Tree mode of death and mortality risk factors across Amazon forests

The study addresses what drives tree mortality across Amazonian tropical forests and how modes of death (standing vs. broken/uprooted) and risk factors vary across regions. Contextually, Amazonia stores 150–200 Pg C and contributes ~12% of the terrestrial carbon sink; mortality strongly shapes spatial carbon storage and temporal sink variation. Yet, mechanisms behind tree death at biome scale are poorly understood, limiting representation in Earth-system models. Physiological failures (senescence, competition stress, pathogens, hydraulic failure) often lead to standing death; structural failures (storms, treefalls) lead to broken/uprooted deaths. Tree size and growth are expected to influence mortality (e.g., larger trees vulnerable to drought hydraulics and lightning; small trees to light competition), and individual slow growth may precede death. Species vary in baseline mortality and tolerance to stress; theory predicts a growth–survival trade-off (fast-growing species with low wood investment have higher mortality; slow-growing, denser-wood species have lower mortality). Prior evidence for this trade-off is strong in juveniles but mixed for adults and rarely tested across large geographic scales. The authors hypothesize spatial patterns in causes and risk factors align with climate gradients and forest structure, and test the roles of tree-level attributes (size, relative growth) and species traits (mean growth, max size, wood density, drought affiliation) in predicting mortality risk across Amazonia.

The paper synthesizes prior work on mechanisms of tropical tree mortality, including physiological vs. structural causes and their indicators (standing vs. broken/uprooted), and regional expectations along precipitation seasonality gradients. It reviews evidence that larger trees are more drought- and lightning-vulnerable, while small trees face light competition, and that slow growth often precedes death. It highlights the growth–survival trade-off framework, with fast-growing, low wood density species having higher mortality and shorter lifespans, versus slow-growing, dense-wood species with lower mortality, noting strong juvenile evidence and limited adult-scale, site-specific studies. Previous localized Amazon studies on mode of death are cited, as well as work indicating drought resistance shapes species distributions and can influence mortality. Basin-wide structure–function variation mediated by soils and climate, and observations of drought-induced mortality events, are referenced to motivate a comprehensive, spatially explicit assessment.

Data: 189 long-term RAINFOR forest inventory plots across lowland, terra firme, intact Amazonian forests (total area 331.05 ha; average plot size 1.23 ha; average census interval 2.8 years). Plots with census intervals >10 years were excluded; plots <0.5 ha were excluded or joined if <1 km apart. Standardized protocols measured all trees and palms with DBH ≥10 cm at each census; lianas and specific non-woody arborescent taxa were excluded. Dataset included 124,571 trees and 23,683 deaths for descriptive analyses; survival models used 158 plots with ≥3 censuses, 116,431 trees and 21,272 deaths. Mortality rate calculation: For each census, mortality rate (% yr−1) computed as m = (1 − (Na/Ni)) × 100, accounting for interval length T between censuses; plot-level rates are means across censuses weighted by interval length. Regional analyses grouped plots into four geological/climatic regions: Northern, East-Central, Western, Southern Amazonia. Regional comparisons used bootstrapped, area-weighted means with 95% CIs and post hoc Tukey tests. Mode of death: A standardized field protocol classified dead trees as standing vs. broken/uprooted. Analyses of mode proportions used 125 plots where at least 50% of dead trees and ≥5 individuals had mode recorded (total 16,599 dead trees). Because longer census intervals increase the chance standing dead trees break, authors corrected for census-interval length by modeling plot-level mode proportions as functions of mean census interval and using adjusted intercepts; regional proportions were estimated including interval as covariate. Species traits and tree-level predictors: Species traits assembled from prior studies included wood density (g cm−3), maximum stem diameter (95th quantile, mm), species mean diameter growth (mm yr−1), and water-deficit affiliation (WDA, mm; abundance-weighted mean of sites’ maximum cumulative water deficit). Missing trait values were imputed at genus, family, or plot mean levels as needed. Tree-level predictors were stem diameter at penultimate census (D) and relative diameter growth rate between antepenultimate and penultimate censuses: rel. growth = (Dt1 − Dt0)/(t × Dt0). Palms were excluded from survival analyses; trees with biologically implausible rel. growth < −5% yr−1 were removed. Statistical analysis: Cox proportional hazards mixed-effects models estimated mortality risk as a function of predictors, with plot as a random effect to account for nested structure and site-level conditions. Size effects were modeled as linear and quadratic terms (U-shaped possibility). Predictor set: tree-level (D, D2, rel. growth) and species-level (max D, mean growth, wood density, WDA). Collinearity was assessed via variance inflation factors; all predictors retained (VIF < 10). Model selection used stepAIC to minimize AIC. Importance of predictors assessed by ΔAIC from models excluding each factor and by χ2 statistics. Analyses were performed basin-wide and separately for each region using the best basin model structure, and separately for standing vs. broken/uprooted deaths. R 3.5.2 with survival, MASS, stats, multcomp, and rms packages was used.

