Low growth resilience to drought is related to future mortality risk in trees

Forests provide essential ecosystem services, yet are threatened by climate change and increasing drought frequency, duration and severity. While severe drought can reduce productivity and trigger tree mortality, some trees survive repeated droughts, suggesting that resilience to drought may influence long-term survival. Growth resilience (the capacity to return to pre-drought growth rates) comprises resistance (limiting the immediate impact) and recovery (regaining pre-drought growth). Prior studies have documented physiological mechanisms of drought-induced mortality and empirical growth indicators, and shown differences in drought legacies between gymnosperms and angiosperms. However, a direct linkage between past growth resilience to drought and future mortality risk at the individual tree level had not been empirically tested at broad scales. This study asks whether low growth resilience to past severe droughts is associated with increased future mortality risk, whether the relevant resilience component differs between angiosperms and gymnosperms, and how these relationships depend on site aridity, drought intensity, soil properties, and tree size.

Previous work has sought mechanistic indicators of mortality risk based on plant carbon and water economy (e.g., hydraulic failure and carbon starvation), and empirical indicators from radial growth patterns that often precede death. Resilience concepts from ecology have been applied to tree growth, with resistance and recovery as key components. Large-scale dendroecological studies reported pervasive drought legacies and differences in growth responses and legacies among taxa (e.g., Pinaceae vs. Fagaceae), and mixed evidence on how local climate (precipitation/soil moisture vs. PET) relates to resistance, recovery, and overall resilience. Despite this, no prior study directly linked quantified growth resilience to past drought to subsequent mortality risk across species and regions, largely due to difficulty in evaluating resilience and mortality in the same individuals.

Design and data: The study leveraged a pan-continental tree-ring database (Cailleret et al.) of coexisting surviving and now-dead trees sampled at the same sites, primarily in temperate, Mediterranean, and boreal biomes of North America and Europe. After criteria-based site selection, 118 sites with 22 species (8 angiosperms, 14 gymnosperms) and >3,500 trees (2,456 surviving; 1,454 now-dead) were included. Two growth metrics were used: tree-ring width (TRW, mm) and basal area increment (BAI, mm²).

Drought characterization: The Standardized Precipitation Evapotranspiration Index (SPEI) was used as a multi-scale drought metric (CRU TS 3.22 precipitation and FAO-56 Penman–Monteith PET; 0.5° resolution; 1901–2013). For each site, the optimal SPEI time window (1–24 months ending in a summer/early autumn target month; NH: June–October; SH: Dec–May) was selected by maximizing the fit of linear models between SPEI and the site residual chronology (1931–1980).

Growth series processing: Individual TRW series were standardized using cubic smoothing splines (frequency response 0.50 at 67% series length), autoregressive modeling, and normalization. Site residual chronologies were built using Tukey’s biweight robust means.

Drought event selection: One severe drought event per site was selected within 10–40 years before the first recorded mortality at the site, with criteria: (1) SPEI below the 10th percentile of the site’s SPEI distribution in the selected time window, and (2) abnormal low growth in the same or following year (mean TRW reduced >5% relative to the average of the 4 pre-drought years). Sites failing criteria were excluded (final N=118 sites).

Resilience indices: Following Lloret et al., indices were computed over 4-year windows pre- and post-drought: Resistance = Dr/PreDr; Recovery = PostDr/Dr; Resilience = PostDr/PreDr = resistance × recovery. Indices were computed from raw TRW and BAI.

Covariates: Climate aridity used the Global Aridity Index (1970–2000; precipitation/PET). Soil properties (ten variables from ISRIC WISE30sec and SoilGrids) were summarized via PCA; PC1 (55% variance) captured soil fertility (positively associated with N, organic C, available water capacity). Tree diameter at breast height (DBH) in the drought year, time elapsed between the drought and the series end (Δtime), and drought intensity contrasts (SPEI differences corresponding to each metric) were considered.

Statistical analysis: Linear mixed models (Gaussian; identity link) with log-transformed resistance, recovery, or resilience as response variables. Fixed effects initially included tree status (surviving vs. now-dead), taxonomic group (angiosperm vs. gymnosperm), DBH, drought intensity (SPEI during drought and relevant SPEI differences), Δtime, aridity index, soil fertility (PCA PC1), and interactions between tree status and each covariate. Due to multicollinearity among certain SPEI terms, reduced sets were applied per model. Random intercepts were included with site nested in species, and species nested in genus. Models were simplified by stepwise removal based on AIC. Analogous models were fitted for TRW- and BAI-based indices. Additional models tested including mortality source and group × aridity interactions; these did not improve fit and were excluded from final models.

