Conservation Biology

Conserving migratory birds effectively requires full annual cycle strategies that identify where on-the-ground efforts can have the greatest impact. Here, we present a hemispheric spatial framework to identify priority areas for 112 migratory bird species across the Americas. Building on full annual cycle prioritizations, we defined finer-scale spatial planning units that reflect differences in migratory and congregational behaviors between shorebirds and landbirds. We compiled population data for each planning unit and focal species and applied conservation planning tools to design area-efficient portfolios of sites and landscapes that secure 10% of each species population within the Americas flyways. The resulting minimum area portfolios include 175 shorebird sites and 80 landbird landscapes optimized to meet the species-specific 10% representation targets across breeding, non-breeding, and passage seasons. We also identified a broader set of complementary solutions, ranked by an importance score, to provide decision-makers with flexible options for strategic resource allocation. This framework provides the scientific foundation for the Americas Flyways Initiative (AFI), which aims to catalyze investment in nature-based solutions and bird-friendly infrastructure to enhance the conservation of migratory birds and strengthen the resilience of the Americas flyways by 2050.

The number of species introduced outside of their historical ranges by human activity continues to rise. A subset of these species establishes, form self-sustaining populations, and some (invasive alien species) go on to cause substantial harm to biodiversity and ecosystems. Preventing new invasive alien species from establishing is the key focus of interventions, because post-establishment management is costly and often fails. However, it remains unclear how effective multilateral efforts have been in curbing the rise. Here we show that the emergence of new invasive alien species across countries is slowing, and trends are similarly negative across geographically diverse countries. Using data and modelling advances, we find a 35% reduction in the establishment of new invasive alien species over a policy-relevant 50-year time frame. The findings directly inform the assessment of progress for the invasive alien species target of the Kunming-Montreal Global Biodiversity Framework, and provide a global baseline for monitoring rates of invasive alien species establishment. Furthermore, the slowdown suggests that policy and investment over recent decades to prevent invasive alien species from entering and establishing in countries have had a positive effect.

1. Global restoration targets, including the Kunming-Montreal Global Biodiversity Framework and the European Union's Nature Restoration Regulation, call for extending restoration beyond protected area boundaries to reconnect isolated ecosystems, yet implementing these landscape-scale interventions forces trade-offs between biophysical effectiveness and social feasibility. 2. In the Grenoble region of the French Alps, we used a gradient analysis of 12 ecosystem services across protected area interfaces to design and compare two restoration scenarios. Both were built from the same set of co-selected interventions but differed only in their spatial allocation: a 'Technical' scenario targeting areas of low ecosystem service continuity around protected areas, and a 'Participatory' scenario co-designed with local stakeholders. 3. Both scenarios physically modified about 14.5% of the landscape, mostly through a forest-based adaptation strategy, yet their effects reached up to more than 60% of the territory for the most connectivity-dependent services, showing that benefits propagate well beyond the area directly modified. 4. The two scenarios diverged mainly in how they treated agricultural land. The Technical scenario reduced abrupt drops in ecosystem service provision at the protected area edge nearly twice as often as the Participatory one (46% versus 26% of the sharp declines it addressed), and at six locations removed them entirely. However, it concentrated over 99% of its non-forest effort in agricultural diversification, such as diversifying crops and adding hedgerows, the action stakeholders ranked hardest to implement, whereas the Participatory scenario favoured more feasible actions such as river restoration, which met less resistance but missed several critical discontinuities in ecosystem service supply. 5. These contrasts show that biophysical potential and social feasibility can pull restoration in different directions, and that the more effective spatial configuration is not necessarily the more implementable one. 6. Closing the implementation gap for nature-based solutions depended less on expanding the area restored than on where interventions were placed. Gradient-based diagnostics appear most useful not as fixed prescriptions but as tools to steer existing sectoral funding in agriculture, urban planning and forestry toward high-impact locations, an approach transferable to other regions working to reconnect protected areas with their surrounding landscapes as coupled social-ecological systems.

