1. Positive relationships between structural diversity and forest productivity have been documented in controlled experiments and early secondary forests, however, negative relationships have also been observed in late successional forests. The mechanisms causing observed relationships between structural diversity and productivity are not well established, but complementarity among crowns and asymmetric competition have been suggested.

2. We used LiDAR and repeated census data to examine relationship between canopy structural diversity and productivity in nine 1-ha subtropical forest plots along a disturbance gradient in southeastern China. We quantified the relative importance of community composition, species diversity, canopy structural diversity, leaf area index (LAI), and disturbance regime on productivity using piecewise structural equation modelling. We also tested how vertical leaf area distribution effected productivity.

3. Contrary to many prior observations, we found a negative relationship between canopy structural diversity and forest productivity. The negative effect may stem from asymmetric competition between overstory and understory leaves, leading to a lower leaf area efficiency (i.e., wood production per leaf area). Asymmetric competition was suggested by a negative relationship between understory leaf area and total productivity. Changes in community composition over the disturbance gradient, but not species diversity, had a significant effect on productivity.

4. Synthesis. Our study suggests that leaf area and canopy structural diversity have contrasting effects on productivity in this subtropical forest, and this need to be considered when estimating rates of carbon sequestration in secondary forests. The negative effect of asymmetric competition on productivity is comparable to that of the shift in species composition over succession, highlighting the role of canopy structural diversity in shaping forest productivity.

We established nine 1-ha (100m × 100m) permanent plots in GNNR that included twice-logged forest, once-logged forest and old-growth forest (three 1-ha plots for each type). We divided each 1-ha plot into 16 25m×25m quadrats and calculated coarse woody productivity in each quadrat using all individuals with DBH larger than 5cm. We estimated the above-ground biomass (AGB) of each individual stem using species-specific allometric equations for some species (Lin et al., 2015) and the non-species specific allometric equation of Chave et al. (2005) for other species (detailed descriptions can be found in Lin et al. (2015)). The coarse woody productivity (CWP) per quadrat was quantified as the sum of the increment of individual alive and the AGB of recruits divided by the census length in years following the methods of Chisholm et al. (2013). LiDAR data of the ~3 km² of the GNNR were collected in September 2016 using a UAV-borne LiDAR system with a Velodyne Puck VLP-16 sensor, yielding a wavelength of 905 nm and an average point density of 19.7 points m⁻² from approximately 200 m above ground. More details about LiDAR data acquisition can be found in Guo et al. (2017). We clipped each nine 1-ha plot with a 20m buffer from the entire LiDAR data. To eliminate the effects of point density at different plots, we subsampled plots with higher point density to ensure all plots had as similar point density as possible. We used first returns to calculate LAI, canopy height, and several indices representing canopy structural complexity such as entropy and standardized deviation of canopy height (height.sd) at a spatial grain of 25 m × 25 m.