Edge effects on tree architecture exacerbate biomass loss of fragmented Amazonian forests

The study investigates how forest fragmentation and associated edge effects alter the three-dimensional architecture and allometry of Amazonian trees and the implications for aboveground biomass (AGB) and carbon cycling. Tree architecture regulates carbon allocation, respiration, hydraulic safety/efficiency, light capture and mechanical stability, thereby influencing biomass and carbon storage. Fragment edges experience higher light availability, elevated temperatures, lower water availability, and greater wind turbulence, all of which could drive architectural adjustments (e.g., branching patterns, crown shape, hydraulic path length, symmetry) and modify allometric relationships used to estimate AGB. Traditional allometry based on DBH and height may be biased if edge-induced architectural changes are not considered. Using terrestrial laser scanning (TLS) to quantify fine-scale architectural traits and develop edge- versus interior-specific allometries, the authors test two hypotheses: (1) both pre-existing (surviving) trees and post-fragmentation recruits (colonising trees) exhibit architectural and allometric changes near edges; and (2) these allometric changes measurably impact forest AGB at larger spatial scales.

The paper situates its work within several strands of literature: (a) tree architecture and its links to carbon cycling, hydraulics, light capture, and mechanical stability in tropical forests; (b) fragmentation edge effects on microclimate (higher temperature, lower soil moisture), wind turbulence, mortality of large trees, lateral light penetration, and consequent changes in species composition and crown structure; (c) observed biomass declines in Amazonian forest fragments attributed largely to elevated mortality of large trees; and (d) the emerging use of terrestrial LiDAR to quantify 3D tree structure, derive architectural traits, and improve non-destructive allometric models. Prior studies indicate strong edge effects on forest structure up to hundreds of meters, potential underestimation of biomass by traditional methods, and species- and size-dependent architectural responses. However, uncertainties remain about how fragmentation alters architecture across ontogenetic stages and species, and how this propagates into allometry and AGB estimates—gaps this study addresses with TLS-based trait extraction and plot-level AGB calculations.

Study area: Central Amazonia near Manaus, Brazil, within the Biological Dynamics of Forest Fragments Project (BDFFP), with forest fragments originally isolated in 1980–1983. Fragments (1, 10, 100 ha) experience hotter, drier, windier edge microclimates and higher understory light compared to interiors. TLS data acquisition: RIEGL VZ-400i TLS collected April 2019. Resolution 40 mdeg (point spacing ~34 mm at 50 m), 600 kHz pulse rate, up to eight returns, range up to 350 m. Six transects: five near edges (three 100×10 m, two 50×10 m) perpendicular to edges; one interior transect (30×10 m) at 500 m from any edge. Each transect had three parallel scan lines, scans every 5 m within and between lines; additional tilted scans (scanner 90°) to better capture canopy. Total 1,188 scans. Co-registration in RiSCAN PRO; automatic onboard alignment refined post hoc. Tree segmentation and QSM: Automatic individual tree extraction using a shortest-path method to the base layer with directional constraints, followed by extensive manual correction to remove lianas and fix misassignments. Severely damaged/dead trees or trees with inseparable crowns were excluded. 315 trees were segmented. Quantitative Structural Models (QSMs) fitted cylinders to point clouds to derive topology and geometry (diameters, lengths, angles, volumes). To limit small-branch bias, branches <2 cm diameter were trimmed (following Jackson et al.). QSMs provided DBH, total height, woody volume (total and partitioned into trunk/branches), and architectural traits. Tree sizes spanned DBH 2.4–72.3 cm, height 5.0–35.8 m, volume 0.01–13.0 m³. Architectural traits retrieved: surface area per unit volume (m² m⁻³) for trunk and for branches (indicative of thickness/maintenance costs), asymmetry (wood allocation shifted from vertical axis), path fraction (umbrella-shaped crown/light-optimizing path lengths), relative crown depth (vertical crown extent per unit height; often linked to multi-stemming), and relative crown width (horizontal crown extent per unit DBH). Ecological interpretations summarized in Table 1 of paper. Classifying trees by establishment time: Using long-term BDFFP records, trees >20 m height were deemed survivors established before fragmentation (pre-1990). Of trees <20 m, ~66% were post-fragmentation recruits. A 20 m height threshold separated colonising (<20 m) from surviving (>20 m) trees. Edge-effect extent estimation: Mixed linear models assessed trait dependence on edge vs interior category, tree height, and their interaction, with nested random effects for region, fragment size, and plot. Edge-effect distances (1–100 m) were scanned to find the distance of maximal absolute t-value for the “edge effects” term per trait, defining the depth of edge influence used to classify trees for subsequent analyses. Allometric modeling: Linear mixed-effects models (log–log) predicted woody volume from DBH²×H (Eqs. 1–2) and from DBH only (Eqs. 3–4), separately for edge and interior trees, including nested random effects (region, landscape/fragment size, plot). Quadratic terms were tested via AIC; 200 permutations with 80/20 train/validation quantified model stability and uncertainty. AGB estimation across plots: Using 44 1-ha permanent plots (28 edge, 16 interior) across four BDFFP sites (Dimona, Colosso, Florestal, Porto Alegre), stem censuses (>10 cm DBH, >12,000 stems, 1026 species) and site as a random effect, linear mixed models estimated plot AGB. Differences using interior allometry (Eq. 3) for both edge and interior plots quantified AGB change due to forest structure alone within 100 m of edges. Differences using interior (Eq. 3) vs edge (Eq. 4) allometries quantified additional AGB change due to edge effects on tree allometry within 55 m of edges. Confidence intervals incorporated parameter uncertainty.

