Conversion of Pinus nigra Plantations with Natural Regeneration in the Slovenian Karst: The Importance of Intermediate, Gradually Formed Canopy Gaps

Since the mid-19th century, Pinus nigra plantations have played a key role in the restoration of degraded European landscapes. Nowadays, these plantations are aging and prone to natural disturbances, insect infestations, and diseases. For their successful gradual conversion, knowledge of optimal gap spatiotemporal dynamics is crucial. We studied herb and natural regeneration patterns along with site factors on 477 subplots within 44 plots distributed over four stand types: closed stand (14% diffuse light), open stand (21%), gap edge (23%), and gap (57%). Despite the abundant Quercus petraea, Q. cerris, and Q. pubescens mast year, no one-year seedlings were recorded, which is likely due to the summer drought. Short seedlings (h < 20 cm) of Quercus sp., Fraxinus ornus and Ostrya carpinifolia were more successful within closed stands. Short Quercus seedlings were positively associated with soil depth and negatively associated with soil nutrients, distance to seed trees, and Sesleria autumnalis coverage. Taller Quercus seedlings required more light than both of its strongest competitors and were positively related to humid soils and less rocky sites. Ungulate overbrowsing significantly impeded natural regeneration. The results indicate a satisfactory Quercus density for conversion and the importance of advanced regeneration, which should be gradually, but persistently, released by progressively widening gaps.


Introduction
Globally, there are many former forest lands that have been degraded due to clearing, burning, and overgrazing. Afforestation of these areas could improve soil productivity, forest climate and biodiversity, as well as mitigate climate change [1]. Good practices from the past are important for encouraging new forest restoration projects and for learning [2]. Afforestation of the Karst region above the Gulf of Trieste from the mid-19th century represents one of the first successful examples of forest restoration on almost completely barren land exposed to drought and strong winds [3]. Before human colonization, the landscape was overgrown with deciduous forests consisting mainly of oaks (Quercus petraea (Matt.) Liebl., Quercus cerris L., and Quercus pubescens Willd.; hereafter Quercus), and, to a lesser extent on more mesophilic sites, Fagus sylvatica L. [4].
Due to the degraded sites in the Karst, Pinus nigra J.F. Arnold was used for restoration. Although it is not a native species on Karst, it is a specialist on shallow rocky calcareous soils and xerophytic sites. Later, P. nigra also subspontaneously expanded. There are about 16,000 ha of P. nigra stands alone within the Slovenian Karst region. Nowadays, P. nigra plantations across Europe are aging,

Stand and Site Characteristics
The study was carried out in the Karst region of southwest Slovenia, as seen in Figure 1. The analyzed P. nigra plantations are located on flatlands or gentle slopes above an altitude of 235-450 m a.s.l. Carbonate parent material dominates, but there are some small pockets of silicate in the form of chert, flysch, and sands. The soils are eroded and mainly in the form of shallow rendzina [16]. In some places, especially in depressions, deeper brown soils have developed. In addition to soil variability, land degradation has also occurred at different intensities; therefore, the potential natural vegetation is difficult to determine [4]. In more favorable ecological conditions, stands were most likely dominated by sessile oak (Q. petraea) (Seslerio-Quercetum sensu Braun-Blanque typology), while less favorable sites were dominated by downy oak (Q. pubescens), hop hornbeam and flowering ash stands (Ostryo-Quercetum pubescentis). Pinus nigra stands in Slovene Karst experience severe droughts in summer and higher precipitation than most natural P. nigra stands in Europe, which makes them prone to fungal diseases. Therefore, even greater management risks are expected in the future. At present, most stands are two-storied, with 87% P. nigra in the overstory and about 10% F. ornus and O. carpinifolia in the understory. Quercus and the noble broadleaves amount to only 3% in the total growing stock. Average growing stock varied from 148-223 m 3 ha −1 and increment 3.3-5.1 m 3 ha −1 year −1 . Stands were between 80 and 120 years old, and the upper stand height was about 25 m. The mean annual temperature in the period 2000-2011 was 12.0 • C, and the mean sum of yearly precipitation was 1234 mm, with extensive droughts in summer (Godnje meteorological station, 760 m a.s.l.; [17]). The climate is sub-Mediterranean.

