Study sites

This study was conducted in both the Bawangling Nature Reserve (BWL, 18°52′ – 19°12′ N, 108°53′ – 109°20′ E) and the Diaoluoshan Nature Reserve (DLS, 18°40′ – 18°49′ N, 109°45′ – 109°57′ E) in Hainan Island, China (Fig. 1). BWL is ca. 500 km², with an altitude range of ca. 100 – 1654 m. The mean annual temperature is 23.6°C, and the precipitation range is 1676 – 2553 mm. DLS is ca. 380 km², with an altitude range of ca. 50 – 1499 m. It has a mean annual temperature of 24.4°C, and the precipitation range is 1870 — 2760 mm. Both the Reserves are characterized by tropical monsoon climates, with a distinct wet season from May to October and a dry season from November to April. Soil types in both places are mountain red soil and mountain yellow soil. Vegetation, from low to high altitude, is tropical lowland rainforest, tropical montane rainforest, tropical montane evergreen forest and tropical montane dwarf forest [28].

Fig. 1.

A map showing the locations of the seventeen 0.25-hm² plots in Bawangling Nature Reserve (BWL) and Diaoluoshan Nature Reserve (DLS).

Data Collection

We established 8 plots in the forests in BWL in areas where shifting cultivation had taken place, and 9 plots in DLS after selective logging had occurred. Each plot was 0.25 hm². We determined the vegetation and land use history for each plot using forestation materials provided by the local Forestry Bureau. We then divided the plots into three successional stages comprising young forest (about 30 years old: BWL1, BWL2, BWL3, DLS1 and DLS2), middle-aged forest (about 60 years old: BWL4, BWL5, BWL6, DLS3, DLS4, DLS5, DLS6 and DLS7) and old-growth forest (more than 120 years old: BWL7, BWL8, DLS8 and DLS9) (See fig. 1). The elevation of plots ranged from 300 m to 700 m in BWL and DLS, while the gradient ranged from 12° to 21°. Before shifting cultivation and selective logging had taken place, the primary vegetation in each site was tropical lowland rain forest, and the dominant species included Vatica mangachapoi and Cyclobalanopsis patelliformis. Each plot underwent natural restoration and there was no evidence of any additional disturbances. Because shifting cultivation had destroyed canopy cover and led to degraded, dry soils, both the young and middle-aged forests in BWL were dominated by drought-resistant deciduous trees, such as Syzygium cumini and Liquidambar formosana. The dominant species in old-growth forests in BWL, however, were evergreen trees such as Vatica mangachapoi, Koilodepas hainanense, and Cyclobalanopsis patelliformis because soil moisture had improved as the forest community recovered. The dominant species in plots in DLS, on the other hand, were evergreen trees such as Schefflera heptaphylla and Sarcosperma laurinum in the young forests, Vatica mangachapoi, Gonocaryum lobbianum, Gironniera subaequalis and Maclurodendron oligophlebium in the middle-aged forests, and Vatica mangachapoi and Cyclobalanopsis patelliformis in the old-growth forests.

We measured the height of all individual trees and shrubs (to the height of the highest sucker or shoot) in each of the study plots (excluding those clearly suckering from other trees), with DBH ≥ 1 cm. We used a clinometer to measure from the base to the highest point on a tree. Species names and their growth form were recorded in accordance with Flora of China (English edition:  http://www.efloras.org).

Following standard vertical stratification division classification in tropical forests methods [29], and using Whittaker's Growth Form Classification System [30], the vertical community was divided into four sub-communities. These comprised the ground layer, where plant height was more than 1 m but less than 3 m; the regeneration layer, where plant height was more than 3 m but less than 5 m; the succession layer, where plant height was more than 5 m but less than 10 m; and the canopy layer, where plant height was more than 10 m. Species in each community were grouped, according to their growth forms, into 1) hierarchy species (plants which grow in a layer under canopies but do not appear in the canopy layer when they are mature), 2) regeneration species (plants that grow in low layers but also appear in the canopy layer), and 3) succession species (plants which grow in low layers but can also potentially appear in the canopy layer).

