Warming, drought, and disturbances lead to shifts in functional composition: A millennial-scale analysis for Amazonian and Andean sites

The study addresses how millennial-scale changes in climate (precipitation, temperature anomalies, El Niño frequency) and disturbances (fire activity and general disturbance) have driven shifts in functional trait composition in Neotropical forests. Motivated by observed recent changes in productivity, species, and trait composition in tropical forests, the work aims to provide a mechanistic understanding using functional traits over centennial-to-millennial timescales, which better capture slow forest demographic processes than decadal studies. The authors hypothesize that increased rainfall and reduced drought increase the prevalence of drought-vulnerable, acquisitive taxa (low wood density, larger leaves, taller stature, larger seeds), that disturbance regimes can either favor fast-growing pioneer strategies or stress-tolerant traits depending on disturbance frequency and severity, and that climate-trait relationships are elevation-dependent (precipitation and El Niño stronger in lowlands; temperature stronger in highlands). They test these hypotheses across eight Amazonian and Andean sites spanning 0–3200 m elevation.

Prior work has linked pollen taxon changes to ecological niches to infer environmental drivers, but functional trait approaches can provide more mechanistic insights into species and ecosystem responses. Studies have shown recent shifts in tropical forest traits and composition and highlighted the roles of climate change, CO2, and disturbances. Trait-based community-weighted means have been used over decadal scales, while a prior millennial-scale case study in Peru found precipitation decreased community wood density and increased leaf size, and fire activity increased tree height and seed mass. Elevational gradients strongly structure Neotropical trait distributions, with lowlands more rainfall-limited and highlands temperature-limited, suggesting differential responses to climate drivers. The literature also indicates heterogeneous El Niño impacts across the Amazon-Andes and varying human disturbance histories affecting resilience, underscoring the need for multi-site, trait-based analyses over long timescales.

- Study sites and paleo-records: Eight forested sites across Amazonia and the Andes (lowlands 5–600 m, mid-elevation 1250 m, highlands 2780–3150 m) with fossil lake-sediment pollen records spanning at least the last 2000 years (mean ~6830 years; max ~10,020 cal yr BP). A total of 1200 pollen samples (typically ~300 grains per sample) were included. Pollen grains identified predominantly to genus were used; sampling coverage varied by site but was consistent through time within sites. - Functional traits: Four traits representing key axes of plant function were compiled for Neotropical tree genera present in the pollen records: wood density (from global WD databases), leaf area (including all leaflets; from BIEN and prior datasets), adult height (from BIEN and prior datasets), and seed mass (from Kew SID). Trait data were aggregated to genus means using Neotropical records, requiring ≥3 species per genus (or ≥1 for genera with <9 species). Phylogenetic signal was strong for WD, height, and seed mass, and weaker but significant for leaf area. For genera with high within-genus trait variability (CV>1), species occurrences outside the site’s elevational range were excluded using GBIF data. - Community-mean traits: For each pollen sample, community-weighted mean trait values were computed by weighting ln-transformed genus trait values by ln-transformed pollen abundances to reduce biases from highly productive pollen taxa and extreme trait values. The emphasis is on the direction of temporal change (standardized slopes), not absolute trait values. - Climate proxies (independent of pollen records): (1) Mean annual precipitation reconstructed from nearby speleothem δ18O records (values multiplied by −1 so higher values indicate higher precipitation). (2) Mean annual temperature anomalies over the Holocene from Kaufman et al. (Temperature 12k), using the southern tropics composite (0–30° S) at 100-year bins. (3) El Niño event frequency from Lake Pallcacocha (western Ecuador). Given spatial heterogeneity of ENSO impacts, exploratory sign-adjustments for Ecuadorian sites were considered but main analyses used the original series. Climate time series were interpolated to match pollen sample ages. - Disturbance proxies: (1) Fire activity quantified as charcoal concentration from the same sediment cores (often higher-resolution than pollen). When necessary, the charcoal measurement at or immediately preceding the pollen sample time was used. Because charcoal methods vary among studies, charcoal was standardized within records (z-transform after per-site scaling). Sites with virtually no charcoal (e.g., Lake Pata; Lake Kumpaka) reflect minimal fire. (2) General disturbance estimated as proportion of Cecropia pollen (relative to total pollen, including non-tree and unidentified), indicative of canopy openness from disturbances (e.g., land use, landslides, hurricanes). Sensitivity analyses excluding Cecropia from community trait calculations yielded similar results. - Data processing and standardization: All response and predictor variables were standardized within site (mean 0, SD 1) to enable effect size comparisons and merge heterogeneous datasets. Elevation was included to test interaction effects; main elevation effects are not expected after within-site scaling. - Statistical analyses: Cross-site models used linear mixed-effects models (nlme) with one model per trait. Fixed effects: standardized precipitation, temperature anomaly, El Niño frequency, fire activity, general disturbance (Cecropia), and elevation, plus interactions between elevation and each environmental/disturbance driver. Random effects: site-specific intercepts and year as random slope (to address repeated measures/temporal autocorrelation). Non-significant interactions were dropped. Site-specific models used generalized least squares (gls) with corCAR1 temporal autocorrelation for each trait at each site, including the same fixed predictors. Analyses were conducted in R 3.6.1. Results using a uniform 2100-year window were similar to using each site’s maximum window. - Most common pollen genera: High-elevation sites featured Acalypha, Cecropia, Hedyosmum, Hypericum, Miconia, Podocarpus, Weinmannia; mid-elevation: Cecropia, Miconia; lowlands: Alchornea, Caesalpinia, Cecropia, Iriartea, Mauritia, Trema.

