Forest fragmentation impacts the seasonality of Amazonian evergreen canopies

The study addresses long-standing controversy about the magnitude, timing, and controls of leaf phenology in Amazonian evergreen forests. Prior field and remote sensing work indicates seasonal variation in leaf demography and canopy structure, with many species flushing new leaves during the dry season linked to higher solar radiation and increased productivity. Yet, when water stress occurs, leaf development decreases and leaf/branch abscission increases. Phenology is also shaped by genetic and functional strategies. Forest fragmentation may exacerbate climatic impacts via edge-induced microclimate changes (higher temperature, wind, lower soil moisture) and altered species composition, but ground observations have shown mixed signals, leaving uncertainty about fragmented forest responses. Passive optical sensors mainly capture upper canopies and cannot disentangle vertical stratification. The authors apply repeated terrestrial LiDAR (TLS) to resolve vertical canopy dynamics and link them to microclimate. They hypothesize: (1) vertically stratified phenology in undisturbed forests varies with seasonal microclimatic conditions; (2) understory phenology depends on seasonal changes in the upper canopy; and (3) fragmentation alters phenology, with hotter, drier edge conditions exacerbating upper canopy losses during the dry season.

Previous studies have shown that a majority of Amazon forests respond to climatic variations, with evidence of seasonal dynamics in evergreen canopies, including dry-season leaf flushing associated with increased radiation and productivity. Under drought or high evaporative demand, leaf development declines and leaf/branch shedding increases. Phenology is influenced by genetic and functional traits optimizing performance under dry-season conditions while reducing competition and herbivory. Fragmentation alters species composition toward early successional, resource-acquisitive taxa and changes microclimate (higher temperatures, wind exposure, potentially lower soil moisture), potentially increasing leaf and branch turnover; however, litterfall seasonality near edges has been reported as mild, and species drought resistance and acclimation complicate predictions. Passive optical remote sensing offers limited insight into vertical stratification, often reflecting upper canopy responses. LiDAR-based approaches and prior studies have suggested stratified responses where understory growth occurs when upper canopy abscission increases light penetration. This work builds on those insights, leveraging high-frequency TLS to quantify stratified phenology and fragmentation effects.

Study area: Central Amazonia (2°20′30″S, 60°05′37″W) within the Biological Dynamics of Forest Fragments Project (BDFFP), a long-term fragmentation experiment. A 100-ha fragment was sampled along transects spanning distances from edges (0–500 m), with edges dominated by early successional species and the interior comprising undisturbed primary forest. Data acquisition: TLS surveys using a RIEGL VZ-400i were conducted between April and October 2019 approximately every 15 days (with a 40-day gap between late April and early June), covering two 100 × 10 m transects near edges and one 30 × 10 m interior transect (total ~0.52 ha). Scanner settings included 40 mdeg angular resolution in azimuth and zenith, 600 kHz pulse repetition, up to 350 m range, and up to eight returns per pulse. Multiple scan lines (5 m spacing) and additional tilted scans (zenith range 30–130°) were used to minimize occlusion. A total of 276 scans per campaign were co-registered (RiSCAN PRO v2.9), producing one point cloud per transect per date. Deriving structural metrics: A common 0.5 m DTM was generated from first-survey ground returns (lidR/LAStools). Plant Area Density (PAD; m² m⁻³) was computed using AMAPVox with a voxel size of 1 m³ and a 5 m buffer, applying the Free Path Length estimator and assuming spherical leaf angle distribution (G(θ)=0.5). Directional gap probability was modeled as P(θ,l)=exp(−λθkl). Plant Area Index (PAI; m² m⁻²) was obtained by summing PAD along vertical columns (X,Y). Across all campaigns, 230,609 voxels were monitored 11 times. PAD profiles indicated two main vertical strata: understory (<15 m) and upper canopy (≥15 m); PAI was computed separately for each stratum and combined (total PAI). Edge effects and strata definition: Nonlinear mixed models (nlme in R) related PAI to distance from edge with an asymptotic term exp(−2x) and transect as a random intercept. A piecewise linear (“hockey-stick”) model (SiZer) identified an edge threshold at ~37 m; forests within 40 m of margins were classified as edges; interiors were ≥40 m from edges. Understory contributed ~62% (interior) and ~68% (edge) of total PAI during the study period. Microclimate and climate data: Daily PAR (direct, diffuse, total) was derived from MODIS MCD18A2 V6 (5 km). Rainfall came from NASA POWER (0.5°) and was aggregated to 30-day totals to define the dry season (running 30-day precipitation <200 mm). Microclimate was measured with 22 TMS data loggers recording air temperature (15 cm above ground) and soil moisture via TDT (8 cm depth) every 15 min from 27 April to 16 October 2019 (435,798 paired readings). Soil moisture was calibrated to volumetric content using site-specific texture and density. For visualization, fifth-order polynomials were fitted to microclimate time series. Statistical analysis: Linear mixed-effects models predicted PAI as a function of time (categorical by measurement date), edge effects (edge vs interior), and their interaction, with random intercepts for edge effects nested within transect and varIdent weights to account for differing sampling effort. Model selection was based on AIC; separate best models were identified for understory, upper canopy, and total PAI. Performance and uncertainty were assessed by 200 random 80/20 calibration/validation splits, generating 95% confidence intervals. Relationships between strata were tested with simple linear regression of understory PAI vs upper canopy PAI by habitat (edge vs interior).

