The coordinated impact of forest internal structural complexity and tree species diversity on forest productivity across forest biomes

Abstract

Forest structural complexity can mediate the light and water distribution within forest canopies, and has a direct impact on forest biodiversity and carbon storage capability. It is believed that increases in forest structural complexity can enhance tree species diversity and forest productivity, but inconsistent relationships among them have been reported. Here, we quantified forest structural complexity in three aspects (i.e., horizontal, vertical, and internal structural complexity) from unmanned aerial vehicle light detection and ranging data, and investigated their correlations with tree species diversity and forest productivity by incorporating field measurements in three forest biomes with large latitude gradients in China. Our results show that internal structural complexity had a stronger correlation (correlation coefficient = 0.85) with tree species richness than horizontal structural complexity (correlation coefficient = -0.16) and vertical structural complexity (correlation coefficient = 0.61), and it was the only forest structural complexity attribute having significant correlations with both tree species richness and tree species evenness. A strong scale effect was observed in the correlations among forest structural complexity, tree species diversity, and forest productivity. Moreover, forest internal structural complexity had a tight positive coordinated contribution with tree species diversity to forest productivity through structure equation model analysis, while horizontal and vertical structural complexity attributes have insignificant or weaker coordinated effects than internal structural complexity, which indicated that the neglect of forest internal structural complexity might partially lead to the current inconsistent observations among forest structural complexity, tree species diversity, and forest productivity. The results of this study can provide a new angle to understand the observed inconsistent correlations among forest structural complexity, tree species diversity, and forest productivity.

Graphical abstract

1. Introduction

Forest structural complexity describes the structural heterogeneity of forest vegetation canopy [1,2]. It can mediate the interactions between trees and abiotic environments, and has a direct impact on forest biodiversity and carbon storage capability [3,4]. Revealing the correlations among forest structural complexity, tree species diversity, and forest productivity is essential for predicting the responses of forest ecosystems to the global climatic change and making management plans to protect and restore forest ecosystems [5,6].

Current consensus is that a higher forest structural complexity can raise tree species diversity and increase forest productivity [7,8], and therefore forest structural complexity has been used as a proxy of tree species diversity [9,10]. However, contradictory observations have been reported, which posed challenges to this assumption (Table S1). For example, van der Sande et al. [11] and Yuan et al. [12] found weak or negative correlations between forest structural complexity and tree species diversity. One potential argument that has been used to explain the observed inconsistent correlations was that environmental conditions can contribute to tree species diversity as well, which may balance out the contribution of forest structural complexity [11]. Despite the evident contribution of environmental conditions to tree species diversity, these studies usually used canopy horizontal heterogeneity (e.g., canopy cover) and vertical heterogeneity (e.g., mean height) to represent forest structural complexity [13], while the internal information of forest canopy has been often neglected.

Although forest vertical structural complexity attributes have been reported to have stronger correlations with tree species diversity and productivity [14], they are both largely controlled by trees dominating the upper canopy, and may neglect the contribution of lower canopy vegetation [15]. Forest internal structural complexity that incorporates both vertical and horizontal structural complexity of the forest canopy (Fig. S1) [16], instead, is designed to depict the vegetation structural heterogeneity within forest canopies [17,18]. Here, we hypothesized that forest internal structural complexity might have tight corrleations with tree species diversity and forest productivity, and it should be evaluated separately to better understand the correlations among forest structural complexity, tree species diversity, and forest productivity.

Additionally, forest structural complexity and tree species diversity are scale-dependent parameters [19]. Studies have shown that forest structural complexity and tree species diversity varied significantly with the size of sampling plots, and an optimal plot size should be investigated to better describe forest structure and species heterogeneities [19]. Therefore, we hypothesized that the correlation between forest structural complexity and tree species diversity might be scale dependent as well. However, current studies mostly focused on a single spatial scale in temperate forests. Cross-scale and cross-biomes correlations between forest structural complexity and tree species diversity have been rarely investigated. One of the major obstacles is the lack of accurate and efficient forest structural complexity measurement tools for large-scale studies [20].

