Di ﬀ erential Early Performance of Two Underplanted Hardwood Tree Species Following Restoration Treatments in High-Graded Temperate Rainforests

: Raul í ( Nothofagus alpina (Poepp. & Endl.)) and Ulmo ( Eucryphia cordifolia Cav.) are mid-tolerant tree species in the Coihue-Raul í -Tepa (ca. 0.55 mill ha) and Evergreen (ca. 4.1 mill ha) forest types in south-central Chile, respectively. These species have been selectively logged in old-growth forests especially during the 20th century, Raul í mostly for its highly valuable timber, and Ulmo for its highly demanded ﬁrewood and bark for the tannery industry. Natural regeneration of these species occurs mostly through canopy gaps, but it can be retarded, or even inhibited, when the cover of the understory vegetation becomes unusually dense, such as in high-graded forests. Although underplanting is possible for these species, the knowledge about their growth in forest understories is scarce, and necessary to inform restoration programs. Therefore, we evaluated short-term responses (two years) of underplanted containerized seedlings in root-collar diameter, height, stem volume, and in the slenderness index, as a function of canopy openness (%, continuous variable) and three restoration treatments (categorical variables, plus one control treatment) at two di ﬀ erent sites with high-graded old-growth forests for each forest type. By using generalized linear mixed-e ﬀ ects models (GLMMs) we determined that Raul í was more sensitive to the inﬂuence of both canopy openness and restoration treatments, while Ulmo was mostly inﬂuenced by canopy openness. Speciﬁcally, Raul í was positively inﬂuenced by canopy openness and restoration treatments in all response variables except for the slenderness index. Conversely, Ulmo was inﬂuenced by canopy openness in all response variables except the slenderness index, which was inﬂuenced by both predictor variables (canopy openness and restoration treatments). Thus, prospects for restoration with these species are discussed, including possible ontogenetic changes in their responses to light that may demand continuous silvicultural operations to recover the productive and functional roles of these species in these forest ecosystems. í -Tepa forest types. We hypothesized that canopy openness and the restoration treatments di ﬀ erentially enhance the early growth of Ulmo and Raul í in these forest types.


Introduction
Forest degradation is estimated to affect nearly one-half of the world's forests, which are less biodiverse and productive than well-conserved forests [1,2]. Commonly, forest degradation delays

Study Sites
The present study was carried out in two of the major forest types within the Valdivian Temperate Rainforests of South America in south-central Chile, the Evergreen and the Coihue-Raulí-Tepa [39] ( Figure 1). We selected two sites for each forest type in which we identified old-growth stands that were high-graded 30-50 years ago (PJ Donoso, personal observation). These stands have a scattered distribution of medium-to large-sized trees, and a patchy distribution of poled-sized trees mostly of poor quality of low-value tree species and dense thickets of Chusquea spp. (C. quila in the Evergreen forests and C. coleou in the Coihue-Raulí-Tepa forests), and ferns or other understory competing plant species (especially in Evergreen forest type). Different competing understory species and poor tree regeneration are also noted [40]. The general characteristics of the studied forest types are presented in Table 1.