Businesses are increasingly encouraged to disclose their nature-related dependencies, impacts, risks and opportunities. A common component of sustainability reporting is screening operational sites for ecologically sensitivity to identify locations for further evaluation and action. However, with 600+ biodiversity metrics available, selecting and interpreting appropriate metrics remains challenging for business. We developed a simple screening framework informed by the Taskforce for Nature-Related Financial Disclosures guidance, grouping eleven widely used global biodiversity metrics into four complementary [&prime]baskets[&prime], representing different aspects of biodiversity. We created hypothetical but realistic mining, onshore wind energy and agricultural companies, to assess how metric choice, buffer size, scoring approach and sensitivity thresholds influence screening outcomes. Our basket framework consistently identified similar high-priority sites across metric combinations, but site rankings varied with methodological choices. We recommend clearer guidance on metric selection and application, alongside greater transparency from business regarding assumptions, methods and limitations when screening sites for ecological sensitivity.

Biodiversity loss is driven by unsustainable human activities, yet the contextual conditions and underlying drivers of threats remain poorly understood. We assessed how protected areas, socioeconomic conditions, and biophysical factors explain global patterns of threat probabilities across six major threat types and four vertebrate taxa. We identified key explanatory variables and their associations with threats using Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP). Socioeconomic conditions, specifically human development and income inequality, were the strongest predictors. Their associations were complex and non-linear: notably, high human development index (HDI) was associated with both higher and lower threat probabilities, depending on inequality and regional context. Second, land cover and biophysical variables, such as shrubland cover, tree cover, and elevation range, explained additional, but taxon-specific variation. Finally, protected areas showed limited ability to explain threat patterns. By linking threat probabilities to their contextual and socioecological conditions, we aim to build a better understanding of the systemic drivers of biodiversity loss.

Over the past two decades, significant population recoveries of tigers and leopards have driven a substantial rise in human-carnivore conflict across South Asia. Sustaining the long-term conservation of both species demands effective, adaptive conflict management. However, existing research is largely confined to isolated landscapes or protected areas, hindering the identification of spatial conflict drivers across broader, heterogeneous regions. To address this gap, this study compiled 4,457 human leopard conflict (HLC) and 2,563 human tiger conflict (HTC) records spanning 2000 to 2024 across Bangladesh, Bhutan, India, and Nepal. We evaluated conflict locations against 33 landscape, anthropogenic, climatic, topographic, and hydrological predictors across three spatial scales using binomial generalized linear regression models and three alternative background sampling methods. Model performance peaked at the largest spatial scale for both species and remained consistent across sampling strategies. HTC correlated positively with forest fragmentation and proximity to human settlements, but decreased with higher overall anthropogenic disturbance. Conversely, HLC concentrated in highly productive, high-precipitation, human-modified landscapes. These contrasting spatial patterns highlight distinct ecological responses to habitat modification and carry unique conservation implications. Ultimately, this study underscores the need to integrate carnivore abundance, wild prey densities, and socioeconomic factors to achieve a holistic understanding of conflict dynamics.

Information on the long-term dynamics of species distributions is essential for assessing global biodiversity change and its causes, and for informed conservation decisions. A valuable source of such historical information are atlases. They provide spatial standardisation, extensive geographic coverage, temporal replication, documented sampling protocols, and span multiple species. Here we present model-based occupancy probabilities for 1,055 bird and 58 butterfly species, derived from nine gridded atlas datasets spanning several editions over the past 50 years. These are: Japan, New Zealand, New York state (US), the states of Alberta, Quebec, and The Maritimes (Canada), Czech Republic, United Kingdom, and Europe. Despite standardisation and coordination during data collection, the raw atlas data still contain some variation in sampling effort. We accounted for this using a two-step process: (1) we estimated a proxy for sampling effort in each grid cell within each atlas period using the Frescalo Algorithm, and (2) we ran a Bayesian spatial occupancy model for each species and atlas period, using the estimated sampling effort as the single predictor of detection in the observation process of the model. As the main product, we provide occupancy probabilities for species across grid cells and atlas periods. We also report uncertainty around each probability estimate and flag ~21% (rare) species per atlas for which the model did not fit well. The probability distributions are fit for purpose and served in a user-friendly database with standardised geographic projection, metadata, and taxonomy.

Despite the popularity of carbon credits as a tool to offset greenhouse gas emissions, there are growing concerns over whether their climate mitigation claims are robust. Most attention has focused on natural risks to forest carbon projects, such as wildfire, while social risks (e.g., those driven by fraud, mismanagement, and land conflicts) remain largely unquantified. Here we analyze 244 forest-based carbon projects to assess the vulnerability of the voluntary carbon market to social risks and find that projects systematically underestimate their risk of failure. Even under optimistic assumptions, projects are roughly twice as likely to fail within 25 years as their self-assessed risk implies over a 100-year horizon. Projecting these trends forward, we show that credits issued to failed projects could exceed the reserves registries hold to mitigate these risks within 1-2 decades. Overall, our results suggest stronger mechanisms and reserves are needed to ensure carbon credits deliver the promised climate benefits.