Sampling Design and Recordings
The study object was the entire area of P. nigra plantations on comparable sites with mature stands and stands in regeneration, in which there had been no logging in the five years prior to the study. We randomly selected eight locations from the study object and installed 44 plots 25 × 25 m in size, as seen in Figure 2. The plots were randomly selected within four stand types: closed stand, partially open stand, the edge of a canopy gap and the center of a canopy gap. The gaps ranged from 1.2 to 4.4 ha in size. Until recently, mature stands were mostly lightly thinned, and managers only rarely formed gaps deliberately. Thus, it was not possible to select a balanced set of treatments in each location. For each plot, we sampled the main ecological features (relief type, rockiness, slope inclination, and orientation) and typical dendrometric data (e.g., tree species coverage, full inventory of dbh and heights).
Across the center of each plot, six systematic herb subplots were placed, and two rows of Quercus subplots were arranged to the left and right of the central transect, as seen in Figure 2. The first two herb subplots were located 1 m south and north of the plot center. The minimum distance between the two herb or Quercus subplots was 2 m. The center of the Quercus subplots was defined by the presence of Quercus seedlings older than one year. The size of the subplots was 1.5 × 1.5 m. The herb subplots were installed and recorded on all plots (N = 264), while the number of Quercus subplots varied from 0 to a maximum of 12 (N = 214). On each subplot ground vegetation coverage according to species was estimated visually from above and recorded to the nearest 10% from 10% to 100% and to the nearest 1% from 1% to 10%. Thus, the cumulative cover of herbs and shrubs often exceeded 100%. The same scale was also used for the estimation of coarse woody debris (CWD) and rock coverage (rockiness) on the forest floor. Microrelief (flat, concave, convex), slope inclination and orientation, soil depth, and cumulative depth of the organic horizons (L-litter, F-fermentation horizon and H-humus horizon; hereafter referred to as LFH) were measured. Regeneration density according to species was estimated corresponding to four height classes (one-year old seedlings, seedlings with height <20 cm, seedlings with height ≥20 and <130 cm and seedlings taller than 130 cm). Seedlings were categorized as browsed if the leading shoot was damaged.
On a sample of plots (n = 232), relative diffuse light 1.3 m above each subplot was assessed with two point quantum sensors (LI-COR, Inc., Lincoln, NE, USA), one placed above the subplot and the second in a nearby open area. Measurements were taken on overcast days [18]. On the same sample of subplots, topsoil water content (0-10 cm) was measured using time domain reflectometry [19] in September 2012 (moisture1) and October 2012 (moisture2). Each time measurement was taken after a period of dry weather. In 2012, only 400 mm of rainfall fell in the period from May to September, which is 73% of the average in the 2001-2011 vegetation period. At the same time, the average July

Sampling Design and Recordings
The study object was the entire area of P. nigra plantations on comparable sites with mature stands and stands in regeneration, in which there had been no logging in the five years prior to the study. We randomly selected eight locations from the study object and installed 44 plots 25 × 25 m in size, as seen in Figure 2. The plots were randomly selected within four stand types: closed stand, partially open stand, the edge of a canopy gap and the center of a canopy gap. The gaps ranged from 1.2 to 4.4 ha in size. Until recently, mature stands were mostly lightly thinned, and managers only rarely formed gaps deliberately. Thus, it was not possible to select a balanced set of treatments in each location. For each plot, we sampled the main ecological features (relief type, rockiness, slope inclination, and orientation) and typical dendrometric data (e.g., tree species coverage, full inventory of dbh and heights).