Recent advances in light detection and ranging (lidar) provide a novel means to measure forest structure accurately and efficiently, which has the potential to help address this gap. Three-dimensional lidar data have been successfully used to quantify horizontal and vertical structural attributes [21,22], and have been shown to be connected to reveal forest internal structural information [17,23]. In this study, we collected unmanned aerial vehicle (UAV) lidar data and field inventory data from three forest biomes across China: a temperate needleleaf and broadleaf mixed forest in Changbai Mountain (CBS) (25 ha), a subtropical evergreen broadleaf forest in Gutian Mountain (GTS) (24 ha), and a tropical rainforest in Xishuangbanna (XSBN) (20 ha). Three kinds of structural complexity attributes including horizontal, vertical, and internal structural complexity attributes were calculated from the UAV lidar data. Ultimately, we aim to incorporate these lidar-derived structural complexity attributes with field measurements to test the abovementioned hypotheses by addressing the following three questions. (1) How do forest horizontal, vertical, and internal structural complexity attributes correlate to tree species diversity across forest biomes? (2) How do the correlations between forest structural complexity and tree species diversity vary with the spatial scale? (3) Which is the forest structural complexity attribute having a coordinated contribution with tree species diversity to forest productivity?

2. Data and methods

2.1. Study area

This study was conducted at three forest sites (i.e., the CBS site, the GTS site, and the XSBN site) across a large latitudinal gradient of China with significantly distinct climate conditions (Fig. 1 and Table S2). All three sites belong to the Chinese Forest Biodiversity Monitoring Network, which were selected in undisturbed natural forests for long-term forest biodiversity monitoring and investigation [24]. The CBS site is a mixed forest located in northeastern China with a size of 25 ha (500 m × 500 m). It has the lowest mean annual temperature (MAT) of 3.6 °C and mean annual precipitation (MAP) of 700 mm among the three study sites. The GTS site is a subtropical evergreen broadleaf forest with a size of 24 ha (400 m × 600 m), and has a MAT of 15.3 °C and a MAP of 1963.7 mm. The XSBN site is a rainforest in southwestern China with a size of 20 ha (400 m × 500 m), and has a MAT of 21.8 °C and a MAP of 1963.7 mm. The dominant tree species of each study site are listed in Table 1.

Fig. 1.

The location and forest type of the three study sites. CBS, GTS, and XSBN represent the Changbai Mountain site, Gutian Mountain site, and Xishuangbanna site, respectively. The three elliptical subfigures are unmanned aerial vehicle (UAV) light detection and ranging (lidar) profile examples at the three study sites. The background forest type information is derived from the updated vegetation map of China (1:1000,000) [25]. The map was drawn based on the standard base map GS(2020)1044 provided by the National Platform for Common Geospatial Information Service.

Table 1.

Summary of the size, climate, terrain, and dominant tree species information of the three study sites. CBS, GTS, and XSBN represent the Changbai Mountain site, Gutian Mountain site, and Xishuangbanna site; MAP and MAT represent mean annual precipitation and mean annual temperature; and Pko., Tam., Qmo., Fma., Cey., Ssc., Pma., Pch., Pke., and Gac. represent Pinus koaiensis, Tilia amurensis, Quercus mongolica, Fraxinus mandshurica Rupr, Castanopsis eyrie, Schima superba, Pinus massoniana, Parashorea chinensis, Pittosporopsis kerrii, and Garcinia cowa.

Site	Area (ha)	MAP (mm)	MAT ( °C)	Altitude (m)	Forest type	Dominant tree species
CBS	25	700	3.6	801.5	Mixed forest	Pko., Tam., Qmo., Fma.
GTS	24	1963.7	15.3	580.6	Subtropical evergreen broadleaf forest	Cey., Ssc., Pma.
XSBN	20	1493	21.8	580	Tropical rainforest	Pch., Pke., Gac.

2.2. Field data

Two field campaigns were conducted in 2009 and 2014 at the CBS site, 2010 and 2015 at the GTS site, and 2007 and 2012 at the XSBN site. For each field campaign, study sites were first divided into 20 m × 20 m plots, and the diameter at breast height (DBH), species, and relative location of each tree with a DBH ≥ 1 cm were collected and recorded in each plot. Two tree species diversity indices, i.e., richness and evenness, were then calculated from the most up-to-date field campaigns (i.e., 2014, 2015 and 2012 field campaigns of the CBS, GTS, and XSBN sites, respectively). Here, we calculated tree species richness and evenness at three spatial scales (i.e., 20 m × 20 m, 40 m × 40 m, and 100 m × 100 m) to evaluate the scale effect of the correlations among structural complexity, tree species diversity, and forest productivity (Fig. 2). Tree species richness was defined as the number of species within a forest stand, and species evenness was estimated using the Pielou's J evenness index (the ratio of the observed Shannon's diversity of a stand to its maximum value with the same number of species) [26]. Besides tree species diversity indices, we also calculated aboveground biomass from tree species information and DBH measurements of each field campaign using allometric equations at the abovementioned three spatial scales (Fig. 2). Forest coarse woody productivity (CWP) at the CBS, GTS and XSBN sites was calculated as the difference between the two aboveground biomass estimations divided by the time interval of the two field campaigns.