- Edge-effect extents varied by trait: strongest within ~10 m for relative crown width/depth, ~20 m for path fraction, ~40 m for trunk/branch surface area per unit volume, and ~55 m for asymmetry. - Surviving tall trees (>20 m) at edges: had higher trunk surface area per unit volume (thinner trunks) than interiors (CI95% edges 24–26 m²/m³ vs interiors 14–16), greater symmetry (CI95% edges 14–18 vs 11–13), and lower path fraction (CI95% edges 0.63–0.67 vs 0.69–0.73). Relative crown width and depth were broadly similar to interiors when normalized by DBH and height. - Colonising short trees (<20 m) at edges: exhibited thicker branches and trunks (lower surface area per unit volume: branches CI95% edges 170–200 vs interiors 250–400 m²/m³; trunks CI95% edges 74–77 vs 80–100 m²/m³), higher asymmetry (CI95% 3.0–3.2 vs 2.0–2.5), higher path fraction (CI95% 0.59–0.63 vs 0.53–0.57), deeper relative crowns (CI95% 0.60–0.75 m⁻¹ vs 0.50–0.55), and smaller relative crown width due to larger trunks (CI95% 0.41–0.50 m cm⁻¹ vs 0.51–0.56). - Architectural variation correlated with woody volume; a principal component linked to thicker trunks/branches explained nearly half of the trait variation and associated with higher woody volume. - Allometry with DBH²×H (Eqs. 1–2): For given DBH and height, surviving tall edge trees had similar woody volume to interiors; short colonisers at edges had up to 50% higher woody volume than interior counterparts. Example: 10 m height, 10 cm DBH: 0.18 m³ (edge) vs 0.12 m³ (interior, +50%). - Allometry with DBH only (Eqs. 3–4): Many edge trees had disproportionately lower height for a given DBH, reducing volume predictions. Example: 70 cm DBH tree: 6.27 m³ (edge) vs 8.14 m³ (interior, −30%). For a 33 m, 70 cm DBH surviving tree: 7.4 m³ (edge) vs 7.7 m³ (interior; non-significant difference when height is included). Model fits: R²interior ≈ 0.90 (DBH²H), 0.88 (DBH-only); R²edge ≈ 0.89 (DBH²H), 0.87 (DBH-only). - Plot-scale AGB: Mixed models across 44 ha show a significant AGB reduction of 24.7 Mg ha⁻¹ near edges (t = −3.1; P = 0.003), about 10% relative to interior AGB (282.2 ± 15.3 Mg ha⁻¹). Components: −18.7 Mg ha⁻¹ (6.6%) due to forest structure changes within 100 m of edges, and an additional −6.0 Mg ha⁻¹ due to edge effects on tree allometry within 55 m of edges—one-third of total edge-related AGB loss. - Regional implications: Given vast total edge area (~176,555 km²), edge-induced architectural and allometric changes likely represent a substantial, previously under-quantified component of carbon losses in fragmented Amazonian forests.

