Secondary succession of an unmanaged coppice woodland adjacent to late-successional, lucidophyllous forest in western Japan

ABSTRACT The combined effects of management history and ecosystem connectivity make it difficult to predict future dynamics of abandoned and unmanaged ecosystems. In Japan, unmanaged, secondary forests (satoyama) face risk of arrested or diverted succession, due to extensive human influence across the landscape. Proximity to climax forests, which function as seed sources, could determine the course of succession of abandoned satoyama. Here, we investigated spatial/temporal variation of species composition and stand structure of abandoned satoyama adjacent to a mature lucidophyllous forest in warm-temperate Japan to elucidate the course of succession after abandonment. Two study plots were established in the unmanaged, secondary forest at varying distances from the mature lucidophyllous forest. We calculated vegetation similarity indices among the plots to elucidate spatial variation and temporal change of species composition and stand structure and visualized relationships using nMDS (non-metric multidimensional scaling) ordination. Over the past 15 years, species composition and stand structure of the secondary forest have changed following the normal course of succession. This was because shade-intolerant shrubs, such as Rhododendron, were replaced by recruitment of climax species originating from the lucidophyllous forest. However, Quercus serrata (deciduous oak) and shade-intolerant evergreen trees continued to dominate the upper-canopy. Although the adjacent lucidophyllous forest is an effective seed source for recruitment of climax species, it may take several more decades for the secondary forest to reach late-successional composition and structure, due to legacy effects of past management.


Introduction
Traditional agricultural landscapes are rapidly disappearing due to socio-economic changes and modernization (Agnoletti 2007;Dobrovodská et al. 2019).However, effects of past land management practices (legacy effects) can persist for decades to centuries after traditional practices are abandoned, influencing the subsequent trajectory of ecosystem dynamics, such as vegetation succession, carbon stocks, nutrient cycling, etc. (Foster et al. 2003;Hermy and Verheyen 2007).In addition, as the human-domination of the landscape proceeds, natural ecosystems become more fragmented, negatively influencing landscape-scale ecosystem processes, such as metapopulation dynamics and forest succession (Duelli and Obrist 2003;Martín-Queller and Saura 2013;Baiamonte et al. 2015).The combined effects of ecological history and ecosystem connectivity complicates succession and dynamics of abandoned, unmanaged ecosystems (Foster et al. 2003).
Satoyama are secondary forests traditionally managed under coppice forestry to obtain firewood, organic fertilizer and other forest resources in the traditional agricultural landscape of Japan (Takeuchi and Washitani 2003;Morimoto 2011;Yokohari and Bolthouse 2011).Recent research has shown that traditional agricultural management associated with satoyama landscapes maintains a mosaic of forests at various stages of succession (Yamasaki et al. 2000;Katoh et al. 2009;Iwachido et al. 2020).Traditionally, coppice management had prevented succession of satoyama to climax forest (Kobori and Primack 2003) and maintained open forests at early stages of succession dominated by shade-intolerant species (Nagaike et al. 2003;Takeuchi 2010).Traditional agricultural landscapes in Japan are rapidly disappearing (Washitani 2001;Takeuchi 2010).Many satoyama forests in Japan were abandoned in the 1960-70s as firewood and organic fertilizer were replaced by fossil fuel and chemical fertilizer (Jiao et al. 2019).Thus, these secondary forests have been unmanaged for 50-60 years and are now in mid-succession.
In warm-temperate Japan, the unmanaged forests are transitioning from dominance by shade-intolerant, deciduous broad-leaved trees (e.g.Quercus variabilis, Quercus serrata) to increasing abundance of shade-tolerant, evergreen broadleaved trees invading the understory and shrub layers (Ito 2007;Hirayama et al. 2011;Nakajima et al. 2018).Similar dynamics are observed in abandoned coppice-managed woodlands in Europe (e.g., Douda et al. 2017;Hedl et al. 2010;Keith et al. 2009).In warm-temperate Japan, unmanaged secondary forests are expected to succeed to lucidophyllous forest, the potential climax vegetation (Nakagoshi and Hong 2001;Miyawaki 2004;Nakajima et al. 2018).However, recruitment limitation due to the lack of seed sources could arrest vegetation dynamics at mid-succession (e.g.Acácio et al. 2007).Furthermore, invasion by non-native species could divert the course of vegetation succession (plagiosere, Moriyama et al. 1984;Tojima et al. 2004).The future course of succession of unmanaged secondary forests to potential climax vegetation, therefore, is uncertain.
Because of extensive human influence across the landscape, climax forests are rare globally (Rackham 2008).In warm-temperate Japan, the few climax forests in populated areas are preserved in shrines and temples, for purposes of religious worship (Kamada 2005;Ishii et al. 2010).Such forests are rare specimens of the potential climax vegetation (Miyawaki 1998) and could function as seed sources of climax species.Here, we investigated spatio-temporal variation in species composition and stand structure of an unmanaged secondary forest adjacent to a preserved lucidophyllous forest at a temple in western Japan.Observations indicated that the secondary forest, which has been unmanaged for nearly 60 years, is being invaded by shade-tolerant, evergreen species originating from the lucidophyllous forest.This has resulted in a spatial gradient of successional stages with increasing distance from the lucidophyllous forest (Azuma et al. 2014).To take advantage of this unique setting, we established research plots along this spatial gradient to investigate spatial variation in species composition and stand structure as well as temporal change over 15 years.Our objective was to elucidate the course of succession of abandoned satoyama to lucidophyllous forest in warm-temperate Japan.