Fig. 2.

Illustration of (a) field data processing, (b) UAV lidar data processing, and (c) forest structural complexity attribute extraction.

2.3. UAV lidar data

The UAV lidar data of the CBS, GTS, and XSBN sites were collected in August 2015, September 2016, and March 2018 using a GreenValley International LiHawk system, respectively. Although there are time differences between UAV lidar data acquisitions and field campaigns, especially for the XSBN site, it should not have significant influence on this study considering the fact that the three study sites are conserved mature forests. The system is equipped with a RIEGL VUX-1 UAV laser scanner, which has a maximum ranging distance of 1000 m, and provides high-speed data acquisition capability (550 kHz) using a narrow near-infrared laser beam. The nominal horizontal positioning accuracy is 10 cm, and the nominal vertical positioning accuracy is 5 cm. The flight height above ground level varied from 70 m to 300 m considering both tree height and terrain slope, and flight speed was around 15 km/h (Table S2).

The collected UAV lidar data of each study site were then pre-processed following the same protocol, including denoising, filtering, and normalization (Fig. 2). The denoising step aimed to remove noise points in the point clouds. Here, we used the outlier removal tool integrated in the LiDAR360 (GreenValley International Inc., Beijing, China) software, which identified noise points by examining whether the distance of a point to its n neighboring points was larger than a threshold of avg.+n × std. (where avg. and std. were the average and standard deviation of the distances between points and their neighbors, and n was set to 5 in this study). The filtering step aimed to classify ground points and generate a digital terrain model (DTM) from the ground points. An improved progressive triangulated irregular network densification filtering algorithm integrated in the LiDAR360 software was used to extract ground points [27], and a DTM of 5 cm resolution was interpolated using the ordinary kriging method for each study site [28]. Finally, the normalization step aimed to remove the influence of terrain elevation on lidar point clouds by subtracting DTM values from the original point heights.

Based on the normalized lidar point clouds, three kinds of structural complexity attributes were extracted, including horizontal, vertical, and internal structural complexity attributes (Table 2 and Fig. 2). The horizontal structural complexity was represented by canopy cover, which was calculated as the percentage of vegetation points to the total number of lidar points [29]. The vertical structural complexity was represented by mean height, which was calculated as the average canopy surface height of a forest stand. Because there have been few studies investigating internal structural complexity, here we calculated eight lidar-derived parameters that have been reported to reflect forest internal structural variations, including kurtosis, skewness and average absolute deviation of lidar point vertical height distribution, median of median absolute deviation, interquartile height distance, interquartile cumulative height distance, coefficient of variation, and canopy relief ratio [30]. Details on the methods of calculating these parameters can be found in Table 2. To simplify the following analyses, we further grouped the lidar-derived internal structural parameters at each spatial scale using the principal component analysis (PCA). As can be seen in Fig. S2, the first component accounted for over 66% of the total variance at all spatial scales. Therefore, we used it to represent internal forest structural complexity in the following analyses, and referred to it as the internal variation index hereafter. To further evaluate the scale effect between structural complexity and tree species diversity, each structural complexity attribute was calculated at the abovementioned three spatial scales as well.

Table 2.

Descriptions of unmanned aerial vehicle (UAV)-light detection and ranging (lidar) derived forest structural complexity attributes.

Name	Abbreviation	Description of structural attributes
Horizontal structural complexity
Canopy cover	CC	The percentage of vertical projection of forest land area
Vertical structural complexity
Mean height	Hmean	The mean value of canopy surface height
Internal structural complexity
Kurtosis	Kur	The kurtosis of lidar point vertical height distribution
Skewness	Skew	The symmetry of lidar point vertical height distribution
Average absolute deviation	AAD	The average absolute deviation of lidar point vertical height distribution
Median of median absolute deviation	MMD	The median of median absolute deviation of lidar point heights
Interquartile height distance	IHQ	The height of 75% the lidar points minus the height of 25% the lidar points in a statistical unit. Note that lidar points need to be sorted from low to high first.
Interquartile cumulative height distance	ICHQ	The cumulative height of 75% the lidar points minus the cumulative height of 25% the lidar points in a statistical unit. Note that lidar points need to be sorted from low to high first.
Coefficient of variation	CV	STDheight/Hmean, where STDheight and Hmean are the standard deviation and mean of height in a statistical unit.
Canopy relief ratio	CRR	(Hmean- Hmin)/ (Hmax- Hmin), where Hmean Hmax and Hmin are the mean, maximum and minimum heights in a statistical unit.