Data Analysis
Data were analyzed in R Version 3.6.1 [21]. Seedling density in relation to ecological factors was modeled with a series of generalized mixed-effects models (GLMMs) with negative binomial distribution and a logarithmic link function [22] since the data were overdispersed. All final models were selected following a top-down approach [23]. Different models with all combinations of ecologically meaningful fixed effects (comp. Chapter 3.4.), were built. These models were then compared against simpler nested models based on the ANOVA likelihood ratio test of fits for GLMMs [22]. Research units (i.e., plots) were considered as random factors. For model diagnostics, we inspected confidence intervals of parameters and examined sets of graphical summaries [24]. Across the center of each plot, six systematic herb subplots were placed, and two rows of Quercus subplots were arranged to the left and right of the central transect, as seen in Figure 2. The first two herb subplots were located 1 m south and north of the plot center. The minimum distance between the two herb or Quercus subplots was 2 m. The center of the Quercus subplots was defined by the presence of Quercus seedlings older than one year. The size of the subplots was 1.5 × 1.5 m. The herb subplots were installed and recorded on all plots (N = 264), while the number of Quercus subplots varied from 0 to a maximum of 12 (N = 214). On each subplot ground vegetation coverage according to species was estimated visually from above and recorded to the nearest 10% from 10% to 100% and to the nearest 1% from 1% to 10%. Thus, the cumulative cover of herbs and shrubs often exceeded 100%. The same scale was also used for the estimation of coarse woody debris (CWD) and rock coverage (rockiness) on the forest floor. Microrelief (flat, concave, convex), slope inclination and orientation, soil depth, and cumulative depth of the organic horizons (L-litter, F-fermentation horizon and H-humus horizon; hereafter referred to as LFH) were measured. Regeneration density according to species was estimated corresponding to four height classes (one-year old seedlings, seedlings with height <20 cm, seedlings with height ≥20 and <130 cm and seedlings taller than 130 cm). Seedlings were categorized as browsed if the leading shoot was damaged.
On a sample of plots (n = 232), relative diffuse light 1.3 m above each subplot was assessed with two point quantum sensors (LI-COR, Inc., Lincoln, NE, USA), one placed above the subplot and the second in a nearby open area. Measurements were taken on overcast days [18]. On the same sample of subplots, topsoil water content (0-10 cm) was measured using time domain reflectometry [19] in September 2012 (moisture1) and October 2012 (moisture2). Each time measurement was taken after a period of dry weather. In 2012, only 400 mm of rainfall fell in the period from May to September, which is 73% of the average in the 2001-2011 vegetation period. At the same time, the average July air temperature at the Bilje meteorological station was 1.6 • C above the long-term average (2001-2011). Measurements of topsoil moisture in 2012 thus indicated topsoil moisture following hot and dry summers [20]. We also sampled features of Quercus seed trees and their distance to subplots in a radius of 80 m from each plot.

Data Analysis
Data were analyzed in R Version 3.6.1 [21]. Seedling density in relation to ecological factors was modeled with a series of generalized mixed-effects models (GLMMs) with negative binomial distribution and a logarithmic link function [22] since the data were overdispersed. All final models were selected following a top-down approach [23]. Different models with all combinations of ecologically meaningful fixed effects (comp. Section 3.4), were built. These models were then compared against simpler nested models based on the ANOVA likelihood ratio test of fits for GLMMs [22]. Research units (i.e., plots) were considered as random factors. For model diagnostics, we inspected confidence intervals of parameters and examined sets of graphical summaries [24]. Light and topsoil moisture measurements were carried out on a sample of subplots (n = 215); thus, separate GLMMs for this subsample were carried out.
Mean Landolt indicator values (LIVs for temperature regime, continentality, light quantity, topsoil moisture, soil reaction, nutrient content in the soil, humus content, and soil aeration) were estimated as a weighted average of the Landolt indicator values [25] of all present understory layer species, and their abundances were used as weights [26]. Principal coordinate analysis (PCoA) was used to ordinate subplots based on the 2012 vegetation survey. The distance matrix was calculated by applying the Bray-Curtis distance measure to the square root of the plant species coverage data. Based on inspection of the eigenvalues, we selected only the first and second principal coordinate for presentation. Ecological variables (light and topsoil moisture measurements, distance to beech seed trees, etc.), mean LIVs, and seedling density according to species groups were fitted onto an ordination plot using the vegan package [27]. In identifying the indicator species of each group, we used the Indicator Value Index and its components [28] using the package indicspecies [29]. Differences in ecological factors between stand types and subplot types were tested with the nonparametric Kruskal-Wallis and Wilcoxon tests.

Variability of Ecological Factors within and between Stand Types
The median of relative diffuse light amounted to 13.8% below the closed stand and increased through the open stand (20.8%) and gap edge (23.1%) to the gap, at 57.3%, as seen in Figure 3. The lowest variability of light conditions was measured below the closed stand, followed by the gap edge, the open stand and lastly the gap, which exhibited considerably higher variability. The light distribution was followed by the response of the cumulative coverage of herbs and shrubs with the lowest values below the closed stand (55.0%), followed by the open stand (91.3%), the gap edge (100.0%), and the gap (100.1%). The overall topsoil humidity was lower in September compared to October 2012, with medians of 10.0% and 22.4%, respectively. In September, topsoil moisture was higher under the closed stand when compared to the other three stand types, while in October, the differences were smaller, as seen in Figure 3. The LIV for light followed a similar gradient to that of measured diffuse light, with increasing levels from the closed stand to the gap (K-W p < 0.001). LIV for continentality (K-W p < 0.001) and nutrients (K-W p = 0.0013) also indicated increasing values towards the gap.

