canopy cover and height up and down by 20% to simulate a plausible range of scenarios and enable testing of the sensitivity of statistical models to these potential errors (Extended Data Fig. 4). Adjusting the canopy cover and height classification boundaries downward had the effect of increasing the number of pixels classified as high SCI (values [14][15][16][17][18], effectively simulating overestimates of canopy cover and height (Supplementary Table 8). Adjusting the cover and height classification boundaries upward had the opposite effect. As with the original SCI, we overlaid the HFP on both reclassified SCI rasters to generate FSII rasters incorporating the assumed ±20% error in mapping canopy cover and height. We then performed our statistical analyses with these reclassified SCI and FSII datasets, as detailed with the original datasets, to examine whether our model estimates remained robust to the simulated range of potential error.

Forest cover loss and change in structural condition
We created a raster for forest cover loss by identifying pixels lost between 2012 and 2018, which were initially classified as forest in 2012 on the basis of canopy cover and height thresholds >25% and >5 m, respectively. We also created a change in structural condition dataset for the same timeframe by subtracting the baseline 2012 SCI pixel values from the 2018 values. Given this short window over which we have temporal data, little change in structural condition was observed (<1% of pixels show change in SCI), which may affect statistical power. Efforts are ongoing to update the SCI and FSII datasets to allow change analyses over a longer timeframe (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017), which would allow a stronger analysis of the association between change in forest quality and recent genuine changes in extinction risk, for example, ref. 39 . We note that we did not use a change in forest integrity dataset because HFP layers matching the 2012-2018 timeframe are unavailable.

Predictor variables
We calculated the relative difference between the area of high and low SCI forest within the humid tropical range of species j as , where H sci j and L sci j are the areas of high and low SCI forest, respectively, and C j is the area of humid tropical forest cover within the range of species j. Likewise, we calculated the relative difference between the area of high and low FSII forest as The calculated values range between −1 and +1 and represent the relative percentage difference between the areas under high and low SCI and FSII forests within the humid tropical range of a species. Thus, a value of −1 indicates that ~100% of the humid tropical range of a species is encompassed by low SCI or low FSII forest, whereas a value of +1 means ~100% of the humid tropical range of a species is encompassed by high SCI or high FSII forest (a gradient of increasing SCI/FSII). For the analysis of HFP, we similarly calculated the relative difference between the areas of high and low HFP within the humid tropical range of species j as d hfp Here, a value of −1 means ~100% of the humid tropical range of a species is encompassed by low HFP areas, whereas a value of +1 signifies that ~100% of the humid tropical range of a species is encompassed by high HFP areas (a gradient of increasing HFP). For the analyses including forests of moderate structural condition and integrity with those of high structural condition and integrity, we recalculated these relative difference values as , where M sci j is the area of moderate SCI forest within the range of species j that was summed with H sci j and similarly for M fsii j . We used the lowest SCI value of 1 to identify stands <5 m tall, disturbed since 2012 or with canopy cover <25%, which are considered highly disturbed, and categorized the area of this pixel value Article https://doi.org/10.1038/s41559-022-01915-8 as non-forest 9 . We categorized and summed the remaining SCI values from 2 to 18 as forest. We then calculated the relative difference between the area of forest cover and non-forest within the humid tropical range of species j as d C where C j and NC j are the area of forest cover and non-forest within the range of species j and the denominator sums the humid tropical range area of species j. Similar to the SCI and FSII relative difference values, these calculated values of forest cover also range between −1 (signifying 100% of the humid tropical range of a species consists of non-forest) and +1 (signifying that 100% of the humid tropical range of a species is forested). We thereby brought all predictor variables in this study (forest cover, condition, integrity and human footprint) to a consistent scale for further analyses. However, for the analysis of change, we used proportion loss in forest cover and proportion change in structural condition as predictor variables because of the sparse data on change.