2.4. Statistical analyses

Correlations among structural complexity, tree species diversity, and CWP were evaluated by statistical test and the Pearson correlation coefficient (r). Since the spatial scale of 100 m × 100 m generally had the strongest correlations among them (Section 3.2), only the results at the scale of 100 m × 100 m were presented here. Additionally, structural equation modeling (SEM) was further used to quantify the contributions of forest structural complexity attributes and tree species diversity indices to forest productivity. SEM is a multivariate casual modeling method that allows users to quantify the path coefficients in a designed model [31], and has been frequently used in ecological studies to quantify the logical and methodological relationship between correlation and causation. Since the causality between forest structural complexity and tree species diversity is still controversial [32], here we constructed two SEM models by using each of them as the causal agent. The maximum likelihood method was used to calculate the standard path coefficients in each SEM model. The significance of each path coefficient was evaluated by the significance level of α = 0.05. The adequacy of a SEM model was evaluated by the p-value of the chi-square statistics (P), the comparative fit index (CFI), and the standardized root mean square residual (SRMR). If P was larger than 0.05, CFI was larger than 0.9, and SRMR was smaller than 0.08, the model could be accepted statistically [31]. Similar as the abovementioned correlation analyses, both SEM models were built at the spatial scale of 100 m × 100 m in this study. It should be noted mean height was transformed by a logarithm function in the SEM model to avoid the influence of unbalanced scalar ranges between mean height and other forest structural diversity indices.

3. Results

3.1. Correlations among structural complexity, species diversity, and productivity

Overall, the XSBN site had the richest tree species composition, followed by the GTS site and the CBS site, and their differences increased with spatial scale (Fig. 3a). At the spatial scale of 100 m × 100 m, the average species richness of the XSBN site was around twice as high as that of the GTS site, and seven times as high as that of the CBS site (Table S3). In terms of species evenness, the GTS site was the highest, followed by the XSBN site and the CBS site, and their differences became smaller with the spatial scale (Fig. 3b and Table S3). Among the three structural complexity attributes, canopy cover displayed the smallest variations among study sites (Fig. 3c). The distribution of canopy cover was highly concentrated in all three study sites (standard deviation < 1%), and the average canopy cover was all higher than 98% (Table S3). Mean height and internal variation index both increased from the CBS site to the XSBN site in means, and increased from the GTS site to the XSBN site in standard deviations (Fig. 3d,e). Their differences became larger with spatial scale as well. At the spatial scale of 100 m × 100 m, the average mean height and the average internal variation index of the XSBN site were around 44% and 67% higher than those of the GTS site, and their standard deviations were around 331% and 209% higher than those of the CBS site (Fig. 3d and Table S3). Forest productivity increased from the CBS site to the XSBN site (Fig. 3f). At the spatial scale of 100 m × 100 m, the average CWP of the XSBN site was around 30 folds of that of the CBS site, and six folds of that of the GTS site (Fig. 3f and Table S3). Moreover, the standard deviation of CWP at the XSBN site was also around 6–8 times higher than those at the GTS and CBS sites (Fig. 3f and Table S3).

Fig. 3.

Violin plot of tree species diversity, forest structural complexity, and forest coarse woody productivity (CWP) at three study sites. (a-f) are violin plots of tree species richness, tree species evenness, canopy cover, mean height, internal variation index, and CWP at each study site. Note that 20 m × 20 m, 40 m × 40 m, and 100 m × 100 m represent corresponding spatial scales.