Findings support both hypotheses. Edge environments alter tree architecture, with life-stage dependence: surviving tall trees near edges show lower path fractions and greater symmetry—traits consistent with improved hydraulic safety and mechanical stability under hotter, drier, windier conditions—while maintaining similar woody volume for given DBH and height. In contrast, colonising short trees prioritize light capture via deeper crowns, increased asymmetry, and higher path fraction, and allocate more wood to branches and trunks, increasing woody volume. However, despite greater aboveground allocation in colonisers, some large edge trees exhibit disproportionately reduced height for a given DBH, likely due to microclimatic stress, wind damage, liana loads, and collateral damage from neighboring tree failures. This height suppression drives a 30% decline in large-tree woody volume when only DBH is considered, reducing plot AGB. Edge effects on allometry contributed an additional 6.0 Mg ha⁻¹ to AGB losses (one-third of total 24.7 Mg ha⁻¹), highlighting that structural changes alone (mortality, recruitment, size structure) underestimate biomass declines if allometric shifts are ignored. The apparent contradiction between higher plant area index (PAI) at edges and reduced AGB is resolved by the disproportionate role of large trees in AGB and the dominance of acquisitive, low wood-density pioneer species at edges, which elevates PAI without increasing biomass. The work underscores TLS’s capability to capture fine-scale architectural adjustments that affect biomass estimation and to refine allometric models for fragmented forests. It also suggests potential selection for individuals/species with traits suited to edge conditions. Broader generality should be assessed across Amazonian gradients and less controlled landscapes where edge effects may penetrate farther and interact with logging and fire.

This study demonstrates that edge effects in fragmented Amazonian forests significantly modify tree architecture and allometry, with life-stage-specific responses: colonising trees increase woody volume via light-capturing architectures, while some surviving large trees have reduced heights for their DBH, decreasing their woody volume. Incorporating edge-specific allometry reveals that altered allometry alone accounts for roughly one-third of edge-related AGB losses, emphasizing the need to adjust biomass estimation methods in fragmented landscapes. TLS-based 3D measurements provide a powerful, non-destructive means to quantify these changes and update allometric models. Future research should: (1) monitor architectural trajectories of colonising trees to assess long-term survival, height growth, and mechanical/hydraulic risks; (2) test edge-effect penetration and architectural responses across wider Amazonian environmental gradients and disturbance regimes; (3) integrate species identity, functional traits, genetics, and physiology with TLS-derived architecture to elucidate mechanisms; and (4) improve remote sensing approaches to reconcile PAI and AGB signals in fragmented forests.

- Within-plot variability dominated trait variation, reflecting local edge micro-environment, species composition, and ontogenetic effects, potentially limiting generalizability. - Analytical uncertainties (measurement, scan co-registration, tree extraction, QSM fitting), especially for small branches, could not be fully quantified; branches <2 cm were trimmed to reduce bias. - The 20 m height threshold to classify survivors vs recruits is an empirical approximation based on BDFFP history; some misclassification may persist among short trees. - BDFFP’s controlled conditions (limited logging, fire, hunting) may yield conservative edge-effect estimates relative to more disturbed Amazonian landscapes. - Edge-effect penetration distances (e.g., 55 m for allometry) may be larger elsewhere; AGB losses could thus be underestimated. - Wood density variation and species turnover affect AGB; although plot models account for site-level differences, residual uncertainty remains in scaling to the region.