Data Analysis

Hypothesis 1: Were the patterns of species diversity within each vertical community consistent with those of the overall community after shifting cultivation and selective logging?

Species diversity within the overall community and within each of the four vertical communities were measured for the three successional stages using the Observed Species Richness (OSR) system, the Shannon-Wiener Index (SW) and the Inverse-Simpson Index (I.Sim), respectively. The abundance (N) of each species in the overall community and in each of the four vertical ones was measured. Species dominance and evenness of growth within each vertical community were assessed using a Species Rank-Abundance Distribution Curve with data log_e transformed. During the three successional stages differences in species diversity in the overall community and within each vertical community were assessed using a Randomization-Based Unilateral Test.

Hypothesis 2: Were the changes in species turnover between the vertical communities under the canopies and within the canopy communities similar after shifting cultivation and after selective logging?

First we took the Chao-Jaccard and Bray-Curtis dissimilarity coefficients to calculate Non-Metric Multidimensional Scaling (NMDS) in order to assess species dissimilarity in the three successional stages and in the different vertical communities. Next, we disaggregated the species dissimilarity coefficient into one part that showed changes in species composition and another part that shows changes in species abundance between vertical communities under canopies and canopy communities. Calculation of species dissimilarity coefficients is typically based either on binary data or on current species abundance estimations. Errors may appear when the two methods are used to calculate species dissimilarity between different vertical communities, as more species can be counted in low layers while higher layers are less accessible or even impenetrable. For example, the Sørensen and Bray-Curtis dissimilarity coefficients may count the species in low layers but, because of problems of access, cannot properly calculate the number of the species in the higher layers. This results in an overestimate of the dissimilarity between the layers. The Sørensen and Bray-Curtis dissimilarity coefficients used in this study, therefore, were divided into one part caused by constraints on species growth and another part caused by species replacements. The disaggregation processes were listed as follows.

The formulas for calculating the Sørensen dissimilarity coefficient (β_S, [31]) and the Bray-Curtis dissimilarity coefficient (β_B, [32]) were ${\beta }_S=\frac{b+c}{2a+b+c}$ and ${\beta }_B=\frac{\sum Bi+\sum Ci}{2\sum Ai+\sum Bi+\sum Ci}$, respectively. ‘a’ indicated the OSR common to two plots while ΣAi indicated the sum of N of the species common to two plots. ‘b’ indicated the OSR that appeared in the first plot but not in the second one, while ΣBi indicated the sum of N of the species that appeared in the first plot but not in the second one. ‘c’ indicated the OSR which appeared in the second plot but not in the first, while ΣCi indicated the sum of N of species which appeared in the second plot but not in the first. For the formulas, we first defined ‘a’ as the OSR of regeneration species and ΣAi as the sum of N of regeneration species in this study. Second, we let b = b₁ + b₂, ΣBi = ΣBi₁ + ΣBi₂, and defined b₁ as the OSR of succession species and ΣBi₁ as the sum of N of succession species. We also defined b₂ as the OSR of hierarchy species and ΣBi₂ as the sum of N of hierarchy species. Third, we defined ‘c’ as the OSR of species which appear in high layers but not in lower ones, and ΣCi as the sum of N of species which appear in high layers but not in low ones. Finally, the Sørensen dissimilarity coefficient was divided into β_S1 and β_S2 using binary data, and the Bray-Curtis hierarchy dissimilarity coefficient was divided into β_B1 and β_B2 using community abundance data. β_S = β_S1 + βS₂, and β_B = β_B1 + β_B2, β_S1 indicated the dissimilarity in species composition and abundance caused by species replacements, and β_S2 showed dissimilarities in composition and abundance caused by growth form restriction.