Forest structural complexity attributes showed significant correlations with tree species richness, except canopy cover (Fig. 4a). Both mean height and internal variation index were positively correlated with tree species richness, but the correlation of internal variation index (r = 0.85) was higher than mean height (r = 0.61) (Fig. 4a). Variations in tree species evenness had significant correlations with structural complexity attributes as well, except mean height (Fig. 4a). With the increase of canopy cover, species evenness showed a significant decreasing pattern (r = −0.51), while with the increase of internal variation index, species evenness showed a significant increasing pattern (r = 0.36) (Fig. 4a). Forest productivity showed significant positive correlations with mean height and internal variation index, and showed no significant correlation with canopy cover (Fig. 4a). In addition to forest structural complexity attributes, tree species richness showed a strong positive correlation with CWP as well (Fig. 4a). The correlation between CWP and internal variation index (r = 0.89) was higher than that between CWP and mean height (r = 0.84) (Fig. 4a). Despite the apparent cross-biome correlations among forest structural complexity, tree species diversity and forest productivity, large discrepancies still existed within different forest biomes (Fig. 4b–d). At the CBS site, canopy cover was the only factor having a significant correlation (positive) with tree species richness (Fig. 4b). Internal variation index and mean height had no significant correlations with tree species richness, tree species evenness, and CWP, despite the relatively large r for the correlations between internal variation index and tree species richness (r = 0.33) and CWP (r = 0.25) (Fig. 4b). At the GTS site, canopy cover had strong negative correlations with tree species richness and tree species evenness, while internal variation index had strong positive correlations with them (Fig. 4c). Mean height had no significant correlations with tree species richness and tree species evenness (Fig. 4c). At the XSBN site, all forest structural complexity attributes showed no significant correlations with tree species richness, tree species evenness, and CWP, despite the relatively large r for the correlations of internal variation index with tree species richness (r = −0.34) and CWP (r = 0.44) (Fig. 4d).

Fig. 4.

Pearson correlation coefficients between forest structural complexity attributes, tree species diversity indices and CWP at (a) all three study sites, (b) the CBS site, (c) the GTS site, and (d) XSBN site. Note that ^ns, *, **, and *** represent insignificant correlations, significant correlations with a p-value < 0.05, significant correlations with a p-value < 0.01, and significant correlations with a p-value < 0.001, respectively.

3.2. Scale effect of correlations among structural complexity, species diversity and productivity

Correlations among forest structural complexity, tree species diversity, and forest productivity did show a scale effect. The absolute r values of correlations among them all increased with spatial scales, if there were significant correlations (Fig. 5). Specifically, from the scale of 20 m × 20 m to 100 m × 100 m, the absolute r of correlations of mean height and internal variation index with tree species richness increased from 0.29 to 0.61 and from 0.57 to 0.85 (Fig. 5a), the absolute r of correlations of canopy cover and internal variation index with tree species evenness increased from 0.14 to 0.51 and from 0.06 to 0.35 (Fig. 5b), and the absolute r of correlations of mean height and internal variation index with CWP increased from 0.38 to 0.70 and from 0.37 to 0.85 (Fig. 5c). Canopy cover did not have significant correlations with tree species richness and CWP across all spatial scales (Fig. 5a,c), and mean height did not have significant correlations with tree species evenness across all spatial scales (Fig. 5b). Overall, a scale of 40 m × 40 m was the smallest spatial scale suggested to observe the correlation between forest structure complexity and tree species diversity.

Fig. 5.

Inter-site correlations between forest structural complexity attributes with (a) tree species richness, (b) tree species evenness, and (c) CWP at three spatial scales. Note that 20 m, 40 m, and 100 m represent the spatial scale of 20 m × 20 m, 40 m × 40 m, and 100 m × 100 m, respectively, and ^ns, *, **, and *** represent insignificant correlations, significant correlations with a p-value < 0.05, significant correlations with a p-value < 0.01, and significant correlations with a p-value < 0.001, respectively.

3.3. Coordinated contribution of structural complexity and species diversity to forest productivity

To better understand the coordinated contribution of structural complexity and species diversity to forest productivity, the SEM method was further used to quantify the path coefficients among forest structural complexity attributes, tree species diversity indices, and CWP. As can be seen in Fig. 6, both SEM models had a P larger than 0.05, a CFI larger than 0.9, and a SRMR smaller than 0.08, indicating that they could be accepted statistically. In the SEM model using structural complexity as the causal agent, forest structural complexity explained 73.5% of the variations in tree species richness, and 50.3% of the variations in tree species evenness (Fig. 6a). Internal variation index was the only factor had a significant (positive) contribution to tree species richness (Fig. 6a). Both canopy cover and internal variation index had significant contributions to tree species evenness, but the absolute value of the path coefficient for internal variation index was around three times as high as that for canopy cover (Fig. 6a). Canopy cover, mean height, internal variation index, and tree species richness had significant direct positive contributions to CWP, and the contribution of tree species richness was the largest, followed by internal variation index, mean height and canopy cover (Fig. 6a). Moreover, internal variation index was the only forest structural complexity index that had a coordinated contribution with tree species richness to CWP (Fig. 6a). Overall, the coordination of forest structural complexity and tree species diversity explained 91.2% of the variations in CWP (Fig. 6a).