1. Occurrence records underpin biodiversity analyses, but their coordinates are often erroneous or missing. Misplaced coordinates distort environmental predictors and range estimates used in conservation assessment. Existing tools flag suspect records or georeference locality text, but rarely correct coordinates, provide calibrated uncertainty, or quantify downstream consequences. 2. We present GeoMend, a deterministic, rule-based method that flags and corrects suspect coordinates and georeferences records lacking coordinates from locality text, assigning each a precision tier or uncertainty radius. It uses no machine learning or artificial intelligence. We evaluate it against expert coordinates for 16,913 records and 736 species on Hispaniola, compare georeferencers using 5,000 localities from 72 countries, and measure changes in environmental predictors and the extent of occurrence. 3. On held-out Hispaniola records, GeoMend flags 92% of coordinates more than 5 km from expert truth, versus 27% for CoordinateCleaner. Among the accepted corrections with paired coordinates, the median error falls from 2.19 to 0.82 km, and 68% of the complete held-out pairs improve. On records placed by both GeoMend and GEOLocate that also have original coordinates, the median error is 0.67 km for GeoMend versus 10.69 km for GEOLocate. For records lacking coordinates, evaluated uncertainty radii contain expert truth about 92% of the time in well-supported gazetteer classes. Correction reduces the mean standardized environmental distance from expert truth by 38% and moves the extent of occurrence toward expert truth by a median of 8% (mean of 79%). Corrections and newly georeferenced records give 17 Hispaniola species their first range estimate. 4. On the global benchmark, GeoMend is more accurate than GPT-4o for two-thirds of the records placed by all four comparators, with median errors of 4.4 and 10.8 km, respectively. Applied to Global Biodiversity Information Facility (GBIF) records lacking coordinates, it places records from 82,180 previously unmapped species across 233 countries, ranging from precise locality interpretations to administrative centroids. Automated placement makes the global backlog more approachable, while explicit tiers distinguish uses supported by precise and coarse coordinates. Further georeferencing requires regional expertise to combine automated placement with local knowledge.

The global market concentrates extractive pressure on lands held by Indigenous peoples, including peoples living in isolation, for whom free, prior and informed consent cannot be obtained and protection must therefore rest on territorial instruments. Halmahera, Indonesia, holds some of the worlds largest lateritic nickel reserves beneath a lowland rainforest inhabited by the Hongana Manyawa, yet the trajectory of land use and cover change across the island has not been quantified. We characterised land use and cover change over the 17,437 km2 island between 2014 and 2024 using MapBiomas time series, and projected a business-as-usual scenario to 2054 with a stochastic cellular-automata model implemented in Dinamica EGO, calibrated with weights of evidence on eight variables describing mining and logging concessions, transport infrastructure, settlements and previous clearing. Forest covered 83.0% of the island in 2014 and 82.1% in 2024; under unchanged policy it falls to 73.7% by 2054, a net loss of 162 thousand ha, or 11.2% of the 2014 baseline, at gross rates of 47,000-51,000 ha per decade. Deforestation probability is highest within 500 m of previous clearing and declines with distance from settlements, cities and mining sites, while proximity to national parks carries a negative weight of evidence. The frontier is self-propagating and spatially predictable, and legally designated territory retains forest within it. Protecting the Hongana Manyawa consequently depends on excluding extractive licensing from the interior forest ahead of the frontier rather than behind it.