Statistical analyses
The response variables in this study are binary-threatened/ non-threatened and declining population/not declining in population. To achieve this binary classification, we defined species in the IUCN 'critically endangered', 'endangered' and 'vulnerable' categories as threatened and species in the 'near threatened' and 'least concern' categories as non-threatened while discarding species in the 'data deficient' category (see Extended Data Fig. 7 for analyses of alternative threatened definitions). For the IUCN population trend data, we defined species in the 'decreasing' category as declining in population and species in the 'increasing' and 'stable' categories as not declining in population while discarding species in the 'unknown' category 22 .
We used a generalized linear modelling framework, specifically logistic regression, for statistical inference. Our primary units of analyses-species-cannot be considered as independent because of the variable degree of evolutionary relatedness between the species in each taxonomic group. To account for the potential effect of evolutionary dependence, we first obtained phylogenetic trees for mammals 61 , birds 62 , reptiles 63 and amphibians 64 and matched the species lists from the previous steps to discard species not in the respective phylogenetic trees. Our list of species after this step comprised 3,217 mammals, 6,674 birds, 3,735 reptiles and 5,069 amphibians, for a total of 18,695 species of vertebrates. We further discarded 2,299 data-deficient species for a final total of 16,396 species in the analyses of threatened status. We also discarded 5,842 species of unknown population trend for a final total of 12,853 species in the analyses of declining population (Supplementary Table 1). For each taxonomic group, we partitioned species into rainforest-obligate and rainforest-associated categories. Next, we randomly sampled 100 trees out of 10,000 available full phylogenetic trees for each taxonomic group, as recommended by ref. 62 , to construct covariance matrices enumerating the proportion of the evolutionary path shared between each pair of species. We used these covariance matrices in phylogenetic logistic regression models to generate inferences corrected for phylogenetic signal 20 .
We parameterized identical models for rainforest-obligate and rainforest-associated species in each taxonomic group to estimate the relative importance of high-integrity forests within species ranges in being associated with reduced odds of species: (1) being threatened and (2) having a declining population, compared with structurally intact forests and forest cover alone. Before analyses, we standardized each predictor variable (forest cover, condition and integrity) to have a mean of 0 and a standard deviation of 1 (z-transformation). We examined the effects of the three predictor variables on the respective response variable (threatened status or declining population) by parameterizing them as additive effects in multiple phylogenetic logistic regression models and used the standardized partial coefficient of each predictor variable as a measure of its effect on the response variable 22,27 . In this form of multiple logistic regression, the exponentiated standardized partial coefficient of a given predictor variable represents the odds of a 1-unit increase in that variable on the response, controlling for the effects of the other predictor variables by statistically holding them at their average values 27 . Given the correlated nature of the predictor variables (Supplementary Table 20), standardized partial regression coefficients can provide unbiased estimates of the relative importance of forest cover, condition and integrity on the odds of species being threatened or having a declining population 28 . We estimated 95% CIs for the estimated standardized coefficients in each regression with 2,000 parametric bootstraps as recommended by ref. 20 and made inferences based on the median of 100 regressions, each regression being performed with one phylogenetic tree randomly drawn from 10,000 available trees 62 .
We then considered two-way interactions between forest cover × condition and forest cover × integrity with phylogenetic logistic regression models to test whether the effects of forest integrity and condition depend on the amount of forest cover within species ranges. We also tested whether the effect of forest integrity depends on the amount of high structural condition forest within species ranges by considering interactions between forest condition × integrity. A positive coefficient for the interaction term (for example, forest cover × condition) would suggest that the effect of forest condition on the probability of being threatened or having declining populations was stronger when species ranges contained low proportions of forest cover in intact structural condition, as opposed to when species ranges contained high proportions of forest cover in degraded structural condition. In contrast, a negative coefficient for the interaction term would indicate that the effect of forest condition was stronger when species ranges contained high proportions of forest cover in intact condition, as opposed to low proportions of forest cover, irrespective of its structural condition. Interaction models were otherwise parameterized in an identical manner to the additive models described above.
To examine variation in extinction risk across biogeographic realms for each taxonomic group, we first fit a model with realm as a categorical predictor variable. We included rainforest dependency as an additive effect in this model to estimate variation in extinction risk between rainforest-obligate and rainforest-associated species. This model enabled estimation of baseline probabilities of species being threatened or having a declining population in each realm. Next, we parameterized a model testing for an interaction between forest integrity and realm, with a positive interaction coefficient suggesting lower extinction probability with increasing forest integrity and the relative strength of the interaction indicating variability in this probability between realms. Given that the amount of forest cover within the humid tropical range of a species can also influence and have a confounding effect on the integrity × realm interaction, we included a second interaction between forest cover and realm in this model, thereby statistically controlling for the effects of forest cover. We did not consider forest structural condition in this analysis because forest integrity was mostly of greater importance in predicting extinction risk than was forest condition (Fig. 1).
We implemented all phylogenetic logistic regression analyses via the package 'phylolm' 65 in the R (v.4.0.3) statistical programming language 66 . To limit bias in maximum likelihood estimates of logistic regression coefficients, we used the maximum penalized likelihood method with Firth's correction implemented in the phyloglm function via the parameter logistic_MPLE 20,65 . We conducted our analyses across thousands of species with three predictor variables in the case of additive models and two predictor variables for interaction models, which risks inflating type 1 error rate. Therefore, we used a procedure adjusted for FDR 67 which corrects for multiple comparisons in comparative extinction risk modelling. We calculated FDR-adjusted P-values with the p.adjust function in R 68 . We obtained centroids of species humid tropical ranges in ArcGIS Pro 2.5.0 and thereafter used the packages 'spdep' 69 and 'ncf' 70 in R to calculate geographic distance between species humid tropical range centroids and perform spatial autocorrelation analyses of model residuals.

