Sensitivity of South American tropical forests to an extreme climate anomaly

The study investigates how tropical forests in South America respond to extreme climate anomalies, focusing on the 2015–2016 El Niño event. The central research question is whether forests with hotter and drier long-term baseline climates are more resistant due to pre-adaptation or more vulnerable because they operate closer to physiological thresholds. The context is the importance of intact tropical forests as major carbon sinks and the uncertainty in vegetation and coupled climate–carbon models regarding forest sensitivity to warming and drying. Prior work indicates a long-term Amazonian biomass carbon sink that has been declining and temporary suspensions of the sink during past Amazon droughts (2005, 2010). Less is known about forest carbon dynamics outside Amazonia, including Atlantic and transitional forests. The 2015–2016 El Niño produced record heat and drought across South America, offering a natural experiment to quantify impacts of both baseline climate and short-term anomalies on aboveground biomass carbon dynamics (net carbon, gains, and losses) and tree mortality patterns across 123 long-term plots. The purpose is to determine regional vulnerabilities, mechanisms (growth vs mortality), and whether baseline dryness confers resistance or heightened susceptibility.

Long-term plot measurements showed Amazonian forests have been a substantial carbon sink for decades but with a decline since the early 1990s and potential sink cessation before 2040 (Brienen 2015; Hubau 2020). Ground data documented temporary shutdowns of the biomass sink following the 2005 and 2010 Amazon droughts. Beyond Amazonia, evidence suggests seasonal Atlantic forests may lose sink capacity and Southeast Amazon forests may act as a net carbon source, likely driven by degradation. Mechanistic studies highlight growth and mortality sensitivities to water availability and temperature, with uncertainties in models concerning warming/drying impacts. The literature frames two contrasting hypotheses: drier, hotter forests could be more resistant due to adaptation or more vulnerable due to proximity to physiological limits. The 2015–2016 El Niño was the hottest and among the most severe droughts in at least 50 years in South America, making it suitable for testing these hypotheses across a broad biogeographic range using ground plots (RAINFOR, PPBio).

Study design and plots: 123 long-term, lowland, closed-canopy, mature tropical forest plots across six countries (Bolivia, Brazil, Colombia, French Guiana, Peru, Venezuela) from the RAINFOR and PPBio networks. Inclusion criteria: ≥2 censuses before the 2015–2016 El Niño, ≥1 census after, ≤5 years between the pre- and post-El Niño censuses, and matching seasonality within 120 days. The El Niño census interval was chosen to capture the local climate extremes (drought and, where possible, temperature). Median plot size: 1 ha (mean 1.05 ha, range 0.25–6.25 ha). Mean initial census May 2001, mean pre-El Niño census September 2014, mean post-El Niño census May 2017. Mean pre-El Niño monitoring length 13.3 years; mean El Niño interval length 2.7 years. Only censuses after 1983 were used.

Tree measurements and data curation: All stems ≥100 mm diameter were tagged, identified where possible, and measured at 1.3 m or above buttresses using standardized protocols. For stems with measurement point changes (e.g., buttress growth), growth rates were harmonized across intervals. Recruitment of stems crossing the 100 mm threshold during intervals was recorded. Quality control flagged extreme growth (>40 mm yr⁻¹) or shrinkage (>5 mm) for verification; corrections applied via interpolation/extrapolation or size-class-based averages where necessary (~0.4% of measurements).

Biomass and carbon flux estimation: Aboveground biomass (AGB) per stem was estimated with the Chave et al. (2014) allometry: AGB = 0.0673 × (ρ D² H)^0.976, where ρ is wood density, D diameter, H height; palms used Goodman et al. (2013) equations. Wood density values were assigned per species where available (Global Wood Density Database), otherwise genus, family, or plot means. Tree heights were measured in 108 plots (subsamples by diameter classes) and used to fit regional Weibull height–diameter models (BiomasaFP package) for Amazonian subregions to predict H for all stems. Plot-level AGB (Mg dry mass ha⁻¹), annual aboveground wood productivity (AGWP; Mg dry mass ha⁻¹ yr⁻¹), and AGB mortality (Mg dry mass ha⁻¹ yr⁻¹) were calculated with census-interval corrections (Talbot et al. 2014) to account for unobserved recruitment, growth before death, and mortality of new recruits. Carbon stocks and fluxes were derived assuming a mean wood carbon fraction of 45.6%. For each plot, Δ net carbon, Δ carbon gains, and Δ carbon losses were the differences between El Niño interval means and pre-El Niño monitoring means. Primary analyses used relative changes (percent of initial aboveground carbon); absolute-change analyses were also conducted as sensitivity checks.

