Global patterns and climatic controls of forest structural complexity

Climate change will alter the structure and functioning of boreal, temperate and tropical forest ecosystems with contrasting, yet unclear impacts on biodiversity and ecosystem functions across biomes. Responses of forest biodiversity and ecosystem functions to climate change are strongly linked to changes in forest structural complexity. Understanding these impacts requires insight into the climatic controls on forest structural complexity, as climate shapes forest compositional and functional diversity—key determinants of structural complexity. Forest structural complexity quantifies the three-dimensional distribution of trees and canopies, encompassing tree size diversity, crown morphology, canopy layering and packing, and canopy connectedness. Advances in airborne and terrestrial LiDAR have enabled quantification of 3D forest structure, and structural complexity metrics are strong predictors of net primary productivity due to their links with canopy space occupation, canopy connectedness, and light absorption. Biotic determinants include species composition, complementarity in crown architectures, and tree size diversity; these depend on functional traits like shade tolerance and belowground resource acquisition, which are constrained by functional diversity. Climate, particularly humidity and temperature, controls the spectrum of plant functional strategies (physiological tolerance hypothesis), suggesting climate influences structural complexity via compositional and functional diversity. However, global patterns and climatic determinants of forest structural complexity remain largely unexplored. This study aims to quantify global variation and climatic drivers of forest structural complexity across biomes, map its global patterns, and estimate responses to climate change, focusing on primary forests to minimize disturbance effects. The authors hypothesize that global variation in structural complexity is mainly determined by climate factors controlling compositional and functional diversity: light availability during the growing season (solar radiation), mean growing season temperature, and water availability (mean annual precipitation, precipitation seasonality, and water balance as MAP minus PET). They also consider edaphic factors (soil water holding capacity, soil nitrogen, cation exchange capacity) to control for soil effects.

The paper situates forest structural complexity as a 3D attribute extending beyond traditional structural metrics like biomass, leaf area, or canopy height. Prior work shows higher structural complexity with greater diversity of tree sizes and complementary crown morphologies, yielding multi-layered, densely packed, and connected canopies. LiDAR technologies have driven development of new complexity metrics, which are linked to productivity through occupied canopy space, canopy connectedness, and light absorption. Studies indicate tree species diversity enhances structural complexity through niche complementarity in crown architectures and canopy packing. Yet, species coexistence and vertical stratification depend on traits such as shade tolerance and crown plasticity, and are constrained by functional diversity. Climate is a major control on compositional and functional diversity (e.g., physiological tolerance hypothesis), implying indirect climate-structure links. Canopy height, a limit on vertical space, also relates to water availability (hydraulic limitation hypothesis). Despite these theoretical underpinnings and regional studies, comprehensive global assessments tying climate to structural complexity across biomes have been lacking.