Influence of phylogenetic correlation
In phylogenetic logistic regression, the parameter α measures the strength of the phylogenetic correlation. When α = 1, evolution is approximately by Brownian motion on a given phylogeny, with α > 1 indicating lower phylogenetic correlations among species 20 . In most cases across all taxonomic groups and models, the estimated phylogenetic signal α was close to zero (Supplementary Table 21), suggesting that the predictor variables included in the models induced phylogenetic signal in the residuals.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All datasets used in this paper are openly available via the citations identified in the Methods. Processed spreadsheets can be accessed at Zenodo https://doi.org/10.5281/zenodo.7036360.

Code availability
Python and R code to replicate geospatial and statistical analyses can also be accessed through the same Zenodo repository https://doi. org/10.5281/zenodo.7036360. Additional Python code to process species range maps before raster overlay and tabulation of area can be accessed at https://doi.org/10.5281/zenodo.5525586.
Article https://doi.org/10.1038/s41559-022-01915-8 Extended Data Fig. 1 | The relative importance of forest integrity, structural condition, the human footprint or pressure, and forest cover on the odds of mammals, birds, reptiles, and amphibians being threatened and having declining population trends (for sample sizes, see Supplementary Table  1a). Structural condition and human pressures considered together (that is, FSII) tended to be associated with lower odds of species extinction risk and declining population trends than either structural condition or human pressures individually. Point estimates represent median standardized odds ratios of species being threatened (circles) or having a declining population (squares), generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions (Supplementary Table 3). The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and rainforest-associated species for each response variable. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy. Fig. 2 | Predicted probabilities of tropical rainforest-obligate and associated mammal, bird, reptile, and amphibian species being threatened and having declining population trends as a function of two-way interactions between forest structural condition and integrity. Species tended to be at higher risk of being threatened and having declining populations when high proportions of forest cover within their ranges were structurally intact but of low integrity (that is, under high human pressure) than when their ranges contained low proportions of forest cover in high structural condition and low human pressure. Median predicted probabilities were generated from 100 phylogenetic logistic regressions. See Supplementary Tables 1a and 6 for sample sizes and model estimates, respectively. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy. Fig. 3 | Predicted probabilities of rainforest-associated mammals, birds, reptiles, and amphibians being threatened and having declining population trends across the four biogeographic realms within the tropical rainforest biome. The bar plots show the baseline probabilities in each realm estimated without consideration of either forest cover or integrity. The adjacent line plots show the probability of being threatened and having a declining population with increasing forest integrity after statistically controlling for the effects of forest cover (that is, among species with average area of forest cover within their ranges). Data points (1, threatened/declining; 0, not threatened/not declining) are vertically and horizontally jittered to reduce overlap. The bars and lines represent median predicted probabilities from 100 phylogenetic logistic regressions. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Error bars and the shaded areas of the lines represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. These results were mirrored in rainforest-obligate vertebrates (Fig.  3). See Supplementary Table 1b and Table 7 for sample sizes and model estimates, respectively. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy.