Study Sites
The present study was carried out in two of the major forest types within the Valdivian Temperate Rainforests of South America in south-central Chile, the Evergreen and the Coihue-Raulí-Tepa [39] (Figure 1). We selected two sites for each forest type in which we identified old-growth stands that were high-graded 30-50 years ago (PJ Donoso, personal observation). These stands have a scattered distribution of medium-to large-sized trees, and a patchy distribution of poled-sized trees mostly of poor quality of low-value tree species and dense thickets of Chusquea spp. (C. quila in the Evergreen forests and C. coleou in the Coihue-Raulí-Tepa forests), and ferns or other understory competing plant species (especially in Evergreen forest type). Different competing understory species and poor tree regeneration are also noted [40]. The general characteristics of the studied forest types are presented in Table 1. The sites for the evergreen forest type were located in two public forest reserves, Llancahue and Pumillahue, both in the Loncoche's transversal mountain range in the Chilean intermediate depression. Similar climate (coastal oceanic with a Mediterranean influence according to Koeppen classification, [40]) but some soil differences appeared between both sites (Table 1). In Llancahue, the soil series corresponds to Los Ulmos (Typic Paleudults: Ultisol), which are dominated by clay. At Pumillahue, the soil series is Correltúe (Andic Palehumults: Ultisol) characterized by silt loam soils [41]. These soils are derived from old volcanic ashes and have suffered intense weathering processes that lead to low soil fertility (Table 1). Correltúe has a better rooting development compared to Los Ulmos, and also a higher proportion of organic carbon (12% and 6.5%, respectively). Correltúe also has a higher phosphorus content up to 13 ppm [41]. Therefore, the Pumillahue site has better soil properties for seeding establishment than Llancahue.
The sites for the Coihue-Raulí-Tepa forest type are located in the San Pablo de Tregua experimental forest station belonging to Universidad Austral de Chile, and in Riñimahuida (private property). Both sites are in the Andes range and correspond to the same Liquiñe soil series (Acrudoxic Hapludands: Andisol). The soil in San Pablo de Tregua has an infertile pumice horizon over basaltic- The sites for the evergreen forest type were located in two public forest reserves, Llancahue and Pumillahue, both in the Loncoche's transversal mountain range in the Chilean intermediate depression. Similar climate (coastal oceanic with a Mediterranean influence according to Koeppen classification, [40]) but some soil differences appeared between both sites (Table 1). In Llancahue, the soil series corresponds to Los Ulmos (Typic Paleudults: Ultisol), which are dominated by clay. At Pumillahue, the soil series is Correltúe (Andic Palehumults: Ultisol) characterized by silt loam soils [41]. These soils are derived from old volcanic ashes and have suffered intense weathering processes that lead to low soil fertility (Table 1). Correltúe has a better rooting development compared to Los Ulmos, and also a higher proportion of organic carbon (12% and 6.5%, respectively). Correltúe also has a higher phosphorus content up to 13 ppm [41]. Therefore, the Pumillahue site has better soil properties for seeding establishment than Llancahue.
The sites for the Coihue-Raulí-Tepa forest type are located in the San Pablo de Tregua experimental forest station belonging to Universidad Austral de Chile, and in Riñimahuida (private property). Both sites are in the Andes range and correspond to the same Liquiñe soil series (Acrudoxic Hapludands: Andisol). The soil in San Pablo de Tregua has an infertile pumice horizon over basaltic-andesitic rocks [41], presenting high phosphorus retention and aluminum levels due to the presence of alophan. In Riñimahuida, the soil has a fertile horizon of about 30 to 40 cm in depth, but its main limitation is excessive drainage, which makes it susceptible to erosion by water or wind [40,41]. As both soils have Andic properties, it is not unusual to find low bulk density, acidic soil conditions, high porosity and fast drainage [42]. Sources: [40,41,[43][44][45] These forests had between 775 to 914 trees per hectare and basal areas between 25.4 to 65.5 m 2 per hectare (Table 1). These basal areas represent between one-fourth to two-thirds of the common basal areas expected in old-growth forests in the region [43].

Study Design and Restoration Treatments
Four 2000 m 2 plots, divided in four quadrants, were established at each site. In each quadrant, three restoration treatments were randomly assigned and implemented, leaving one quadrant as a control (no treatment) ( Table 2). This design was not completely orthogonal, but it represents increasing intensity to promote theoretically better conditions for the regeneration of light-demanding tree species. After implementation of the restoration treatments, a plantation was systematically conducted in each treated quadrant. In order to capture the greater variability in canopy openness, 20 seedlings per quadrant were established. Table 2. Description of the restoration treatments implemented in the experiment.

Treatments
Description and Activities

Improvement cut
Underplanting plus improvement cut of trees with lowest quality and thinning among clumped groups of low-diameter trees with the goal to homogenize light penetration into the understory to stimulate growth of seedlings.