Fig. 6.

The structural equation model among forest structural complexity attributes, tree species diversity indices, and CWP (a) by using forest structural complexity as the causation and (b) by using tree species diversity as the causation. Solid lines in red represent significant positive paths, solid lines in blue represent significant negative paths, and insignificant paths were represented by dotted lines. P, CFI, and SRMR represent the p-value of the chi-square statistics, the comparative fit index, and the standardized root mean square residual, and richness and evenness are tree species richness and tree species evenness in short.

In the SEM model of using tree species diversity as the causal agent, tree species richness had significant positive contributions to mean height and internal variation index; while tree species evenness had significant negative contributions to canopy cover and mean height, and a significant positive contribution to internal variation index (Fig. 6b). Tree species richness and all three structural complexity indices had significant positive correlations with CWP, and the path coefficient of tree species richness was the largest (Fig. 6b). The coordination of forest structural complexity and tree species diversity explained 91.2% of the variations in CWP (Fig. 6b).

4. Discussion

Forest structural complexity can alter the light and water distribution within a forest canopy, and therefore influence tree species diversity [33,34]. Correlations of horizontal forest structural complexity (i.e., canopy cover in this study) and vertical forest structural complexity (i.e., mean height in this study) with tree species diversity have been widely investigated [35,36]. However, forest internal structural complexity (i.e., internal variation index in this study), as a major factor contributing to the light and water distribution within a forest canopy, has been largely neglected. Here, we found that internal variation index is the only factor correlating to both tree species richness and tree species evenness across forest biomes (Fig. 4a). Canopy cover has an insignificant correlation with cross-biome variations in tree species richness (Fig. 4a), which might be caused by the fact that canopy cover, as the representation of horizontal structure complexity, is mainly controlled by trees dominating the upper canopy and is easily saturated (Fig. 3c) [37]. Mean height has an insignificant correlation with tree species evenness (Fig. 4a). Moreover, although it has a significant correlation with tree species richness, it may not be used as a proxy to reflect the magnitude of tree species richness. For example, the mean height at the CBS site is higher than at the GTS site, but its tree species richness is around twice lower than that at the GTS site (Table S3).

Correlations between forest structural complexity and tree species diversity may be slightly different when treating them as the causal agent or the result differently (Fig. 6), suggesting that there may be mutual influences between forest structural complexity and tree species diversity instead of one-way influences [7]. When treating forest structural complexity as the causal agent, internal variation index has significant positive contributions to both tree species richness and evenness (Fig. 6a). In areas that can support to form a internally complex forest stand, different tree species may take advantage of vertical ecological niches to survive and therefore may form a richer and evener tree species composition [38,39]. Canopy cover has a significant negative correlation with tree species evenness (Fig. 6a). In areas with low canopy cover, increasing availability of light resources is helpful for trees to develop horizontal ecological niches and form evener tree species compositions (Fig. 4a) [15,40]. In areas with high canopy cover, large trees may intercept more light in the upper canopy, and lead to size-asymmetric competition [37]. Under this circumstance, large trees can suppress the growth of small trees by reducing their light availability, and may even lead to exclusion of small trees, which therefore may ultimately decrease tree species evenness [36]. When treating forest structural complexity as the result, tree species richness has significant postive contirbutions to both mean height and internal variation index (Fig. 6b), indicating a richer tree species composition may have a taller and internally more complex canopy so that trees can fully take advantage of the vertical space to develop finer ecological niches and therefore compete for the limited light resources [7,40]. Tree species evenness has a significant negative contribution to both canopy cover and mean height (Fig. 6b). In an area with uneven tree species compositions, it might be dominiated by a few tree species, which may therefore reduce the chance of forming a closed and tall canopy (i.e., larger canopy cover and mean height) [41]. Moreover, this effect may also counterbalance the positive contribution of internal variation index to tree spcies evenness. Although tree species evenness still has a positve contiribution to internal variation index (Fig. 6b), its path coefficicent is around four times smaller than that of the SEM model with forest structural complexity as the causal agent (Fig. 6a).