Extended Data Fig. 4 | Original and reclassified SCI and FSII datasets. (a)
The original SCI and FSII raster datasets from Hansen et al. 2019, 2020 and used in the main analyses presented here. The tropics lie between 23.5° N and 23.5° S latitudes (indicated by the dotted lines) but the tropical rainforest or humid tropical biome extends into the subtropics in some areas. (b) A reclassified SCI raster generated by simulating a + 20% error in canopy cover and height derived from multispectral satellite imagery (left). This +20% error reduced the number of pixels classified as high SCI (values [14][15][16][17][18], effectively simulating underestimates of canopy cover and height. (c) A reclassified SCI raster simulating a -20% error in canopy cover and height measurements (left). This -20% error increased the number of high SCI pixels, effectively simulating overestimates of canopy cover and height. See Supplementary Table 7 for original and reclassified thresholds of canopy cover and height. As with the original SCI, the Human Footprint was overlaid on both simulated SCI rasters to generate corresponding FSII rasters incorporating the assumed ±20% errors (b, c: right). All raster data were resampled from the original 30 m pixel resolution to 1 km (Methods). See Fig. 1 Fig. 5 | Propagating simulated errors in mapping canopy cover and height with the SCI and FSII datasets to statistical models. The relative importance of forest integrity, structural condition, and forest cover on the odds of mammals, birds, reptiles, and amphibians being threatened and having declining population trends. The underlying structural condition and integrity data for these analyses are a reclassified SCI raster generated by (a) simulating a + 20% error in canopy cover and height derived from multispectral satellite imagery. This +20% error reduced the number of pixels classified as high SCI (values [14][15][16][17][18], effectively simulating underestimates of canopy cover and height (Extended Data Fig. 4b) and (b) simulating a -20% error in canopy cover and height measurements. This -20% error increased the number of high SCI pixels, effectively simulating overestimates of canopy cover and height (Extended Data Fig. 4c). As with the original SCI, the Human Footprint was overlaid on both simulated SCI rasters to generate FSII rasters incorporating the assumed ±20% errors. Our overall conclusions remained robust to this simulated range of potential error in mapping canopy cover and height in the SCI and FSII datasets. Point estimates represent median standardized odds ratios of species being threatened (circles) or having a declining population (squares) generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions. The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and associated species for each response variable. See Supplementary Table 1a for sample sizes, Supplementary  Table 8  Article https://doi.org/10.1038/s41559-022-01915-8 Extended Data Fig. 6 | Pooling moderate structural condition and integrity forests with structurally intact and high-integrity forests. The relative importance of forest integrity, structural condition, and forest cover on the odds of mammals, birds, reptiles, and amphibians being threatened and having declining population trends. In the main text, we calculated the area (km 2 ) of structurally intact and high-integrity forests (SCI and FSII values 14-18), relative to the area of structurally degraded and low-integrity forests (SCI values 2-5 and FSII values 1-5) within species humid tropical ranges (Methods). Here, we conducted an additional analysis pooling the area of moderate structural condition and integrity forests (SCI and FSII values 6-13) with structurally intact and high-integrity forests and thereafter parameterizing an identical set of models as the main analyses. We were thus able to consider the entire gradient of forest quality when examining its effects on species extinction risk and declining populations. Our overall conclusions remained consistent, suggesting forests of moderate structural condition and integrity can have value for biodiversity conservation. Point estimates represent median standardized odds ratios of species being threatened (circles) or having a declining population (squares) generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions. The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and associated species for each response variable. See Supplementary Tables 1a and 11 for sample sizes and model estimates, respectively. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy. Fig. 7 | Alternative definitions of IUCN Threatened Status. The relative importance of forest integrity, structural condition, and forest cover on the odds of mammals, birds, reptiles, and amphibians being threatened. In the main analyses, we considered a species to be threatened if it was classified in any one of the IUCN Red List categories Critically Endangered (CR), Endangered (EN) or Vulnerable (VU). Here, we performed additional analyses considering a species as threatened only if it was classified as (a) CR and (b) either CR or EN. For all analyses, we classified species in the Near Threatened and Least Concern categories as non-threatened. This allowed us to maintain the binary classification (threatened/non-threatened) of the response variable, which was necessary for the logistic regression analyses used in this paper. Our overall conclusions remained consistent across these different degrees of threat.

Extended Data
Point estimates represent median standardized odds ratios of species being threatened, generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions. The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and associated species. See Supplementary Tables 13-14 for sample sizes and model estimates, respectively. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy. Fig. 8 | Excluding species designated as threatened under IUCN criteria A and B. The relative importance of forest integrity, structural condition, and forest cover on the odds of mammals, birds, reptiles, and amphibians being threatened. These analyses were performed after excluding (a) 2,751 species listed as threatened in criterion B of the IUCN Red List of Threatened Species, and (b) 3,745 species listed as threatened in both criteria A and B of the IUCN Red List. We did not include declining population data in this analysis of potential circularity because the IUCN Red List criteria are not used for determining overall population trends. Point estimates represent median standardized odds ratios of species being threatened, generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions. The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and associated species. See Supplementary Tables 15-17 for sample sizes and model estimates. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy.

Extended Data
Article https://doi.org/10.1038/s41559-022-01915-8 Extended Data Fig. 9 | Change in forest structural condition and forest cover loss. The relative importance of change (degradation) in tropical rainforest structural condition and forest cover loss between 2012 and 2018 on the odds of mammal, bird, reptile, and amphibian species being threatened and having declining population trends. Point estimates represent median standardized odds ratios of species being threatened (circles) or having a declining population (squares) generated by exponentiating standardized coefficients (log odds) of 100 phylogenetic logistic regressions. The vertical dotted line represents an odds ratio of 1, denoting statistical non-significance. Error bars represent median 95% confidence intervals generated with 2,000 parametric bootstraps in each regression. Each regression was performed with one phylogenetic tree randomly drawn from 10,000 available trees for each taxonomic group. Separate models were parameterized for rainforest-obligate and associated species for each response variable. See Supplementary Tables 1a and 19 for sample sizes and model estimates, respectively. Illustration credits: Steven Traver, Ferran Sayol, Birgit Szabo, and Jose Carlos Arenas-Monroy.