- Across the eight sites, all four traits responded to climate and disturbance variables, often in elevation-dependent ways. General disturbance and temperature were the strongest overall predictors of functional change. - Adult height: Increased with precipitation; decreased with temperature in lowlands but increased with temperature at high elevations; El Niño frequency tended to decrease height. Fire activity decreased height at high elevations and tended to increase it at low elevations; general disturbance increased height only at high elevations. - Leaf area: Increased with precipitation, temperature, and general disturbance. El Niño frequency decreased leaf area at high elevations but increased it at low elevations; the disturbance effect was especially strong in lowlands. - Seed mass: Decreased with general disturbance overall; disturbance effects were strongest at low elevations. Fire activity decreased seed mass at high elevations but increased it at low elevations. - Wood density: Decreased with temperature and with general disturbance; the negative disturbance effect on WD was strongest in lowlands. - Comparative responsiveness: Adult height and leaf area were the most responsive traits; seed mass and wood density showed weaker or more context-dependent responses, likely reflecting opposing ecological pressures (e.g., drought vs. shade for WD). - Site-level analyses: 34% of 156 site-level relationships were significant, similar to cross-site (35%). Temperature (47% significant) and general disturbance (53% significant) were the strongest site-level drivers with directions broadly consistent with cross-site models. Effects of El Niño frequency and fire activity varied markedly among sites, reflecting regional differences in ENSO impacts and local disturbance histories. - Overall patterns: Wetter conditions (higher precipitation, lower El Niño frequency) favored more acquisitive trait syndromes (taller stature, larger leaves, softer wood). Warming led to taller, softer-wooded communities at high elevations (reduced frost risk, upslope migration) but shorter, softer-wooded communities at low elevations (higher atmospheric drought favoring smaller-stature, fast-cycling taxa). Fire had weak effects in lowlands but promoted stress-tolerant, small-seeded, fast-life-cycle taxa in highlands.

The results support the central hypothesis that climatic and disturbance drivers shape long-term functional composition, with trait responses modulated by elevation. Increased water availability promotes acquisitive strategies (larger leaves, taller stature), whereas warming produces divergent outcomes: in high, cold sites it alleviates growth limitations (promoting taller, larger-leaved taxa and upslope movement of lowland species), while in warm lowlands it exacerbates atmospheric drought leading to shorter-stature communities and lower wood density. General disturbance—captured by Cecropia pollen—emerges as a strong integrator of canopy-opening events that favor disturbance-adapted, fast-growing taxa, increasing leaf area and (at high elevations) height, while reducing seed mass and wood density. Wood density responses likely reflect opposing selection pressures (drought vs. shade tolerance), weakening simple precipitation–WD relationships. The weak and inconsistent lowland fire effects may result from rapid, fine-scale dynamics not captured by pollen resolution, variable fire regimes and intensities, or fire usage primarily in adjacent land uses rather than closed forests. The strong responsiveness of height and leaf area indicates plant size traits are sensitive indicators of long-term environmental change, whereas seed mass and wood density are influenced by multiple, sometimes opposing processes. Site-specific variability in ENSO and fire responses emphasizes the importance of local climate teleconnections, soils, seasonality, human impact histories, and lagged ecological responses.

This first multi-site, millennial-scale trait-based analysis across Amazonian and Andean forests shows that temperature and general disturbance are dominant drivers of functional composition change. Future warming is likely to shift lowland forests toward shorter, softer-wooded, large-leaved, disturbance-adapted communities, while highland forests may become taller with softer wood and larger leaves as frost limitations relax, implying divergent carbon storage trajectories (potential declines in lowlands, increases in highlands). The findings underscore the value of trait-based paleoecology for anticipating ecosystem functional shifts under global change and suggest that monitoring and management should consider elevation-specific vulnerabilities and disturbance regimes. Future work should refine local ENSO proxies, quantify fire intensity from charcoal, and integrate species-level trait variation to improve predictive understanding.

- Pollen-based community means rely on genus-level identifications and genus-average trait values; within-genus variability and differences between pollen and vegetation abundances can introduce uncertainty. The study focuses on standardized slopes (directions of change) rather than absolute trait values. - Cecropia abundance is both a disturbance proxy and part of the community trait calculation, potentially reducing independence; sensitivity analyses suggest minimal bias but residual dependence may remain. - Charcoal methods and particle-size classes vary among records, and charcoal abundance may not reflect fire intensity or severity; spatial heterogeneity in burning within landscapes can obscure ecosystem-scale effects. - ENSO frequency is from a single regional record (Lake Pallcacocha) and may not capture local ENSO hydroclimate impacts; modern correlations vary by region and elevation. - Temporal resolution and age-model uncertainties, plus potential ecological response lags, may blur short-term dynamics (e.g., rapid post-fire changes). - Predictor datasets and lab techniques differ among sites, necessitating within-site standardization that precludes direct comparison of absolute values.