Study design: Primary forests were sampled to minimize disturbance effects. In total, 294 plots were established across 20 primary forest sites spanning five biomes: boreal forests (2 sites), temperate broadleaf forests (6), temperate conifer forests (4), (sub-)tropical savannas and woodlands (3), and (sub-)tropical moist broadleaf forests (5), selected to represent dominant forest types across broad climatic gradients. Plot layout and data collection: At each site, 1-ha plots (100 × 100 m) were laid out, spaced ≥200 m apart. At each plot, five single terrestrial laser scans (TLS) were conducted using FARO Focus 120 or FARO M70 scanners (tripod height ~1.3 m), with 300° vertical and 360° horizontal field of view and ~0.035° angular step. Scans followed a “five on a dice” pattern (center plus four scans 42 m from center toward corners). Angle count sampling (Bitterlich) at each scan position estimated stand basal area. Point cloud processing and SSCI computation: Scans were filtered in FARO SCENE to remove stray points and exported as .xyz (x,y,z). Resolution was downsampled to 1/16 (~0.14° step) for processing. In Mathematica, the Stand Structural Complexity Index (SSCI) was computed: point clouds were split into azimuthal sectors (0.14°), paired across hemispheres to form 1280 cross-sectional polygons. For each polygon, shape complexity (fractal dimension via modified perimeter-to-area ratio) was computed; mean fractal dimension per scan was then scaled by the natural logarithm of the effective number of layers (ENL), derived via the inverse Simpson index applied to 1 m vertical bins, yielding SSCI per scan. Canopy height per scan was the difference between max and min z; canopy openness was computed via stereographic projection with a 60° opening angle. Climate and soil data: Climate variables (1971–2000) were extracted from WorldClim2 at 30 arcsecond (~1 km) resolution: mean annual precipitation (MAP), precipitation seasonality (coeff. of variation), mean annual temperature (MAT), solar radiation (kJ m^2 day^-1), and mean growing season temperature (growing season defined by ≥5 °C months in temperate/boreal, or months with precipitation > 0.5× PET in tropical/subtropical). PET from Global Aridity and PET database. Projected 2070 climate from WorldClim2 was used for discussion of climate change impacts. Soil variables: soil nitrogen and cation exchange capacity (SoilGrids); field capacity (soil water holding capacity) from Regridded Harmonized World Soil Database v1.2. Soil variables were aggregated to weighted means for 0–100 cm depth. Statistical analysis: Plot-level scan indices and basal area were aggregated to plot means, then to site means with standard errors. Linear regression and linear mixed-effects models (nlme) tested relationships between SSCI (and canopy height, openness, basal area) and single or combined climate/soil predictors. Collinearity was controlled by excluding predictor combinations with |r| ≥ 0.7. Models required all predictors p < 0.05, were ranked by AICc; automated selection (MuMIn) cross-checked manual selection. Model robustness was evaluated via leave-one-out cross-validation (caret) and by excluding entire biomes to assess performance (R², RMSE). Spatial autocorrelation in residuals was tested using Moran’s I (spdep). Global mapping of potential structural complexity: A 50 km global sampling grid (within sampled biomes and forest/woodland ecoregions per Olson et al.) was generated and filtered to forest/woodland ecoregions (excluding steppe/tundra/meadow/grassland transitions). For each point, MAP and precipitation seasonality from WorldClim2 were used with the best-fit model to predict potential structural complexity (SSCI_pot). Outliers outside the 95% kernel density of MAP-seasonality (per biome) were removed; 21,581 points remained. Differences among biomes were tested (one-way ANOVA, Tukey HSD). Latitudinal patterns were modeled using generalized additive models (mgcv). A 30 arcsecond raster was produced (SAGA GIS) to map SSCI_pot for all eligible pixels, truncating predictions beyond observed SSCI range to ≤2 or ≥9. Uncertainty maps (95% CI) were generated; regions outside climatic range or with different field capacity than study sites were flagged. Hotspot overlap with biodiversity was assessed by combining SSCI_pot with ecoregion-level plant diversity data.

• Water availability dominates: SSCI was strongly correlated with mean annual precipitation (MAP; R² = 0.66), water balance (MAP–PET; R² = 0.71), precipitation seasonality (R² = 0.72), and soil water holding capacity (field capacity; R² = 0.61). Solar radiation (R² = 0.29) and soil nitrogen (R² = 0.45) were weaker; MAT and growing season temperature and cation exchange capacity were not significant as single predictors.
• Best predictive model: Multiple linear regression with MAP + precipitation seasonality explained 89.4% of SSCI variation across biomes (R² = 0.89; AICc = 48.02; RMSE = 0.62), outperforming alternatives (ΔAICc ≥ 9.77). Residuals showed no spatial autocorrelation (Moran’s I ≈ 0.006, p = 0.19).
• Model robustness: Leave-one-out cross-validation achieved R² = 0.86 and RMSE = 0.71. Excluding entire biomes maintained high performance (R² 0.82–0.91; RMSE 0.58–0.64), with the largest reduction when excluding (sub-)tropical savannas/woodlands (R² = 0.82).
• Other structural attributes: No significant climate/soil relationships with canopy height or basal area were found. Canopy openness decreased exponentially with increasing MAP and increased with precipitation seasonality.
• Global potential structural complexity (SSCI_pot): Using 1971–2000 climate, SSCI_pot varied by biome: (sub-)tropical moist broadleaf forests mean 6.79; temperate broadleaf 5.75; temperate conifer 5.15; boreal 4.99; (sub-)tropical savannas and woodlands 4.54. Latitudinally, SSCI_pot peaks at the equator, declines toward the tropics (~23°), rises again to mid-latitude peaks (~40° N/S), then declines toward boreal latitudes.
• Hotspots: Very high SSCI_pot (≥9) in Australasian, Indomalayan, and Neotropical moist forests: Napo and Chocó-Darién (western Amazonia), Borneo and Sumatra lowland rainforests, and New Guinea lowland rainforests. Temperate hotspots include Valdivian temperate forest (southern Chile), Northern Pacific Alaskan coastal forests, and Tasmanian temperate rainforests. Lower SSCI_pot in woodland ecoregions of (sub-)tropical grasslands/savannas/shrublands (e.g., Angolan Mopane, Zambezian Baikiaea, Miombo woodlands).
• Biodiversity linkage: Hotspots of high SSCI_pot coincide with hotspots of plant diversity; biome-level SSCI_pot differences mirror patterns in vascular plant species richness.
• Mapping coverage: Predictions were confined to sampled forest biomes, excluding tropical/subtropical dry forests, mangroves, and Mediterranean forests/woodlands to avoid extrapolation.