• Spatial mortality rates: Western (2.2% yr−1; 95% CI 2.0–2.3) and Southern (2.8% yr−1; 2.4–3.4) regions had higher mortality than Northern (1.3% yr−1; 1.2–1.4) and East-Central (1.4% yr−1; 1.2–1.6).
• Modes of death: Across Amazonia, 51.2% (48–54%) of deaths were broken/uprooted; 48.4% (45–52%) were standing—statistically similar overall. Regional variation in mode proportions did not consistently track overall mortality rates: Western dominated by broken/uprooted (55% [51–59%]); East-Central had fewer broken/uprooted (39% [28–50%]); Southern, despite high dynamics, had 44% [37–52%] broken/uprooted; Northern showed near parity (49% vs. 51%).
• Predictor importance: Models combining tree- and species-level predictors outperformed those with either alone. Species-level traits alone (ΔAIC 497) outperformed tree-level only models (ΔAIC 3283), indicating species life-history is more predictive.
• Strongest predictor: Species mean growth rate was the single best predictor basin-wide; excluding it raised AIC by ΔAIC = 1734 and it had highest individual χ2. Faster-growing species had higher mortality risk in all regions.
• Tree-level growth: Lower individual relative growth strongly increased risk; excluding tree-level growth worsened AIC by ΔAIC = 260. Slower-growing individuals were especially prone to standing death compared to broken/uprooted in mode-specific models.
• Size effects: Size was important but secondary to growth (excluding size ΔAIC = 226). Effects varied by region: in Western and Southern Amazon, smaller trees had greater risk (consistent with collateral damage and belowground competition); in East-Central, larger trees were at greater risk, aligning with prevalence of standing deaths.
• Wood density and max size: Lower wood density associated with higher mortality risk; species maximum diameter had generally negative coefficients (larger max size, lower risk) except some regional variation. In basin-wide model: WD coef ≈ −0.7 (SE 0.06); Max D coef ≈ −0.002 (SE 0.0001).
• Drought affiliation (WDA): Not significant basin-wide, but regionally important. In Southern Amazon (driest), species with wet affiliation (less drought-tolerant; less negative WDA) had higher risk; WDA coefficient significant and positive in South (≈ +1×10−3, SE 3×10−4). In East-Central, drought-tolerant species paradoxically had higher risk, potentially reflecting flood–drought trade-offs and past wet anomalies.
• Mode-specific analyses: Species traits remained more important than tree-level factors for both standing and broken/uprooted deaths; slower individual growth increased risk of standing death more than broken/uprooted.
• Overall, findings provide basin-scale evidence for the growth–survival trade-off in adult tropical trees and highlight spatial heterogeneity in mortality drivers.

The results demonstrate that Amazon-wide tree mortality cannot be partitioned into a uniform physiological baseline plus spatially varying structural failure; instead, both physiological and structural drivers vary across regions. The dominance of species-level traits, especially species mean growth, affirms the growth–longevity (growth–survival) trade-off as a primary axis structuring adult tree mortality across diverse Amazon forests with different climates and soils. Individual slow growth preceding death underscores the role of stress (light or water) in elevating risk and suggests physiological decline may predispose trees to structural failure. Regional differences in size effects and drought affiliation highlight how local climate and disturbance regimes (e.g., storm-related damage in Western/Southern vs. standing deaths in East-Central) shape mortality patterns. In the Southern dry fringe, higher mortality among wet-affiliated species implies current climates exceed adaptive limits for vulnerable taxa, with implications for community composition and carbon dynamics. The findings provide empirical relationships suitable for incorporation into demographic vegetation models, encouraging explicit representation of longevity strategies and growth-dependent mortality in Earth-system modeling.

This study delivers a pan-Amazonian, mechanistic picture of tree mortality, showing that species life-history traits—especially species mean growth rate—more strongly predict mortality than individual tree attributes, and that individual slow growth is a consistent precursor to death. Modes of death are split roughly evenly between standing and broken/uprooted, with regional heterogeneity not simply explained by overall mortality rates. Drought vulnerability becomes a key predictor at the biome’s drier margins, indicating climate stress beyond species’ adaptive ranges. These insights provide robust, scalable mortality functions for demographic vegetation models and improve projections of carbon dynamics. Future work should track temporal changes in these risk factors to diagnose drivers of the observed rise in Amazon tree mortality and to link mortality equations to environmental variables for predictive modeling.

• Mode-of-death classification was available for a subset (125 plots; 16,599 dead trees) and can be sensitive to census interval length; although corrected statistically, residual biases may remain.
• Causes of death are inferred (standing vs. broken/uprooted) rather than directly observed; multiple interacting processes may underlie each mode.
• Trait datasets had missing values imputed at genus/family/plot levels, potentially introducing uncertainty.
• Palms and certain non-woody arborescents were excluded; results pertain to woody trees ≥10 cm DBH.
• Regional analyses depend on plot distribution and monitoring intensity; unmeasured local factors (e.g., edaphic heterogeneity, disturbance history) may influence results.