Natural Regeneration
The regeneration density on both herb and Quercus subplots was high, at nearly 30,000 trees ha −1 on average, as seen in Table 1. Although 2011 was an abundant Quercus mast year, we did not record any one-year-old seedlings. In a separate sample outside our plots, we dug out 38 Quercus seedlings and found that only 5% were of a vegetative origin. Herb plots give an unbiased assessment of overall regeneration structure, and they indicated domination of F. ornus at 73%, followed by Quercus at 10%, O. carpinifolia at 9%, and Ulmus glabra Huds. aggr. at 2%. The noble broadleaves (Acer pseudoplatanus L., Acer obtusatum Waldst. & Kit ex Willd., Prunus avium L., Tillia cordata Mill., T. platyphyllos Scop., Fraxinus excelsior L., and Sorbus torminalis (L.) Crantz) and other species amounted to 6%, as seen in Appendix A. From the composition of regeneration according to height class, we can observe a slight decrease in the share of F. ornus, a considerable decrease in Quercus and noble broadleaves, and an increasing share of O. carpinifolia, as seen in Figure 4a. It seems that the average growth conditions

Natural Regeneration
The regeneration density on both herb and Quercus subplots was high, at nearly 30,000 trees ha −1 on average, as seen in Table 1. Although 2011 was an abundant Quercus mast year, we did not record any one-year-old seedlings. In a separate sample outside our plots, we dug out 38 Quercus seedlings and found that only 5% were of a vegetative origin. Herb plots give an unbiased assessment of overall regeneration structure, and they indicated domination of F. ornus at 73%, followed by Quercus at 10%, O. carpinifolia at 9%, and Ulmus glabra Huds. aggr. at 2%. The noble broadleaves (Acer pseudoplatanus L., Acer obtusatum Waldst. & Kit ex Willd., Prunus avium L., Tillia cordata Mill., T. platyphyllos Scop., Fraxinus excelsior L., and Sorbus torminalis (L.) Crantz) and other species amounted to 6%, as seen in Appendix A. From the composition of regeneration according to height class, we can observe a slight decrease in the share of F. ornus, a considerable decrease in Quercus and noble broadleaves, and an increasing share of O. carpinifolia, as seen in Figure 4a. It seems that the average growth conditions do not allow recruitment of the first group of tree species to higher regeneration classes. This is even more evident if we look at the regeneration structure according to tree species, height class, and stand type (Section 3.4). Within the first (h < 20 cm), second (20 ≤ h < 130 cm) and third regeneration height class (h ≥ 130 cm), the average share of seedlings damaged by ungulates amounted to 16%, 43%, and 26%, respectively. The overall browsing damage rate was 30%. Virtually all tree species were most damaged in the second height class, as seen in Figure 4b. Among the species, Ulmus was the most damaged, followed by Quercus, F. ornus and O. carpinifolia.

Herb Vegetation and Indication of Ecological Factors
We recorded 28 species of trees, 31 species of shrubs, and more than 120 species of herbs, which is by far more than that recorded in similar research in other phytogeographic regions of Slovenia. Ordination of vegetation indicated two main gradients, as seen in Figure 5. The first seems to be influenced by stand type. LFH thickness, CWD, and rockiness increase in the direction of closed stand conditions, while LIV for soil moisture and minimum and average distance to seed trees decrease. However, the centroids of stand types are located close together, indicating that other factors are also important, e.g., geographical distances between the plots. The second gradient, which is independent of the first, indicates changes in LIV for reaction, continentality and light as well as slope and coverage of Sesleria autumnalis Scop. The density of shorter and taller F. ornus and O. carpinifolia seedlings increases in the direction of closed stand. The seedling density of shorter Quercus seedlings was not significantly related to the ordination axes, while taller seedlings seem to be more associated with LIV for light than O. carpinifolia or F. ornus. The density of species that were significantly related to the ordinal axes was not related to S. autumnalis coverage. Within the first (h < 20 cm), second (20 ≤ h < 130 cm) and third regeneration height class (h ≥ 130 cm), the average share of seedlings damaged by ungulates amounted to 16%, 43%, and 26%, respectively. The overall browsing damage rate was 30%. Virtually all tree species were most damaged in the second height class, as seen in Figure 4b. Among the species, Ulmus was the most damaged, followed by Quercus, F. ornus and O. carpinifolia.