Study site and methods
The study was conducted in an unmanaged secondary forest and adjacent lucidophyllous forest at Taisanji Temple, Hyogo Prefecture, Japan (34°41′N, 135°04′E, 70-200 m ASL, Figure 1(a)).The substrate of the slopes surrounding Taisanji Temple is granite (Kodate and Nakanishi 1986).Historical drawings from 1803 suggest that the surrounding vegetation was open pine forest (Matsushita 1997).Vegetation maps and survey data from 1960 to 1970 indicate evergreen forests had been established surrounding the temple grounds suggesting succession had proceeded (Kodate and Nakanishi 1986).The evergreen forest, directly behind the temple grounds, is a place of Buddhist training and has had minimal human intervention.The mature lucidophyllous forest is dominated by Castanopsis cuspidata and comprises many indigenous species, representative of the climax forest in this region (Ishida et al. 1998).Adjacent to the mature lucidophyllous forest is unmanaged secondary forest (abandoned satoyama), where neighboring farmers had actively utilized the forest until ca.1960s.The secondary forest is dominated by deciduous broad-leaved trees.Observations suggest that after management ceased, evergreen trees from the lucidophyllous forest are slowly invading into the secondary forest, creating a spatial gradient of forest succession with increasing distance from the lucidophyllous forest.
We established two permanent plots to observe vegetation dynamics.The mature lucidophyllous forest (M) plot (50×50 m) was established in 2003 at the foot of the hill on the temple side, while the far-secondary forest (F) plot (50×40 m) was established in 2005 in the unmanaged secondary forest on the opposite side of the hill (Figure 1(b)).These plots are the same as those studied by Azuma et al. (2014).We counted approximately 100 annual rings in a core sample taken at 30 cm height from the trunk of the largest tree in the M plot in 2012 (C.cuspidata, DBH = 78 cm), suggesting that the forest is near climax stage.In the F plot, we counted 60-70 annual rings in core samples from the stems of multi-stemmed Quercus serrata and Quercus variabilis trees, suggesting the secondary forest was last cut around 1950 (Azuma et al. 2014).We measured the diameter at breast height (DBH, 1.3 m above ground) and height of all trees taller than 1.3 m within the research plots using diameter tapes, digital calipers (for DBH<2 cm), telescoping poles (height<8 m), and ultra-sound clinometers (Vertex III, Haglof, Sweden).Diameter measurements were repeated in 2008, 2014 and 2020.Height measurements were repeated in 2014 and 2020.These two plots were used to observe temporal change in species composition and stand structure.
In 2020, we established two additional research plots near the ridge of the hill between the M and F plots to investigate spatial variation of species composition and stand structure with increasing distance from the mature lucidophyllous forest.The young lucidophyllous forest (Y) plot (20×30 m) is located in a relatively young lucidophyllous forest ca. 100 m upslope from the M plot (Figure 1(b)).The closesecondary (C) forest plot (20×40 m) is located in a mixed evergreen-deciduous forest on the mature forest side, upslope of the F plot.Plot sizes are variable reflecting the spatial extent of each forest type.We counted 48-65 annual rings in core samples taken at 30 cm height from the trunks of the largest trees in the Y and C plots (all C. cuspidata), indicating that the Y plot is younger than the M plot and that the oldest C. cuspidata trees established in the C plot soon after management ceased.The DBH and tree height of all trees taller than 1.3 m in Y and C plots were measured using the same criteria and methods as for the M and F plots above.