According to the definition above, we calculated β_S1 and β_B1 between vertical communities under canopies (the ground layer, the regeneration layer and the succession layer) and canopy communities (the canopy layer) for each of the plots, in order to explore dissimilarities in both species richness and community abundance during forest succession. The β_S1 and β_B1 formulas were as follows:

We used presence-absence data to establish plot × species matrices showing species in the vertical communities under canopies and canopy communities, respectively, to calculate β_S1. The calculation not only included dissimilarities in species richness between the vertical communities under canopies and canopy communities of the same plots, but also included dissimilarities in species richness between the vertical communities under canopies of one plot and the canopy communities of the rest of the plots. For the plot in the old-growth forests, β_S1 only referred to the dissimilarities in species richness between the vertical communities under canopies and canopy communities of the same plots. Also, we used abundance data to create matrices showing species in the vertical communities under canopies and canopy communities, respectively, in order to calculate β_B1. Differences in β_S1 and β_B1 among the three successional stages after shifting cultivation and selective logging had occurred were assessed using Randomization-Based Unilateral Tests. All data were analyzed in the statistical package R version 3.2.3.

Patterns of overall species diversity after shifting cultivation and selective logging respectively.

Where shifting cultivation had occurred, N of the old-growth forest was significantly lower than young and middle-aged forests (Randomization-Based Unilateral Test, P = 0.002, P = 0.03; Appendix 1). However, the OSR and SW of the former were significantly higher than the latter (Randomization-based unilateral test, OSR, P = 0.002, P = 0.04; SW, P = 0.01, P = 0.03; Appendix 1), although I.Sim in old-growth forests was only significantly higher than in middle-aged forests (Randomization-Based Unilateral Test, P = 0.03; Appendix 1). In the plots where selective logging had occurred, N in middle-aged forests was significantly higher than in old-growth forests (Randomization-Based Unilateral Test, P = 0.02), and there were no significant differences in species diversity among the three recovery stages (see Appendix 1).

The OSR, SW and I.Sim of each forest community after shifting cultivation were significantly lower than after selective logging at each the recovery stages, while N of the former was significantly higher than the latter (Randomization-Based Unilateral Test, P < 0.001).

Patterns of species diversity in the four vertical communities after both shifting cultivation and selective logging.

For each of the four vertical communities in old-growth forests species evenness was higher while species dominance was lower, compared to growth in young and middle-aged forests after shifting cultivation had occurred (Fig. 2). The rank of species abundance in the three communities under canopies was young forest > middle-aged forest > old-growth forest, and the rank of abundance of canopy communities was middle-aged forest > young forest > old-growth forest. N of each vertical community in old-growth forests was significantly lower than in both young and middle-aged forests, and N of canopy communities in the young forests was significantly lower than in the middle-aged forests (Appendix 1 and 2; Fig. 3). The OSR and SW of each vertical community in old-growth forests, however, were generally significantly higher than in young and middle-aged forests (Appendix 1 and 2; Fig. 3). The first axis of non-metric multidimensional scaling (NMDS) showed changes in species compositions in the three recovery stages after shifting cultivation, while the second axis showed changes in composition among the four vertical communities (Chao-Jaccard: stress = 0.11; Bray-Curtis: stress = 0.11, Fig. 4 a and c).

Fig. 2.

Species rank-abundance distribution curves showing species dominance and evenness in each vertical community after shifting cultivation (SC) and selective logging (SL), respectively, have occurred.

Fig. 3.

Comparisons in Observed Species Richness (OSR), Number of individuals (N), Shannon Wiener's diversity index (SW) and Inverse Simpson's diversity index (I.Sim) in ground layer, regeneration layer, succession layer and canopy layer among young forest (Y), middle-aged forest (M) and old-growth forest (O) after shifting cultivation (in white) and selective logging (grey).

Fig. 4.

Non-Metric Dimensional Scaling (NMDS) plots showing species composition in ground layer (A, ○), regeneration layer (B, •), succession ayer (C, •), and canopy layer (D, •) in each successional stage after shifting cultivation and selective logging have occurred. BWL1-8 and DLS1-9 show the plots after shifting cultivation and selective logging, respectively.