Improvement cut and understory vegetation control
Same as treatment 2, but including understory control in the entire plot, which included manually cutting the shrubs and piling them outside of the plots. This treatment aimed to avoid competition of seedlings with understory vegetation.
4. Improvement cut, understory vegetation control and soil scarification Same as above (3), plus manual topsoil scarification with the goal to remove the litter layer, including small woody debris (<20 cm in diameter), which was piled out of the regeneration plots.

Plant Material
Seedlings used in this study had a range from 35 to 45 cm in total height (h) and 4 to 5 mm in root-collar diameter (d). Neither showed significant differences in these variables nor their stem volume index (v) at the establishment. Seedlings were produced in 16 m tall containers with 130 cm 3 in rooting volume, with a substrate of composted Pinus radiata (Monterey pine) bark mixed with a slow-release fertilizer (e.g., 5 kg per 1 m 3 of composted bark). Details about protocols for seedling production and the characteristics of these seedlings can be found in Bustos et al. [46] and Donoso et al. [47].

Measurements
At the time of planting and following two growing seasons, we measured root collar diameter (d) and height (h) during the dormant season (winter). Stem volume index (v, cm 3 seedling −1 ) was calculated using the cone formula as in Rose and Ketchum [48]. Furthermore, we computed the slenderness index (Slen), at year 2, which is the ratio of h/d, where lower values reflect plants with better biomass distribution, therefore having a greater likelihood of better field performance [47]. Finally, we computed the periodic annual increment (pai) for diameter, height, and volume for each seedling, represented as pai d , pai h , and pai v " respectively.
We computed canopy openness (CO, %) at the apex of each seedling by taking hemispherical photographs processed by the software gap light analyzer (GLA) [49]. Further details about the settings used in GLA are given by earlier studies of Donoso et al. [9,45] and Soto et al. [10].

Statistical Analyses
We analyzed the effects of canopy openness and the four restoration treatments on the response variables pai d , pai h , pai v , and Slen. We fit mixed-effects models for each response variable, by using canopy openness and the treatments as predictor variables; meanwhile, site and the plot were considered as random effects. The mixed-effects model framework allowed us to take into account the hierarchical structure of the data, producing more efficient estimates of the variance components [50][51][52][53]. All the mixed-effects models were fitted by restricted estimated maximum likelihood. The best random variable structure to be used in this experiment was the plot nested within the site (site/plot). The results for the selection of this random structure is presented in Appendix A.
For each response variable, we compared the following four models by modifying the predictor variables: (1) canopy openness only, (2) restoration treatments only, (3) canopy openness and restoration treatments, and (4) the interaction between canopy openness and the restoration treatments. The Akaike information criteria (AIC) as well as the Bayesian information criteria (BIC) were used to compare each of the four models [54]. All statistical analyses were performed using the R packages "nlme", "lme4" [50], and "effects" [55] using R [56].

Results
Two growing seasons following the experiment establishment, both species showed a quite similar growth patterns but different responses to canopy openness and the restoration treatments ( Figure 2). For example, Raulí was significantly influenced by canopy openness and the restoration treatments for root-collar diameter (pai d ), total height (pai h ), and stem volume index (pai v ), but not for the slenderness index (Slen) (Table 3, Figure 2). Conversely, Ulmo was only significantly influenced by canopy openness in pai d , pai h , and pai v , while Slen was significantly influenced by both fixed effects (Table 3). Considering only the canopy openness at a level of 40% (dashed lines in Figure 2), similar growth trends were observed for both species, but Ulmo had better growth and narrower error bands (95% confidence intervals) in pai h and pai v than Raulí (Figure 2). The latter grew slightly better in pai d than Ulmo (Figure 2). Table 3. Akaike's (AIC) and Bayesian information criterion indices (BIC) of the fitted mixed-effects models by the response variables and species. The best-supported model for each variable is presented in bold numbers.