Climate data and anomalies: Continuous monthly climate data were constructed for 1970–2018. Temperature combined ERA5 (2 m, 30 km, 1979–2018) with CRU TS4.03 (0.5°, 1970–1978) harmonized via regression in the overlap period and downscaled to 1 km using WorldClim v2 climatology, with plot-level lapse-rate altitude adjustments (0.005 °C m⁻¹). Precipitation combined TRMM 3B43 V7 (0.25°, 1998–2018) and GPCC v7 (0.5°, 1970–1998) harmonized over 1998–2003 and downscaled to 1 km using WorldClim via monthly ratios. Drought intensity was quantified as maximum cumulative water deficit (MCWD) computed monthly with ET fixed at 100 mm month⁻¹ (consistent with Amazonian estimates) to isolate precipitation-driven deficits. For each plot, pre-El Niño drought state was the mean annual MCWD over the pre-El Niño monitoring period; El Niño drought anomaly was the maximum annual MCWD within the El Niño census interval relative to pre-El Niño mean. Temperature anomaly was the difference in mean monthly temperature between El Niño and pre-El Niño periods using plot census dates.

Statistical analyses: Plots were weighted in regressions based on an empirically derived combination of plot area and pre-El Niño monitoring duration to reduce residual sampling-effort patterns: weights used differed for Δ net carbon, Δ gains, Δ losses. Relationships between climate anomalies/baselines and biomass responses (relative and absolute) were assessed using linear models; nonparametric Kendall’s rank-based tests and robust rank-based regression lines (Rfit) complemented where appropriate. Multicollinearity checks showed low correlations among explanatory variables (|r|<0.24; VIF ≤1.4). Mixed-effects logistic models (glmer, lme4) assessed mortality risk as a function of size class and wood density, including El Niño versus pre-El Niño periods. Multi-model inference (MuMIn dredge and model.avg) quantified standardized effect sizes of pre-El Niño climate (temperature, MCWD), climate anomalies (Δ temperature, Δ MCWD), and their interactions on Δ net carbon, Δ gains, and Δ losses, restricting to the 95% AIC confidence set and averaging coefficients with shrinkage for weakly supported terms. Sensitivity analyses excluded censuses before 2000 and used absolute-change metrics; results were qualitatively similar.

• Climate anomalies: During the El Niño census interval (mean 2014.7–2017.4) vs pre-El Niño (mean 2001.4–2014.7), 119/123 plots warmed by +0.53 ± 0.10 °C (P<0.0001), 99/123 had more negative MCWD (mean −66 ± 25 mm, P<0.0001), and 96/123 had lower annual precipitation (−215 ± 100 mm yr⁻¹, P<0.0001). Temperature and MCWD anomalies were positively correlated (P=0.008).
• Net carbon balance: The long-term pre-El Niño aboveground biomass carbon sink of 0.38 ± 0.16 Mg C ha⁻¹ yr⁻¹ declined to −0.02 ± 0.37 Mg C ha⁻¹ yr⁻¹ (indistinguishable from zero) during the El Niño interval (paired t-test, n=123, P=0.0495).
• Component fluxes: Carbon losses (mortality) increased significantly from 1.96 to 2.41 Mg C ha⁻¹ yr⁻¹ (P=0.02). Carbon gains (growth + recruitment) did not change (2.40 pre vs 2.43 Mg C ha⁻¹ yr⁻¹ during El Niño, P=0.7). Thus, sink cessation was driven by elevated mortality.
• Temperature anomaly effects: Higher Δ temperature significantly increased relative carbon losses (linear model P=0.02) and reduced Δ net carbon (P=0.02). Model slope for relative net carbon: y = −2x + 0.5 (P=0.01), implying a 0.5 °C warming caused ~0.5% AGB carbon loss relative to initial carbon. Losses slope: y = 1.6x − 0.3 (P=0.02).
• Drought anomaly effects: Stronger (more negative) Δ MCWD increased relative losses (rank-based P=0.01; linear model P=0.07) and significantly reduced gains (P=0.01), yielding reduced net carbon (P=0.02) with slope y = −0.008x − 0.08; a 100 mm stronger MCWD anomaly led to ~0.8% AGB carbon loss relative to initial carbon. Results were robust excluding 22 plots that were slightly wetter.
• Baseline climate effects: Plots with drier pre-El Niño climates (more negative MCWD) suffered greater relative net carbon losses (P<0.001), larger reductions in gains (P=0.001), and increased losses (P=0.02). Pre-El Niño baseline temperature showed no significant association with changes in net carbon, gains, or losses.
• Mortality rates and traits: Instantaneous stem mortality rates increased from 1.8% yr⁻¹ pre-El Niño to 3.1% yr⁻¹ during El Niño (+1.3 ± 0.4% yr⁻¹; P<0.0001). Increases occurred across all size classes but were proportionally greater in medium (200–399 mm) and large (>400 mm) trees, which approximately doubled, versus a 50% increase in small trees (100–199 mm). Higher wood density reduced mortality risk overall, with a stronger effect during El Niño. Mixed-effects models confirmed elevated El Niño mortality risk for medium and large trees and for lower-density wood, consistent with hydraulic failure mechanisms.
• Multi-model inference: Drier baseline climate (pre-El Niño MCWD) had slightly larger standardized effect sizes on El Niño impacts than Δ temperature, especially for reductions in gains. Significant interactions reduced gains where it was both hotter and drier (Δ temperature × Δ MCWD) and where baseline was drier and drought anomaly stronger (baseline MCWD × Δ MCWD). Net changes were dominated by mortality increases.
• Comparative context and scaling: The decline in net sink during 2015–2016 (−105% relative to a long-term South American sink of 0.45 Pg C yr⁻¹) implies an average −0.02 Pg C yr⁻¹ net biomass change over the 2.7-year interval; assuming impacts concentrated within the peak 12 months yields an estimated ~0.82 Pg C source for the El Niño year. Compared to African tropical forests, which showed only a 36% sink decline and no strong source over 12 months, South American forests were more sensitive. Despite 2015–2016 being the hottest drought on record in Amazonia, ground-based impacts were not stronger than those in the 2005 and 2010 droughts (declines of 0.73 ± 0.53 and 0.89 Mg C ha⁻¹ yr⁻¹, respectively).