Data analysis
Using the DBH data in each survey year, we calculated the basal area (BA, m 2 ha −1 ) for each tree species in the plots.We used the Chao-Jaccard index to compare species composition among the plots based on abundance (trees ha −1 ) and the Bray-Curtis index (Chao et al. 2005) to compare stand structure based on BA.We chose Chao-Jaccard over the Bray-Curtis index for abundance because the Bray-Curtis index is heavily influenced by the relative abundance of species.The Bray-Curtis index can be calculated based on the relative abundances of species (e.g.De Caceres et al. 2013;Hao et al. 2019), indicating that the independent variables need not be count data.Several previous studies have applied the Bray-Curtis index to assess similarity in stand structure among communities using relative basal area (e.g.Torre-Cuadros et al. 2007;Hotta et al. 2015;Sasaki et al. 2018).
The results were visualized using the non-metric multidimensional scaling (nMDS) ordination using the function "metaMDS" of the package "vegan" in R software (ver.3.4.1,R Development Core Team) (Oksanen et al. 2020).The nMDS is a distance-based ordination technique where relationships among biological communities are drawn on a two-dimensional plane to display graphically similarities among ecological communities.It is suited for ecological analyses because it is nonparametric and can be used to relativize distance measures based on a wide variety of ecological data (McCune et al. 2002).Distance between plots on the nMDS ordination plane represents their relative similarities.Here, changes over time in the coordinates of the F plot relative to the M plot on the nMDS ordination plane were interpreted as the course of succession (Ruiz-Jaen and Aide 2006; Mathews et al. 2010;Hiers et al. 2012).We used type-two permutational multivariate analysis of variance (PERMANOVA) using distance matrices and evaluated plot distances on the nMDS plane to test significant change over time and differences among plots in species composition and stand structure, respectively.PERMANOVA (9999 permutations) was conducted using the function "ado-nis2" of the "vegan" package in R (Martinez-Arbizu 2017).Multiple comparisons for evaluating distances between plots were conducted using the "pairwiseAdonis2" package, where p-values of the pairwise PREMANOVA are corrected using the Holm correction (Hervé 2016).In addition to visualizing similarity among communities, correlations between the original species vectors, abundance/ dominance of a species in each community, and the axis scores of the nMDS ordination can be computed and these correlations scaled to represent the direction and strength of influence of species on each community (Legendre and Gallagher 2001).To infer species that affected the composition and structure of the study plots, we plotted species having significant correlations with axes 1 and 2 on the nMDS ordination plane.

Temporal change of species composition and stand structure
Comparison of DBH distributions between 2003 and 2014 indicated that species composition and size distribution of the M plot had changed very little between surveys and was characterized by dominance of shade-tolerant, climax species, such as C. cuspidata, Aucuba japonica, Camellia japonica and Cleyera japonica (Figure 2).In contrast, species composition of the F plot in 2005 was characterized by dominance of shade-intolerant, deciduous (e.g.Q. serrata, Lyonia ovallifolia) and evergreen (e.g.Ilex pendiculosa, Quercus phillyraeoides) trees and deciduous shrubs (e.g.Rhododendron reticulatum).While stem density remained relatively stable in both plots (ca.2400 and 3800 trees ha −1 , for M and F plots, respectively, Table S1, S2), total BA increased by 22% (from 37.11 m 2 ha −1 in 2003 to 45.27 in 2020) in the M plot and by 37% (21.66 m 2 ha −1 in 2003 to 29.67 in 2020) in the F plot (Table S3, S4).
Comparison of height distributions between 2003 and 2014 indicated that the canopy height of the M plot increased, but the relative vertical distribution of species remained stable (Figure 2).The upper canopy continued to be dominated by C. cuspidata, which had the largest BA (Table S3), and the lower canopy by Camellia japonica.The canopy height of the F plot also increased.While Q. serrata continued to dominate in the upper-canopy as well as in BA (Table S4), marked changes were observed in the vertical distribution of species in the mid-and lower-canopy layers.Q. phillyraeoides and Ilex pedunculosa increased markedly in height and BA and dominated the mid-canopy in 2014, while in the lower canopy, deciduous species (e.g.R. reticulatum, L. ovafolia) decreased and evergreen species (e.g.Camellia japonica, Cleyera japonica) increased.
Abundances of the dominant climax species in the M plot were positively correlated with Axis 1 and negatively correlated with Axis 2 of the abundance-based nMDS ordination plane, while it was vice versa for the dominant species of the F plot (Figure 3).The BA of C. cuspidata was positively correlated, while that of the dominant species in the F plot was negatively correlated with Axis 1 of the BA-based nMDS ordination plane.During the study period, the coordinates of the M plot changed very little in relation to Axis 1 of both the abundance-and BA-based nMDS, reflecting stable species composition and stand structure.In contrast, Axis 1 values of the F plot tended to increase toward the direction of the M plot, although these changes were not statistically significant (Tables 1 and 2).