Growth Responses to Partial Light Conditions.
Overall, Ulmo had a slightly better growth performance than Raulí. The periodic annual increment in d (paid) was similar for both species, averaging 0.3 cm yr −1 at 40% canopy openness, with both species showing a positive paid response to increasing canopy openness, an expected response for these mid-tolerant species, which have similar light compensation points (Ulmo 5.3 ± 1.2 lmol m −2 s −1 ulmo [57], and Raulí 7.0 ± 0.4 lmol m −2 s −1 [58], compared with shade-intolerant Nothofagus species that have values close to 20 lmol m −2 s −1 [58]). However, Raulí had higher variability in paid than Ulmo, For Raulí, the restoration treatments had significant effects on pai d , pai h , and pai v , but not on Slen. Specifically, the three restoration treatments evaluated here had significant effects with respect to the untreated plots (Table 3). The improvement cut (B) and improvement cut with understory vegetation control (C) treatments had 23% higher pai d than the control treatment (A). On the other hand, there was a 35.3% increase in pai d in the improvement cut, understory control, and soil scarification treatment (D) over the control (Figure 2). A similar trend was obtained for pai h , where treatment C was 62.5%, treatment D was 54.1%, and treatment B was 45.8%, higher than the control (Figure 2). Treatment outperformance compared to the control was greatest for pai v , which was 150% higher in treatments C and D, and 135% higher in treatment B with respect to the control treatment ( Figure 2).
For Ulmo, the restoration treatments did not have a significant effect upon pai d , pai h , and pai v , but they did upon Slen (Table 3). Specifically, the control had higher Slen compared to the other treatments, being 2% higher than B, and 11.8% higher than treatments C and D (Figure 2).

Growth Responses to Partial Light Conditions
Overall, Ulmo had a slightly better growth performance than Raulí. The periodic annual increment in d (pai d ) was similar for both species, averaging 0.3 cm yr −1 at 40% canopy openness, with both species showing a positive pai d response to increasing canopy openness, an expected response for these mid-tolerant species, which have similar light compensation points (Ulmo 5.3 ± 1.2 lmol m −2 s −1 Ulmo [57], and Raulí 7.0 ± 0.4 lmol m −2 s −1 [58], compared with shade-intolerant Nothofagus species that have values close to 20 lmol m −2 s −1 [58]). However, Raulí had higher variability in pai d than Ulmo, suggesting that it is more sensitive to local variation in environmental conditions, such as soil quality and canopy openness [28]. Actually, Lusk et al. [59] showed that juveniles (50-200 cm in height) of E. cordifolia developed mostly within a range of 20% to 70% canopy openness in old-growth forests, while Donoso [19] reported that natural regeneration of N. alpina has negligible growth rates under closed canopies in old-growth forests. Nevertheless, in general, both species follow the gap-regeneration mode described by Veblen et al. [25], and may reach the forest canopies in old-growth forests only if partial light conditions are available in the under-and mid-stories [19,22,23,27], such as Raulí does in partially harvested Nothofagus-dominated old-growth forests in the Andes [28]. Therefore, underplanting of these species would succeed within a range of partial canopy openness, and promoting seedling growth into larger size classes as they recruit into the overstory should also be accompanied by controls in canopy openness [9].

Mixed Silvicultural Treatments to Favor the Development of Underplanted Seedlings in High-Graded Forest
The restoration treatments conducted here had marked effects on the growth (pai in d, h, and v) of Raulí but not of Ulmo. Specifically, growth in height and volume for Raulí was higher in the treatment that included improvement cut and understory control, while the lowest growth rates occurred under control plots. Different studies have indirectly shown the sensitivity of Raulí and Ulmo to competition with understory vegetation [9,19,28]. Thus, Raulí is a highly demanding species in terms of niche conditions for its establishment and growing phases [3]. Similarly, Uteau and Donoso [37] suggested that Ulmo has strong competition for light in young underplanted seedlings (asymmetric competition), but that during the sapling stage, competition with its conspecific species occurs mostly for soil nutrients and water (symmetric competition). However, we did not observe this pattern for Ulmo seedlings in this study. While both species are mid-tolerant to shade, the fact that one is deciduous and the other evergreen may have implications upon their responses to resource availability. Evergreen seedlings (such as Ulmo) are more conservative in terms of resource use (e.g., nutrients and water) than deciduous ones [59][60][61][62][63]. This can partly explain the lower sensitivity of Ulmo to the different restoration treatments and to increasing canopy openness.