The analysis shows that the 2015–2016 El Niño’s combination of heat and drought increased biomass mortality sufficiently to suspend the long-term aboveground biomass carbon sink across intact South American tropical forests. Crucially, forests in drier baseline climates experienced the strongest negative impacts—reduced gains and increased losses—indicating that pre-adaptation to seasonal dryness did not confer resistance to this extreme. Instead, forests at the dry margins of the biome were especially vulnerable, with consistent negative responses across biogeographically distinct regions (north Colombia, Minas Gerais, southern Amazon), despite differing seasonal cycles.

Mechanistically, the disproportionate increase in mortality among larger and lower wood-density trees suggests hydraulic failure under elevated vapor pressure deficits and soil moisture stress as a key driver, while growth remained relatively stable. Temperature anomalies per se were important for increasing losses, whereas baseline heat was not, consistent with photosynthetic optimization to local temperatures but vulnerability to extreme vapor pressure deficit events. The dominance of mortality responses over growth changes explains why interactions that reduced gains had limited effect on overall net carbon.

Comparisons with satellite-based CO₂ fluxes show general agreement that the regional net carbon balance weakened or reversed during El Niño but highlight differences in processes measured: ground plots register committed biomass losses from mortality not immediately released atmospherically, while satellites integrate all ecosystem and disturbance fluxes, including fires and soil respiration. The absence of increasing sensitivity compared to past droughts (2005, 2010) suggests no progressive loss of resistance in intact forests despite repeated anomalies, underscoring their continued role in climate mitigation if protected. Nonetheless, South American forests appear more sensitive to El Niño than African counterparts, emphasizing regional differences in climatic exposure and ecosystem traits.

This study provides the largest ground-based assessment to date of how South American tropical forests responded to the extreme 2015–2016 El Niño. It shows that the long-term biomass carbon sink halted due to elevated mortality, with forests in drier baseline climates being most affected, contradicting the notion that pre-adaptation to dryness ensures resistance to extremes. Temperature anomalies intensified mortality independent of baseline temperatures, while baseline dryness strongly modulated vulnerability via both reduced gains and increased losses. Despite record heat and drought, impacts were not stronger than those of earlier droughts, indicating intact forests have maintained resistance so far and remain a critical defence against climate change if conserved.

Potential future research directions include: extending monitoring to capture multi-year legacy effects and recovery trajectories; integrating soil and necromass carbon dynamics to close ecosystem carbon budgets; disentangling roles of edge effects, degradation, and fire; improving mechanistic modeling of size- and trait-dependent mortality under compound heat–drought extremes; and comparing sensitivities across tropical regions to refine Earth system model representations of forest–climate feedbacks.

• Temporal window: The El Niño anomaly was shorter than the mean 2.7-year census interval, so estimates blend extreme and more typical months; impacts likely concentrated within ~12 months, introducing dilution of peak effects.
• Plot climate capture: A few plots’ El Niño census intervals captured only one extreme (usually drought), potentially underrepresenting concurrent heat effects.
• Scope of ecosystems: Analyses focused on intact, closed-canopy forests; results may not generalize to degraded, fragmented, or fire-affected forests where mortality processes differ.
• Process coverage: Ground measurements quantify aboveground biomass stocks and mortality (committed emissions) but not immediate atmospheric fluxes or belowground/soil carbon dynamics; comparison with atmospheric/satellite data involves different processes and uncertainties.
• MCWD assumptions: Drought metric used fixed monthly ET (100 mm), potentially simplifying spatial/temporal variability in evapotranspiration.
• Model explanatory power: Multi-model frameworks explained a modest fraction of variance in responses (e.g., 12–31%), indicating unobserved drivers and stochasticity.
• Affiliation and sampling heterogeneity: Plots varied in size and monitoring length; although weighting addressed this, residual heterogeneity may remain.