It is believed that forest structural complexity and tree species diversity have a coordinated positive effect on forest productivity, but current observations often break this consensus [42,43]. Here, we argue that these inconsistent observations might be partially caused by the fact that only horizontal and vertical structural complexity attributes were examined. In the SEM model with forest structural diversity as the causal agent, horizontal and vertical structural complexities are found to have insignificant or weak coordinated effects with tree species diversity on forest productivity (Fig. 6a). Although vertical structural complexity has a direct positive contribution to forest productivity (Fig. 4a), it only has a direct positive contribution to forest productivity, suggesting taller forest stands tend to have higher forest productivity regardless of tree species compositions (Fig. 6). Interestingly, although forest internal structural complexity exhibits a strong correlation with forest productivity (Fig. 4a), its direct contribution to forest productivity is low (i.e., small path coefficient in the SEM model) (Fig. 6a). Instead, it shows the tightest coordinated contribution with tree species richness to forest productivity among all three types of forest structural complexity (Fig. 6a). This may also explain the phenomenon that tree species richness has the largest direct path coefficient to CWP in this SEM model (Fig. 6a). The total effect of internal variation index to CWP is the highest among all structural complexity attributes and tree species diversity indices (Fig. 6a). In the SEM model with tree species diversity as the causal agent, we can find a similar result that internal variation index has the tightest coordinated contribution to forest woody productivity (Fig. 6b). Although only three study sites are included in this study, their distinct tree species compositions and structural complexity characteristics can still provide concrete evidence of the positive coordinated effect of forest internal structural complexity and tree species richness to forest productivity. Moreover, although mean height has been widely used to reflect vertical structural complexity, it is still a simplified index without considering canopy height variations [44]. Here, we further used canopy height variations (i.e., standard deviation of canopy height) to replace mean height, and redid the SEM analyses. The results showed that internal variation index was still the factor having tight coordinated contributions with tree species diversity to forest productivity (Fig. S3). This further highlights the importance of internal variation index in understanding correlations among forest structural complexity, tree species diversity, and forest productivity.

Correlations among forest structural complexity, tree species diversity, and forest productivity are scale dependent, which increase with spatial scales in general (Fig. 5). With the increase of spatial scale, more variations in tree species diversity, structural complexity, and forest productivity can be captured within forest stands than between forest stands (Fig. 3), which may stabilize their estimates by reducing the influence of random factors (e.g., stand location, tree species composition) [19]. These stabilized estimates may further reduce the influence of random factors on observing the correlations among them, and therefore lead to higher correlation coefficients. Among the three forest structural complexity attributes, internal structural complexity is the least influenced by spatial scale. Internal variation index has significant correlations with tree species richness and CWP across all three spatial scales (Fig. 5a,c), which further confirms the strong capability of internal structural complexity in depicting tree species diversity across forest biomes. Moreover, a minimum spatial scale of 40 m × 40 m is suggested from the current study to observe the correlations among forest structural complexity, tree species diversity, and forest productivity, which is identical to the suggested optimum spatial scale for quantifying tree species diversity by Kukunda et al. [19].

Despite the apparent cross-biome correlations among forest structural complexity, tree species diversity and forest productivity, large discrepancies may still exist among different forest biomes (Fig. 4). This inconsistency might be related to different tree growth strategies under different forest conditions [36]. For example, the CBS site is a mature temperate mixed forest biome with the lowest MAT and MAP among the three forest biomes (Table 1), which may not support its forest canopy developing into multiple vertical layers, and trees within it may first use the horizontal niches to develop [45]. Therefore, canopy cover has a positive correlation with tree species richness, while internal variation index has no significant correlation with tree species richness at the CBS site (Fig. 4b). Moreover, the CBS site has a strong negative selection effect to suppress the growth of other tree species (Fig. 7) [15,46], which may lead to the negative correlations between mean height and tree spcies diversity indices (Fig. 4b). The GTS site is a relatively young subtropical broadleaf forest with a warm and moist environment (Table 1) that can support the growth of forest canopy into multiple vertical layers, and light resources may become the major limitation of tree growth [24,47]. Correspondingly, a richer tree species composition may lead to a more complex canopy internal structure to compete for the limited light resources (Fig. 4b) [16,48]. Moreover, canopy gaps play a highly important role in the tree regeneration of the GTS site [49,50]. Trees intend to take advantages of canopy gaps to quickly occupy the horizontal niches to form rich tree species compositions [50], and therefore canopy cover tends to have strong negative correlations with tree species diversity indices at the GTS site (Fig. 4c).