The findings support that global forest structural complexity is primarily governed by water availability and its temporal distribution (seasonality). Mechanistically, wetter climates support greater functional diversity and a broader range of plant strategies (physiological tolerance hypothesis), fostering complementarity in crown architectures and shade-tolerant species that promote vertical stratification and canopy packing, thereby increasing 3D structural complexity. Water availability also constrains potential tree height (hydraulic limitation hypothesis), expanding the vertical space that can be occupied where precipitation is higher. Thus, climate influences structural complexity through multiple interacting pathways: functional diversity, shade tolerance versus other stress tolerances, and maximum attainable tree size. Disturbance regimes modulate realized complexity: small-scale gaps can enhance understory development and complexity, whereas large-scale fires or storms can simplify structure or delay development of complexity. The study minimized disturbance effects by focusing on late-successional primary forests to isolate climatic controls, but acknowledges that spatial variability in structure is also shaped by disturbance histories and fine-scale edaphic variation. The significant role of soil water holding capacity indicates within-site heterogeneity matters. Predicted shifts in precipitation patterns and increased disturbance frequency/intensity under climate change suggest spatially contrasting future changes in forest structural complexity across biomes. For example, increased drought and seasonality could reduce functional diversity and possible tree size, increasing mortality and reducing complexity in some temperate and tropical regions; in boreal forests, increased wildfire activity may offset potential gains from warmer, wetter conditions. Management and restoration implications are substantial: mapping potential structural complexity provides a benchmark for assessing structural intactness, guiding conservation prioritization, and evaluating restoration effectiveness. Integration with forthcoming satellite LiDAR (GEDI) could enable comparisons between potential and actual complexity, enhancing monitoring of forest management, degradation, and recovery.

Using terrestrial LiDAR-derived measurements from primary forests across major biomes, the study demonstrates that mean annual precipitation and precipitation seasonality explain nearly 90% of the global variation in forest structural complexity. It delivers a global map of potential forest structural complexity (SSCI_pot), revealing distinct biome- and latitude-dependent patterns and identifying hotspots that align with centers of plant diversity. This benchmark can inform sustainable, complexity-oriented forest management and restoration, and serve as a reference to assess structural intactness where primary forests have been lost. Future research should expand biogeographic and climatic coverage, incorporate disturbance regimes and fine-scale edaphic heterogeneity into models, and leverage satellite LiDAR (e.g., GEDI) to map actual structural complexity and compare it with potential levels to better predict biodiversity and ecosystem function responses to climate change.

• Sampling focused on primary, late-successional forests to isolate climatic controls; results may not capture dynamics in managed or frequently disturbed forests.
• Biogeographic, climatic, and edaphic coverage is incomplete; some forest biomes (e.g., tropical/subtropical dry forests, mangroves, Mediterranean forests/woodlands) were excluded to avoid extrapolation.
• Soil data resolution (typically 250 m to 1 km) limits detection of fine-scale edaphic variability; field capacity estimates were derived from nearest profiles and aggregated over 0–100 cm.
• Disturbance regimes (frequency, intensity, scale) were not explicitly modeled, though they strongly influence realized structural complexity.
• Some inconsistencies in plot totals across text sections indicate potential reporting typos; however, site-level analyses (n = 20 sites) underpin the statistical models.
• Mapping uncertainty increases in regions outside the studied climatic range or with soil conditions differing from study sites; these areas were flagged, and predictions truncated beyond observed SSCI bounds.