Spatial variation with distance from mature forest
The results of the most recent survey in 2020 indicated that the M and Y plots were dominated by C. cuspidata with Camellia japonica and other evergreen species comprising the mid-to lower-canopy layers (Figure 4).The DBH and height of C. cuspidata in the Y plot were smaller than in the M plot, reflecting the difference in stand age.Compared to the F plot, where the upper-canopy and BA were both dominated by Q. serrata, C. cuspidata dominated the uppercanopy of the C plot and evergreen species contributed larger proportion of the total BA (Table S6).The mid-canopy layer was more developed in the C plot, where I. pedunculosa, Q. phillyraeoides and Clethra barbinervis dominated in the mid-DBH (10-20 cm) and mid-height (7.3-13.3m) classes.Shade-intolerant deciduous species (R. reticulatum, L. ovalifolia) were less abundant in the lower-canopy layer of the C plot.
Abundance-based nMDS indicated that species compositions of the M and Y plots were very similar to each other (Figure 5, Table 3).Species composition of the F and C plots were different from the M and Y plots, as well as from each other.BA-based nMDS indicated that, stand structure of the M and Y plots differed from each other, reflecting the difference in size distribution of the dominant species (Table 4).Stand structure of the F and C plots differed from the M and Y plots, as well as from each other, reflecting the difference in vertical distribution of species.

Discussion
Our results indicated that, in contrast to the dynamic changes observed in the far-secondary forest during the study period, species composition and stand structure of the mature lucidophyllous forest remained relatively stable, suggesting that it is approaching climax state.In the farsecondary forest, shade-intolerant species, such as R. reticulatum and L. ovalifolia, decreased markedly in the lower-canopy layer during the study period and were replaced by shade-tolerant evergreen species.Similar vegetation dynamics have been observed in unmanaged secondary forests in many regions in Japan (Nakajima et al. 2018).For example, in an unmanaged, secondary broad-leaved forest in Kyoto, tree density decreased over a 12-year period as shadeintolerant shrubs were replaced by evergreen species, such as Cleyera japonica and E. japonica, whereas total basal area increased due to growth of the canopy dominant trees (Ito 2007).Hirayama et al. (2011) compared species composition between mid-successional forest dominated by Q. serrata and Q. variabilis (abandoned satoyama), with that of a latesuccessional forest dominated by C. cuspidata, and found that the former had a more developed shrub layer.Morimoto  and Yoshida (2005) found that between 1974 and 1995, native Rhododendron populations in Kyoto City had declined as the number of unmanaged secondary forests increased.
Traditional coppice management maintained the forests at early stages of succession with open canopy conditions (Kobori and Primack 2003).Our results, together with reports from other unmanaged satoyama across Japan suggest that, after management ceased, growth and increasing leaf area of the upper-canopy trees reduce the amount of light penetrating into the forest such that, in the lowercanopy, shade-intolerant species gradually decline and are replaced by shade-tolerant evergreen species.
In warm-temperate Japan, the dominant canopy species are expected to succeed from shade-intolerant, deciduous oaks to shade-tolerant, evergreen oaks (Miyawaki 2004).Seeds of these Fagaceous species (acorns) are mostly gravity and animal dispersed.In secondary forests in Japan, acorns may be transported 20 m to as much as 40 m by rodents (Iida 1996).Birds can also transport C. cuspidata acorns over long distances (Hiroki 2001).In a secondary forest dominated by Q. serrata and Q. variabilis, continuous recruitment and gradual invasion of C. cuspidata into the secondary forest occurred, such that seedlings of C. cuspidata established as far as 40 m away from the nearest adult trees in the adjacent C. cuspidata-dominated forest (Hirayama et al. 2010).Our results indicated evergreen trees and shrubs, originating from the lucidophyllous forest, are invading the lowercanopy of the secondary forest.The direction of temporal change on the nMDS ordination plane suggested that the secondary forest is succeeding toward the climax, lucidophyllous forest following the normal sere for warmtemperate Japan.The marked difference in canopy structure between the mature and secondary forests, however, indicated seedlings and young trees of C. cuspidata in the lower canopy of the secondary forest are far from attaining dominant status, because deciduous oaks and shade-intolerant evergreen trees continue to dominate the upper canopy.Slope position may also affect stand growth and management history.Growth of C. cuspidata may be slower in the upperslope stands (Y and C plots).Because of its lower-slope position (i.e.ease of access), the F plot may have been managed more frequently than the C plot, maintaining the far-secondary forest at early-successional stage until more recently.