Fig. 7.

Bar plots of (a) the total aboveground biomass (AGB) and (b) the total AGB percentages of dominant tree species at the CBS site, and bar plots of (c) the total aboveground biomass (AGB) and (d) the total AGB percentages of dominant tree species at the XSBN site.Tam., Fma., Qmo., Pko., Par., Mez., Gar., Pit., and Bac. represent Tilia amurensis, Fraxinus mandshurica, Quercus mongolica, Pinus koaiensis, Parashorea chinensis, Mezzettiopsis creaghii, Garcinia cowa, Pittosporopsis kerrii, and Baccaurea ramiflora, and others represents all other tree species at the corresponding study site. Note that these bar plots were made at the spatial scale of 100 m × 100 m.

The development of lidar technology provides great opportunities to quantify forest structural complexity in a way that has never been done before. However, there are still some major technical obstacles that have been limiting the next-step investigation of the correlation among forest structural complexity, tree species diversity, and forest productivity. Firstly, UAV lidar technology still has problems to fully penetrate dense rainforests [51], which may influence the accurate estimation of internal structural complexity. At the XSBN site, the percentage of lidar points in the lower canopy zone is much lower than that at the GTS site (Fig. 8). The incomplete vertical profiles, especially the missing of lower canopy information, may lead to an underestimation of internal structural complexity, and therefore may neutralize the contribution of internal structural complexity to tree species diversity (Fig. 4d). Su et al. [52] developed a backpack lidar system with dual laser scanners, which can depict complete vertical canopy profiles in rainforests and show great potentials to solve this problem. Secondly, the rich three-dimensional information provided by lidar point clouds has not been fully used in the quantification of forest structural complexity. Further studies are still needed to investigate how to better quantify forest structural complexity from lidar data.

Fig. 8.

UAV lidar profile examples and their corresponding vertical point density distributions at (a) the CBS site, (b) the GTS site, and (c) the XSBN site. The low point density in lower canopy at the CBS site is caused by the absence of complex under canopy vegetation, and the low point density at the XSBN site is caused by the incomplete penetration of UAV lidar in rain forests.

5. Conclusion

This study investigated the correlations among forest structural complexity, tree species diversity, and forest productivity in three forest biomes with large latitude gradients across China, and UAV lidar technique was adopted to quantify three forest structural complexity attributes, i.e., horizontal, vertical, and internal structural complexity. Three major findings can be concluded. (1) Forest internal structural complexity has stronger correlations with tree species diversity indices than the other two forest structural complexity attributes. The correlation coefficients of internal structural complexity with tree species diversity and forest productivity are larger than those of the other forest structural complexity attributes, and it is the only structural complexity attribute having strong positive correlations with both tree species richness and tree species evenness. Moreover, although correlations among forest structural complexity, tree species diversity, and forest productivity vary with forest biomes, the correlation of coefficients of internal structural complexity are generally larger than those of the other two forest structural complexity attributes. (2) A scale effect can be observed in the correlations among forest structural complexity, tree species diversity, and forest productivity, and their correlations generally become stronger with spatial scale. Forest internal structural complexity is the least influenced by spatial scale. (3) Forest internal structural complexity has a much tighter coordinated contribution with tree species diversity to forest productivity than vertical and horizontal structural complexity, which highlights the important role of long-neglected forest internal structural complexity in mediating tree species richness and forest productivity across forest biomes. We believe that this study provides a new angle to understand the observed inconsistent correlations among forest structural complexity, tree species diversity, and forest productivity, which is beneficial to improving our understanding on forest ecosystem processes and make forest management plans.

Declaration of competing interest

The authors declare that they have no conflicts of interest in this work.

Biographies

Qin Ma is currently a Ph.D. candidate at the State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences. She received her B.S. degree from Huazhong Agricultural University. Her research interests lie in using lidar to understand the role of forest structural complexity in forest ecosystem processes.

Yanjun Su(BRID: 06310.00.75521) is a professor at the State Key Laboratory of Vegetation and Environmental Change, Intitute of Botany, Chinese Academy of Sciences. He received his Ph.D. degree from the University of California, Merced. His-research focuses on the use of lidar and other remote sensing techniques to understand how human activities and global climate change influence terrestrial ecosystems.