Where selective logging had occurred, abundance did not differ in the three successional stages of the four vertical layers (Fig. 2). N of the canopy communities in old-growth forests was significantly lower than in the young and middle-aged forests (Appendix 1 and 2; Fig. 3). No differences in the OSR, SW and I.Sim of each vertical community were found in the three recovery stages (Appendix 1 and 2; Fig. 3). The first axis of NMDS showed changes in species composition among the four vertical communities (Chao-Jaccard: stress = 0.22; Bray-Curtis: stress = 0.21; Fig. 4 b and d). But the second axis did not show any patterns in species composition in the different stages.

Patterns of the βS1 and βB1 after shifting cultivation and selective logging

Both the β_S1 and β_B1 between young and middle-aged forests were lower than those between middle-aged and old-growth forests after shifting cultivation had occurred (Fig. 5; Appendix 3). Additionally, the β_S1 and β_B1 between both the young and middle-aged forests and the old-growth forest were greater than 0.77 and 0.93, respectively. However, neither of these dissimilarity coefficients differed between young and old-growth forests after selective logging (Fig. 5; Appendices 3 and 4). There were also no significant differences in β_S1 and β_B1 between the ground layer in middle-aged forests and the canopy layers in both middle-aged and old-growth forests after selective logging (Appendix 4).

Fig. 5.

Plots showing the changes in β_S1 and β_B1 between three vertical communities under canopies and canopy communities after shifting cultivation and selective logging, respectively. BWL1-8 and DLS1-9 show the plots where the vertical communities under canopies were after shifting cultivation and selective logging, respectively, and B1-8 and D1-9 show the plots where the canopy communities were.

Overall species diversity increases after shifting cultivation but does not differ in the three recovery stages after selective logging.

While overall species diversity increases with community recovery after shifting cultivation, species abundance decreases. This pattern can be explained by the niche differentiation hypothesis. For example, in young and middle-aged forests, pioneer species with powerful sprouting ability [33, 34] rapidly colonize abandoned fields [35], so that although there is an abundance of growth, species diversity is limited. Because of intense competition, plant species in old-growth forests make more efficient use of ecosystem resources by reducing the abundance of their growth in favor of quality and richness. Hence, a large number of species may be more efficient in occupying niches, but niche differentiation stabilizes the coexistence of species [36]. On degraded soils in South China, the succession process is directly related to soil nutrition [37]. Environmental conditions become more heterogeneous during the late stages of succession (that is, in old-growth forests) and create more diverse opportunities for seedling and sapling growth than in previous stages. Thus, old-growth forest is associated with an increase in species diversity in all successional stages. These findings reinforce the point that diversity can increase asymptotically through succession [38] and further demonstrate that diversity increases asymptotically not only in primary succession processes [39] but also in secondary succession processes after shifting cultivation. We found no differences in overall species diversity after selective logging across the three successional stages. One possible reason is that the biological legacy (the seedlings and saplings) left after logging can quickly develop into canopy communities during the forest recovery process. Furthermore, most of the seed bank is usually well preserved [18], and seeds germinate quickly in vertical communities under canopies in young and middle-aged forests.

The fact that overall species diversity at each recovery stage after shifting cultivation is significantly lower than after selective logging shows that the recovery rate of species diversity after selective logging is faster than after shifting cultivation. A possible reason for this is that shifting cultivation in the tropical lowland rainforest of BWL not only destroys plants and propagules on the forest floor, but also damages the seed bank in the soil through repeated burning. The seed source for species recovery therefore comes mainly from the dispersal of seeds from surrounding forest plants, with a correspondingly small recovery rate [40]. In the case of selective logging in the tropical lowland rainforest in DLS, however, stumps, saplings and seedlings on the ground and seeds in the soil are well preserved. These propagules are sufficient for relatively rapid forest recovery, and have a high recovery rate [14, 18]. Our study therefore proves that anthropogenic disturbances (shifting cultivation and selective logging) influence the recovery rate of forest communities differently (Appendix 1): Secondary forests will take longer to return to climax forest after shifting cultivation than after selective logging [25, 41].

Patterns of species diversity in vertical communities are not consistent after shifting cultivation and after selective logging.