Conclusions
Spatio-temporal analysis of species composition and stand structure of unmanaged, secondary forest adjacent to mature lucidophyllous forest using nMDS ordination allowed us to infer the course of succession of unmanaged satoyama.Past management practices, such as coppice forestry in the case of satoyama, may continue to influence the composition and structure of ecosystems for decades to centuries (Bürgi et al. 2013;Perring et al. 2016;Douda et al. 2017).Such legacy effects of historical land use can have lasting effects on  ecosystem and community dynamics (Cuddington 2011).In addition, connectivity, seed flux, and colonization among different communities within a landscape influence species composition (Martín-Queller and Saura 2013).The lack of source populations in human-dominated landscapes can cause recruitment limitation arresting vegetation dynamics at mid-succession (Duelli and Obrist 2003;Acácio et al. 2007).Remnants of natural and semi-natural habitats can function as refugia for rare species and as stepping stones for species dispersal, contributing to restoration of inherent vegetation dynamics at the landscape level (Duelli and Obrist 2003).Although dynamic changes in species composition, especially in the lower-canopy, have been observed in many unmanaged satoyama, the marked difference in canopy structure between the mature and secondary forests in this study suggest, legacy effects of satoyama management can persist for several decades before individual canopy trees are replaced by natural disturbances, such as typhoon, and the unmanaged, secondary forest attains climax composition and structure.In our study site, the adjacent lucidophyllous forest is a seed source for recruitment of shade-tolerant species.However, for secondary forests which lack nearby seed sources of climax species, close monitoring of vegetation change may be necessary to prevent arrested or diverted succession.Further studies should elucidate the dynamics of unmanaged satoyama in the context of landscape-level distribution of natural and semi-natural forests.

Figure 1 .
Figure 1.Location of the study site, Taisanji Temple, in Kobe City, Japan (a).The mature lucidophyllous forest is located on the west side of the mountain (b, solid outline).The east side of the ridge-line is secondary forest, which was cut regularly to obtain firewood until ca.60 years ago, after which it was abandoned (b, dotted outline).Climax species originating from the lucidophyllous forest are invading into the secondary forest.Four study plots were established along the spatial gradient of vegetation change with increasing distance from the mature lucidophyllous forest: mature lucidophyllous (M), young lucidophyllous (Y), closesecondary (C), and far secondary (F) plots.Image: Google ©2021 CNES/Airbus, Digital Earth Technologies, Maxar Technologies, Planet.com.

Figure 2 .
Figure 2. Temporal change of diameter and height distributions of species in the mature lucidophyllous and far-secondary forest plots during the study period.The bars for the smallest diameter/height classes in each plot are truncated and numbers next to the bars indicate number of trees.

Figure 3 .
Figure 3. Nonmetric multidimensional scaling (nMDS) ordination of vegetation similarity based on abundance and basal area of the mature lucidophyllous (M, □) and the far secondary (F, ○) plots during the study period.Coordinates of each species (+: deciduous, X: evergreen) reflect correlations with each nMDS axis.Stress values are <0.01 for both abundance and basal area.

Figure 4 .
Figure 4. Diameter and height distributions of species in the four research plots in 2020.The bars for the smallest diameter/height classes in each plot are truncated and numbers next to the bars indicate number of trees.

Figure 5 .
Figure 5. Nonmetric multidimensional scaling (nMDS) ordination of vegetation similarity based on abundance and basal area of the mature lucidophyllous (M, □), young lucidophyllous (Y, ◊), far secondary (F, ○) and close-secondary (C, Δ) plots in 2020.Coordinates of each species (+: deciduous, X: evergreen) reflect correlations with each nMDS axis.Stress values are less than 0.01 for both abundance and basal area.

Table 1 .
Temporal change of species composition as measured by pairwise Chao-Jaccard similarity indices based on the abundance of species.

Table 2 .
Temporal change of stand structure as measured by pairwise Bray-Curtis similarity indices based on basal area of species.

Table 4 .
Spatial variation of species composition as measured by pairwise Bray-Curtis similarity indices based on basal area of species.: P < 0.05; **: P < 0.01, Larger values reflect lower similarity.