Citizen science data are increasingly used to infer biodiversity change, but causal claims based on such data are credible only if sampling effort and its temporal shifts are explicitly modeled. Zhang et al. (1) used citizen science data to conclude that greater photovoltaic policy stringency, measured using the photovoltaic policy stringency index (PSI), reduced county-level bird diversity in China. We reproduced their fixed effects and instrumental variable estimates. However, the observed Shannon diversity derived from pooled citizen science records reflects both bird communities and sampling effort, which the authors' controls do not adequately capture. Accounting for observer count changed the reported statistically significant 2.10% decline in Shannon index to a nonsignificant 0.58% increase (P = 0.288) per one-standard-deviation increase in PSI, and rendered the instrumental variable estimate statistically indistinguishable from zero (P = 0.912). Yet observer count is only one of many sources of sampling bias. PSI was also associated with multiple dimensions of sampling effort, consistent with sampling effort acting as a potential mediator in the PSI-diversity chain. The sampling domain also shifted markedly from 2014 to 2023: recorded county-months increased almost 24-fold, median observer count rose from one to three, and zero-duration records declined from 57.2% to 0.17%. Without adequate adjustment, these shifts confound estimates of temporal change in observed bird diversity. Beyond its inadequate treatment of sampling effort, the original study also misinterpreted its statistical results. Although the reported R{superscript 2} values are high, they are dominated by county and year-month fixed effects, with PSI contributing a partial R{superscript 2} of only 0.048% on observed Shannon index. The PSI-photovoltaic-area correlation is also weak (r = 0.0414) and vanishes after accounting for fixed effects (P = 0.977). Furthermore, the released bird observation data contain many erroneous outliers, raising significant concerns about insufficiently rigorous data preprocessing and quality control. These results show that the released data cannot properly distinguish ecological change from sampling effort change. Robust inference from citizen science data requires checklist-level effort metadata, explicit correction for spatiotemporal sampling shifts, and close collaboration among researchers with complementary methodological and ecological expertise.

Achieving global conservation and restoration targets requires knowing not only how much land to conserve and restore, but also where such actions will have the greatest ecological effect. Conservation and, in particular, restoration priorities are often inferred from habitat condition, opportunity, or broad ecological value, yet these criteria do not necessarily identify where local changes generate the largest landscape-scale consequences. Building on sensitivity-based conservation prioritization, we extend the approach to derive ecological loss and gain potential by differentiating a persistence-relevant connectivity metric with respect to local habitat condition. Sensitivity identifies areas where local degradation or improvement would most strongly affect connected habitat at the landscape scale. Combining sensitivity with current condition distinguishes high loss potential, where degradation would have large ecological consequences, from high gain potential, where ecological improvement could generate large benefits. Applying the framework to national forest-naturalness data in Norway, we show that loss and gain potential are spatially related but far from redundant, and that gain potential is not simply concentrated in the most degraded forests. Both high-loss and high-gain areas were poorly represented within current protected areas, although loss potential was consistently better represented than gain potential. Extending sensitivity-based prioritization in this way provides a general and scalable means of linking local habitat-condition change to persistence-relevant landscape outcomes, while distinguishing the ecological potentials that can inform conservation and restoration decisions.

Target 12 of the Global Biodiversity Framework urges governments to enhance human access to nature to improve human wellbeing. Here, we study nature access using several metrics for a globally representative sample of 140 large Functional Urban Areas, examining variation among regions and over time. We find most urbanites globally do not reach common benchmarks for urban forest cover and nature access. Globally, 3.3 billion people (76% of urbanites) live in urban areas that do not have at least 30% tree canopy cover. Moreover, 1.1 billion people (23% of urbanites) are farther than 1 km from a park as identified in Open Street Map, and 1.5 billion people (36% of all urbanites) are more than 1 km from a green patch as defined by land cover. Over time, urban nature access has declined. Between 1992 and 2020, most urban residents had an increase in the distance to green patches (63% of urbanites increased, average increase 680 m). Similarly, between 1992 and 2020, 31% of urbanites had a decrease in percent natural cover, while only 1.3% had an increase. We show that the variation among cities or regions in the amount of nature access depends on the metric used and the context. On average, mesic climates have more tree cover than do arid climates, and cities with lower GDP or higher population density tend to have lower natural share than other cities. Our results suggest that global urban nature access is insufficient for supporting human health and worsening over time.

Understanding how scientific effort is distributed is essential for evaluating the evidence available to guide species conservation. We assessed the temporal, thematic, methodological, and spatial organization of jaguar (Panthera onca) research across the species range. Using 1,028 bibliographic records, we applied a metadata-based scope filter, defining a primary analytical corpus of 857 articles and an inclusive sensitivity corpus of 989. We analyzed publication growth from 1946-2024, author-keyword networks, high-specificity research-domain and methodological indicators, article-based spatial concentration across nine regions, and anthropogenic context using the Human Impact Index. Publication output increased strongly and nonlinearly, with the negative-binomial generalized additive model explaining 96% of deviance. Population and abundance, human-wildlife conflict and coexistence, and movement and connectivity were the most frequent research domains, whereas camera trapping was the most frequent methodological approach. Spatial analyses included 237 georeferenced terrestrial articles. Research was strongly concentrated in the Pantanal, Mesoamerica, and Atlantic Forest, but was underrepresented relative to range area in the Andes and Choco-Darien, Amazon, and Guiana Shield. These patterns were stable to the broader scope definition. Research concentration showed no strong association with region-wide anthropogenic pressure, although studies within several extensive regions tended to occur in more human-influenced portions than the regional average. Jaguar research therefore shows substantial growth and partial alignment with applied conservation challenges but remains geographically uneven. Expanding representative research in underrepresented regions, while maintaining work in threatened landscapes, would strengthen the evidence base for range-wide conservation planning and improve coordination among countries, institutions, and regional research traditions.