• Across 118 sites and 22 species, trees that later died during droughts had significantly lower growth resilience to earlier, non-lethal droughts than coexisting survivors of the same species and site. This pattern held for both angiosperms and gymnosperms and for TRW- and BAI-derived indices.
• Taxonomic differences in resilience components: In angiosperms, mortality risk was associated primarily with lower resistance (weaker capacity to limit the immediate growth impact of drought), while in gymnosperms it was associated mainly with reduced recovery (poorer capacity to return to pre-drought growth levels). Gymnosperms showed longer-lasting growth legacies (negative effects persisting ≥4 years) in now-dead individuals.
• Soil properties: Higher soil fertility increased resistance but reduced recovery, especially in surviving trees. Net effect on resilience was neutral for now-dead trees and negative for surviving trees. Significant coefficients (standardized β) from LMMs included: Resistance – soil fertility positive (β=0.016, P=0.039); Recovery – soil fertility negative (β=−0.016, P=0.022); Recovery – Surviving × soil fertility negative (β=−0.006, P=0.039).
• Climate aridity and drought intensity: No consistent main effects of aridity on resilience components; recovery was independent of aridity. Resilience showed a positive Surviving × aridity interaction (β=0.063, P<0.001), indicating survivors in more humid sites were more resilient largely via higher resistance. Drought intensity (SPEI magnitude and SPEI differences) did not consistently influence resilience or its components.
• Tree size and temporal effects: Resilience declined with increasing DBH (β≈−0.001, P<0.001), while Δtime between the studied drought and the end of the series had a small positive association with resilience (β≈0.001, P=0.013).
• Group-specific interactions: Resistance model showed a Surviving × gymnosperm interaction (β=−0.035, P=0.008) and Recovery model a Surviving × gymnosperm interaction (β=0.037, P=0.008), consistent with resistance being more decisive in angiosperms and recovery in gymnosperms.
• Sample sizes for final models: up to 3,733 trees across 118 sites; random effects captured hierarchical structure (genus/species/site).

The findings directly address the study hypothesis by demonstrating that low growth resilience to prior droughts is linked to an increased risk of later drought-induced mortality at the individual tree level across species and regions. This establishes growth-based resilience metrics as broadly applicable proxies for mortality risk. Importantly, the resilience components associated with survival differ by taxonomic group, reflecting distinct drought-response strategies and trait syndromes: angiosperms, with narrower hydraulic safety margins, benefit from higher resistance (limiting immediate drought impact), whereas gymnosperms’ survival relates more to post-drought recovery, likely tied to constraints in embolism refilling, lower stem parenchyma fractions, and dependence on carbohydrate reserves for rebuilding xylem. The lack of a universal relationship with aridity and drought intensity suggests species are filtered by and adapted to local climates, partially decoupling vulnerability from exposure. Soil fertility’s contrasting effects (enhanced resistance but reduced recovery) and the negative association of resilience with tree size highlight how site conditions and tree attributes modulate resilience-mortality links. Collectively, these results suggest that monitoring resistance and recovery in growth time series can provide early-warning indicators of mortality risk and improve forecasts of forest die-off under future climates.

This study provides the first empirical, cross-species and cross-region evidence that low growth resilience to past droughts is associated with elevated future mortality risk in trees. It shows that the resilience component most related to survival differs between angiosperms (resistance) and gymnosperms (recovery), aligning with known physiological and life-history differences. The work advances practical, growth-based indicators for assessing individual tree mortality risk and informs forest management under increasing drought stress. Future research should (i) extend analyses to tropical forests as dendroecological data availability improves, (ii) elucidate mechanistic bases of prolonged recovery limitations in gymnosperms, including roles of hydraulic repair and carbon reserves, (iii) integrate micro-environmental variability and competition, and (iv) couple growth resilience metrics with physiological measurements to refine mortality forecasting models.

• Geographic and biome coverage is largely confined to temperate, Mediterranean, and boreal regions due to tree-ring data availability; tropics are underrepresented.
• Observational design limits causal inference; micro-environmental heterogeneity (competition, soil variability) and intrinsic trait differences may confound patterns.
• Tree size and age effects are confounded and cannot be fully disentangled.
• Drought intensity metrics (SPEI) showed limited explanatory power for resilience legacies; alternative or finer-scale drought characterizations may capture impacts differently.
• Sources of mortality co-factors (e.g., biotic agents) were considered but not retained in final models; residual effects may persist.