We found that after shifting cultivation changes in species abundance in the three vertical communities under canopies are consistent with that of the overall community (Fig. 2; Appendix 2). From early to late stages during community recovery the canopy density increases while understory light decreases. Accordingly, the competition for light among the species in vertical communities under canopies increases, and thus the density of plant individuals in these communities decreases [42], which leads to a decrease in species abundance. The evidence is that species abundance increases from young forests to middle-aged forests and this is probably due to improvements in community environments which favor deciduous and heliophilous evergreen plants that grow up into the canopy communities and enable the number of individual plants to increase. Conversely, the fact that species abundance in the canopy communities decreases from middle-aged to old-growth forests may be due to the effects of competition, which result in stable species coexistence in old-growth forests. In the case of selective logging, species abundance in ground and canopy layers decreases from young to old-growth forests (Appendix 1), which is consistent with patterns of species abundance overall. In young and middle-aged forests, there is a biological legacy left after selective logging. When growing into forest communities, the recruitment of this biological legacy results in an increase in individual plant density. Plants in old-growth forests, however, show a uniform distribution pattern because of the effects of competition (Fig. 2), with a relatively low individual density.

The species rank-abundance distribution curves showed that less abundant species are still widespread in forest communities after selective logging but not after shifting cultivation (Fig. 2). Less abundant species have been proven to prefer special habitats [43]. The high DLS richness in these species suggests that environments recovery rapidly after selective logging, enabling the re-establishment of a diverse habitat. Our study also shows that rare species are important in structuring communities after selective logging, and reinforces the point that less abundant species play important roles in the natural ecosystem [44].

Hypothesis 2: Were the changes in species turnover between the vertical communities under canopies and the canopy communities similar after shifting cultivation and after selective logging? The species composition in the canopy communities in middle-aged forests were similar to that of the vertical communities under canopies in young forests (Fig. 5; Appendix 3), demonstrating that the species in the canopy community in a late recovery stage have usually been regenerated from an earlier stage [45, 46]. In the plots recovering from shifting cultivation in BWL, we also found that the dissimilarity coefficients between the three vertical communities under canopies in young and middle-aged forests and the canopy communities in old-growth forests were more than 0.77 and 0.93 (Fig. 5) respectively. This suggests that there were great differences in species composition between early and middle recovery stages and late stages.

Once tropical lowland forests have been degraded by shifting cultivation in BWL they are classified as 'transformed monsoon rainforests' until they become restored to climax forests, with some deciduous species in the communities [47]. In this study, the young and middle-aged forests were transformed monsoon rainforests, and were dominated by Syzygium cumini and Liquidambar formosana. It has been demonstrated that the pattern of species diversity in these degraded forests is closely related to dry environments and poor soil conditions [25]. The dominant species in the old-growth forests are evergreen broad-leaved tree species, such as Vatica mangachapoi, Koilodepas hainanense, and Cyclobalanopsis patelliformis. Through analyzing the patterns of species dissimilarity between old-growth forests and the recovery stages after shifting cultivation, we predict that the present forest communities will succeed according to one of the following two paths, depending on the environmental recoveries: (1) the present forest community can probably recover to climax forest once soil moisture and nutrient levels have recovered, and deciduous species have been replaced by evergreen plant species [13]; or that (2) the present forest community will never recover to climax forest, and will stay in a disturbance climax (i.e. tropical monsoon rainforest) due to difficulties in environmental recovery [48].

Unlike after shifting cultivation, there was a high similarity in species composition between vertical communities under canopies in both young and middle-aged forests and canopy communities in old-growth forests (Fig. 5; Appendix 3 and 4). Seedlings, saplings and small trees in vertical communities under canopies in both early and middle stages have the potential to join canopy communities in late stages. The similarity in species diversity proves that species regeneration after selective logging is unproblematic in DLS. This helps us predict that the present forest communities in DLS can, according to the monoclimax hypothesis, recover to the state of old-growth forests [1]. Also, because the saplings, young trees, stumps and soil seed bank in logged forests are well preserved in DLS, we can predict the rate of forest recovery after selective logging will be higher than after shifting cultivation in BWL [19].