Citizen science observations are abundant, but conservation requires turning uneven records into reliable predictions and directing new surveys to where information is missing. We developed a biodiversity platform for Japan that is updated monthly and integrates 2.32 million records to predict 8,297 species across seven taxonomic groups. Shared representation models outperformed species-specific models in four groups and extended predictions to species with few records. Five independent datasets, including structured monitoring, environmental DNA and complete forest inventories, confirmed that the models ranked observed species and occupied sites above alternatives, with median AUCs of 0.724 to 0.894 across sites and 0.650 to 0.841 across species. For any user-selected area, the platform returns candidate species, distribution predictions, a biodiversity map corrected for uneven observation effort, a conservation priority map for native species and a map recommending where to survey next. This map highlights places where species with few records are predicted to occur despite limited sampling. Independent observations showed that areas ranked highly by this predicted potential contained many such species, indicating that model predictions can help direct surveys toward knowledge gaps. New observations are incorporated into monthly updates, creating a national feedback system connecting citizen science, local conservation decisions and future surveys.

Coral bleaching is among the most infamous climate impacts. Yet the extent of bleaching due to human-caused climate change, as opposed to natural El Nino variability, and the contributions of individual countries and carbon majors (fossil fuel and cement producers) remains unquantified. Here, we leverage coral monitoring globally across 4,479 sites from 1985-2024 to link specific emitters to marine heat stress and bleaching. For events with >10% of corals bleached at a site, we find that 99% (95% CI: 88-100%) were attributable to human-caused climate change. Even during major El Ninos, 97-100% of bleaching events were attributable to human-caused climate change and the extent of damage increases over time from 23% of sites bleached (95% CI: 3-35%) during 1998 to 71% of sites (95% CI: 37-85%) during 2023-2024. Emissions of the United States, the largest historical emitter, contributed to 51% (95% CI: 39-65%) of bleaching events. Emissions linked to Chevron, Saudi Aramco, Gazprom, and ExxonMobil each contributed to 14% (95% CI: 11-19%) of bleaching events -- roughly double the contribution of the 39 countries of the Alliance for Small Island States. It is virtually certain emissions of China, the United States, and Indonesia caused bleaching in their coastal waters. Strong inequalities exist where countries with high socioeconomic dependence on coral reefs experienced 54% of bleaching yet contributed only 5% to bleaching. As global warming rapidly nears levels catastrophic for warm-water corals, our results show the massive contribution from anthropogenic emissions to historical coral beaching and help fill an important evidentiary gap on accountability for non-economic loss and damage from climate change.

Human-wildlife conflicts over shared resources are often interpreted as direct competition, whereby resources consumed by wildlife are unavailable to people. Yet, individuals of wild populations can differ in their propensity to exploit resources, and the demographic, ecological, and economic consequences of such behavioural heterogeneity remain poorly understood. We developed an integrated demographic-bioeconomic model of Atlantic cod (Gadus morhua), grey seals (Halichoerus grypus), and a coastal gillnet fishery. Stage-structured population models incorporated density dependence, predator-prey interactions, adaptive harvesting, and transitions of adult seals into a bold, depredating state. We compared four scenarios of increasing complexity over 50 years across 100 coupled demographic realisations and examined sensitivity to the energetic contribution of depredated catch, harvest levels, and social amplification of depredation behaviour. Cod and seals persisted in all demographic realisations, supporting our ecological-persistence hypothesis. Contrary to our predictions, however, adding bold seals neither reduced long-term fishery performance nor destabilised the system. Median net present value was approximately three-fold greater in the full system than without depredation, and was greater in 76% of paired realisations. The full system was also the most temporally stable: median cod variability declined by 74% relative to the otherwise equivalent fishery without depredation, while seal variability declined by 45%. Behavioural depredation therefore altered predator population structure and imposed direct catch losses, but these effects were counterbalanced by indirect demographic feedbacks. Synthesis and applications. Depredating predators should not necessarily be viewed simply as competitors that reduce contested resources. Behavioural heterogeneity can introduce compensatory pathways that alter population stability and long-term economic outcomes. Management should therefore consider not only predator abundance and immediate catch losses, but also how harvest creates opportunities for behavioural change and how those changes propagate through coupled ecological-economic systems.