Implications for Forests Restoration
Underplanting activities have been proven to be effective methods of artificial regeneration in many forested habitats around the globe [7], and therefore, it is a useful restoration practice to direct succession trajectories in a more predictable way [6,10,64]. The fast-growth patterns of Raulí and Ulmo would enable a more rapid forest recovery of high-graded forests, e.g., biomass accumulation in shorter periods than through natural succession [17]. Restoring the essential forest attributes is pivotal to enhance the forest functions and processes and, not less important, the forest resilience when succession is arrested by recalcitrant understory vegetation after partial overstory disturbances [31], which is the case of the present study. Intermediate overstory densities have been proven to provide the best results in survival and growth of underplanted seedlings in most forest biomes in the world, including mid-tolerant species [7]. In this study, we did not evaluate the growth under very open canopies (>60%). However, Donoso et al. [65] and Soto et al. [10,35] showed that greater growth rates in Raulí occur in open fields when planted with neighboring faster-growing evergreen species (i.e., facilitation mechanisms under partial shade). In addition, Donoso et al. [9,45] showed that the growth in Raulí was indifferent to canopy openness during the seedling stages but during the sapling stages, growth was enhanced with increased light. Therefore, while further studies should continue to illustrate responses of underplanted seedlings to varying site and microsite conditions, this artificial regeneration option seems suitable for these species and convenient to restore high-graded forests.

Conclusions
Two mid-tolerant tree species that dominate the canopy in the forests where they naturally develop showed variable performances when underplanted in high-graded old-growth forests. Their mid-tolerant-to-shade character was reflected in the positive responses to increasing canopy openness, but Ulmo seemed to adapt better to these understory conditions, as reflected by a narrower variance of the growth within different levels of canopy openness. Raulí reacted positively to the restoration treatments and canopy openness, indicating the need of silviculture protocols in these high-graded forests through understory control and tree density reduction. Ecologically, the early results of this study may suggest that Ulmo has symmetric competition for light and Raulí is more sensitive to competition for soil nutrients and water (i.e., a positive effect of restoration treatments and canopy openness), suggesting asymmetric competition. The latter can be attributed to the evergreen nature of Ulmo in contrast to the deciduous Raulí. Acknowledgments: D.P.S. and M.G.-C. thank the "Fondo Semilla 2018 de iniciación en la investigación de la Universidad de Aysén". We also thank students and research assistants that were involved in this research and, the constructive suggestions of 3 reviewers and the academic editor.

Conflicts of Interest:
The authors declare no conflict of interest.

Effects of the Structure Within the Random Effects
Using the restricted estimated maximum likelihood (REML) after a stepwise selection approach, the significance of each component in the random terms were assessed. In general, for both species the proposed nested structure (i.e., the plot "P" within a site "site") for the random effect had a better performance than "site" and "P" alone (Table A1). Nevertheless, the slenderness models for Ulmo and Raulí, and the height growth model for Raulí, showed that the random structure with "site" performed better than the proposed nested ones (lowest AIC and BIC statistics). However, no significant differences among these models compared to the nested structure (p > 0.05) or having >2 AIC units have been found (Table A1). Table A1. Restricted estimated maximum likelihood (REML) after a stepwise selection. Comparison between the best-supported models for each response variable (in bold) with those of the same structure in the fixed terms, with variation in the structure of the random terms. The (site/P) notation illustrates the nestedness of plots (P) within a site (Site).

Species
Response