• A distinct 4-month dry season (running 30-day rainfall <200 mm) coincided with high MODIS-estimated PAR and elevated understory air temperatures; edges were consistently hotter than interiors, while soil moisture showed smaller edge effects.
• Vertical stratification: PAD change profiles revealed two axes of seasonal variation, with positive PAD changes below 15 m (understory) and negative changes above 15 m (upper canopy). Edge effects on PAI were significant within ~35–40 m of margins.
• Interior forests: Understory PAI declined rapidly from April to early June (t = −3.4; P < 0.001), reaching −5.3% (−0.43 m² m⁻²) by late July (t = −4.2; P < 0.001), then recovered fully by September (t = −1.2; P = 0.21). Upper canopy PAI was stable until late September, then decreased by 7.6% (−0.25 m² m⁻²) (t = −3.9; P < 0.001).
• Edge forests: Understory PAI showed a small, statistically non-significant decline of 3.4% (−0.08 m² m⁻²) from April to July (t = −1.7; P = 0.07). Upper canopy PAI decreased earlier and significantly by mid-July (−6%, −0.25 m² m⁻²; t = 2.2; P < 0.05), about three months before interior upper canopy losses.
• Total PAI: Both edges and interiors showed similar total PAI declines early in the dry season (edges −3.2%, −0.25 m² m⁻², t = −2.2; P = 0.03; interiors −2.7%, −0.34 m² m⁻², t = −2.8; P < 0.005) and relative stability thereafter, indicating that unstratified measurements mask contrasting strata dynamics and are dominated by understory PAI.
• Microclimate linkages: Upper canopy PAI losses occurred when maximum daily temperatures exceeded ~35 °C. Edge canopies experienced losses earlier, coincident with 3–5 °C higher temperatures versus interiors throughout the dry season. Understory in interiors increased PAI with enhanced light availability from upper canopy losses, even under warmer and drier conditions.
• Strata coupling: In interiors, understory and upper canopy PAI were strongly negatively correlated (R² = 0.84; P < 0.001), indicating light-mediated compensation. No significant relationship was observed at edges (R² = 0.02; P = 0.29).

The findings confirm that Amazonian phenology is strongly stratified: understories and upper canopies exhibit asynchronous dynamics. In interior forests, upper canopy plant area is sensitive to high temperatures, with losses occurring late in the dry season when maximum temperatures surpass ~35 °C. Conversely, the understory greens up as light availability increases due to upper canopy thinning, even during periods of lower soil moisture and higher atmospheric demand, underscoring light as a dominant control on understory phenology. Fragmentation modifies these controls: consistently warmer edge microclimates accelerate and amplify upper canopy losses by about three months, while edge understories, already in light-rich environments due to lateral light and gap formation, exhibit reduced seasonal dependence on upper canopy changes, decoupling strata. These results highlight that unstratified canopy metrics (e.g., total PAI, passive optical greenness) can obscure critical vertical dynamics and lead to under- or misinterpretation of climatic influences. The study aligns with trait-based understanding that understory trees are more embolism-resistant and maintain function under drought, whereas upper canopy trees are more vulnerable to high VPD and water stress. Comparisons with other Amazon sites suggest site-specific controls (e.g., rooting depth, fertility, canopy height) modulate phenological responses. The stratified, edge-sensitive dynamics have implications for productivity and carbon cycling: earlier and prolonged upper canopy losses at edges may reduce the photosynthetically active period of large trees and potentially increase mortality, contributing to degradation-related carbon emissions. Active LiDAR observations are essential to capture height-stratified responses that passive sensors miss.

Repeated terrestrial LiDAR combined with microclimate monitoring revealed that Amazonian evergreen forests exhibit strongly stratified phenology and that fragmentation alters these dynamics. Interior forests showed dry-season understory declines followed by recovery as upper canopies thinned under high temperatures late in the season; edges experienced significantly earlier and persistent upper canopy losses with a largely aseasonal understory. These findings demonstrate temperature sensitivity of upper canopies, light-driven understory responses, and a decoupling of strata at edges. The work underscores the need to account for vertical structure in phenology assessments and climate–vegetation interactions, and suggests that edge effects may reduce the carbon uptake of large trees. Future research should expand multi-year, multi-site TLS/UAV/airborne LiDAR observations to generalize patterns across Amazonian gradients, integrate species- and tree-level phenological mechanisms via TLS-based segmentation, and link stratified phenology to carbon and water fluxes. Incorporating height-stratified remote sensing (e.g., UAV LiDAR, GEDI) will improve predictions of phenological and carbon-cycle responses under warming and increasing fragmentation.

• Temporal and spatial scope: Observations cover one dry season (April–October 2019) and a limited area (~0.52 ha) within a single BDFFP fragment; generality across years, drought intensities (e.g., El Niño), and Amazonian gradients remains untested.
• PAI composition: PAI includes leaves and woody elements; leaf–wood separation was not performed, so leaf and branch turnover could not be disentangled.
• Optical assumptions: Analyses did not account for potential seasonal changes in leaf angle distribution, transmittance, or water content that could affect LiDAR-derived PAD/PAI.
• Methodological constraints: Assumption of spherical leaf angle distribution (G(θ)=0.5) and random within-voxel element distribution may introduce bias in complex canopies.
• Remote sensing inputs: PAR and rainfall datasets are coarse relative to plot scale; microclimate measurements were understory-focused and may not fully represent canopy-top conditions.