As anthropogenic threats have intensified over the past 500 years, we find ourselves in the midst of a sixth mass extinction, with continued losses of biodiversity threatening ecosystem stability. This biodiversity loss has caused species extinctions across taxa, and placed several others at high risk of functional extinction. These disturbance-driven impacts represent one of the most acute biodiversity crises facing global marine systems. Species exhibiting slow life histories characteristically have low resilience to disturbance. Here, we assess the risk of functional extinction and identify policy pathways for population recovery of the slow-living, Critically Endangered elasmobranch, the spinetail devil ray (Mobula mobular). We develop a stochastic, state-structured Integral Projection Model (IPM) parameterised with demographic data collected from fishery landings data in India, the world's largest mobulid fishery, and supplemented with data on vital rates from published literature. Using the IPM, we estimate that the population is declining at approximately 12% annually, experiencing substantial limiting pressure from fisheries overexploitation and failing to approach its biological maximum growth potential. Our results indicate that populations of M. mobular will be at high risk of functional extinction if 'business as usual' harvest scenario persists for another decade. We further show that long-term population recovery is only possible if survival increases significantly across all size classes, especially among large reproductive females, alongside a concurrent increase in fecundity. We conclude that no single policy measure is sufficient to recover population of M. mobular along the southeastern coast of India. Instead, combined protection through maximum bycatch mitigation and protection of nursery areas in no-take zones will be required for population recovery. This research demonstrates that recovery of overexploited populations often requires integrated resource management across life stages, and that the Critically Endangered M. mobular warrants urgent conservation action to avoid functional extinction.

Incorporating the complexities of climate change into conservation planning can be challenging. Climate change is projected to change the mean and variability of temperature and precipitation, and change dynamics of extreme events of many climate variables globally. These changes will all have compounding effects on ecosystems worldwide, affecting species distributions, seasonal timings, and population dynamics. However, most recent climate-informed conservation planning frameworks only focus on changes in mean conditions, leaving out ecologically important information about seasonality and extreme events. Using a case study of estuaries in the Pacific Northwest, this research seeks to answer the question "How can the effects of climate change on estuaries best be modelled and described for practical use in ecological management?" To answer this question, we used downscaled climate projection data to estimate changes in the mean and variability of temperature and precipitation. Applying extreme value theory to these projection data, we also projected changes in the magnitude of extremes in temperature and precipitation for estuaries in the region. Using descriptive statistics and open source data, our results present a novel, holistic method of understanding climate risk, including estimating extreme events, highlighting a key research gap in conservation planning. These results also highlight that trends in the mean, variability, and magnitude of climate extremes are not consistent with each other, further underscoring the importance of considering multiple dimensions of climate change together. Despite uncertainty given by climate models, the methods presented here provide a reasonable approach to plan for conservation management in the face of climate uncertainty.

Intensifying fire regimes threaten forests globally, but the risk of persistent post-fire forest loss and its potential mitigation remains poorly quantified. We analyzed millions of wildfires worldwide from 2001 to 2024 and tracked recovery in satellite-observed forest structure and ecosystem function. Post-fire persistent forest loss, indicated by modeled non-recovery to pre-fire conditions over decadal timescales, affected 57.1% of burned forest area globally since 2001, with hotspots in Pacific temperate and southern boreal forests. We then identified 'crucial fires' as events exceeding a stringent modeled-risk probability threshold for persistent structural or functional non-recovery, with fire severity strongly predicting this loss. This severity dependence revealed a management pathway, as locations with prior low-severity fire experienced lower severity in subsequent wildfires and had lower modeled probability of becoming crucial. Under a model-based counterfactual scenario, applying the estimated severity attenuation was associated with a 7.6% reduction; the top 1% of road-accessible areas accounted for 35% of this reduction. These results provide a global framework for identifying where wildfire threatens forest resistance and where targeted low-severity fire management like prescribed fire might be used to combat global forest loss.