Herb Vegetation and Indication of Ecological Factors
We recorded 28 species of trees, 31 species of shrubs, and more than 120 species of herbs, which is by far more than that recorded in similar research in other phytogeographic regions of Slovenia. Ordination of vegetation indicated two main gradients, as seen in Figure 5. The first seems to be influenced by stand type. LFH thickness, CWD, and rockiness increase in the direction of closed stand conditions, while LIV for soil moisture and minimum and average distance to seed trees decrease. However, the centroids of stand types are located close together, indicating that other factors are also important, e.g., geographical distances between the plots. The second gradient, which is independent of the first, indicates changes in LIV for reaction, continentality and light as well as slope and coverage of Sesleria autumnalis Scop. The density of shorter and taller F. ornus and O. carpinifolia seedlings increases in the direction of closed stand. The seedling density of shorter Quercus seedlings was not significantly related to the ordination axes, while taller seedlings seem to be more associated with LIV for light than O. carpinifolia or F. ornus. The density of species that were significantly related to the ordinal axes was not related to S. autumnalis coverage.

Drivers of Regeneration Dynamics
From the differences in ecological factors between the herb and Quercus subplots, we can deduce the peculiarities of the microsites where Quercus regenerated. Significant differences were found for topsoil moisture in September (p = 0.0429), Rubus sp. coverage (p = 0.0019), rockiness (p = 0.0099), and LIV for soil reaction (p = 0.0339). The Quercus subplots had slightly moister topsoil, were more overgrown by Rubus sp., were less rocky and had lower soil reaction. Herb species with the highest specificity (association) with herb subplots were Corylus avellana L., Salvia pratensis L., and Euphorbia nicaeensis All., while Rubus ulmifolius Schott, Rosa arvensis Huds. and Lamium galeobdolon L. were associated with Quercus subplots.
Models for the density of shorter Quercus and F. ornus seedlings predicted a significant decrease from the closed stand, through the open stand and gap edge, to the gap, as seen in Figure 6 and Table  2. The lowest density of all three main tree species was recorded within the gap. Quercus seedlings taller than 20 cm were most frequent in the open stand, followed by the gap edge. They required more light compared to shorter Quercus seedlings. Quercus competitors, F. Ornus and O. carpinifolia, Figure 5. Ordination biplot of a principal coordinate analysis (PCoA) for vascular plant cover in the herb layer (Bray-Curtis distance measure) in 2012. Grey circles represent the location of plots on the first and second axes of the PCoA analysis. The arrows show the direction of the gradient, and the length of the arrows is proportional to the correlation between the variable and the ordination. Only factors with a p-value < 0.05 are plotted, with the exception of rockiness (p = 0.1). Smaller angles between the arrow and axis represent a higher correlation between the two values. Green arrows represent regeneration density per species and developmental stage, e.g., F.orn.20 for F. ornus short seedlings. Dark blue arrows represent measured ecological factors, e.g., CWD coverage, LFH thickness, and minimum and average distance to Quercus seed trees. Light blue arrows are for Landolt phytoindication values for light, continentality, temperature, reaction, nutrients, and soil moisture.

Drivers of Regeneration Dynamics
From the differences in ecological factors between the herb and Quercus subplots, we can deduce the peculiarities of the microsites where Quercus regenerated. Significant differences were found for topsoil moisture in September (p = 0.0429), Rubus sp. coverage (p = 0.0019), rockiness (p = 0.0099), and LIV for soil reaction (p = 0.0339). The Quercus subplots had slightly moister topsoil, were more overgrown by Rubus sp., were less rocky and had lower soil reaction. Herb species with the highest specificity (association) with herb subplots were Corylus avellana L., Salvia pratensis L., and Euphorbia nicaeensis All., while Rubus ulmifolius Schott, Rosa arvensis Huds. and Lamium galeobdolon L. were associated with Quercus subplots.
Models for the density of shorter Quercus and F. ornus seedlings predicted a significant decrease from the closed stand, through the open stand and gap edge, to the gap, as seen in Figure 6 and Table 2.
The lowest density of all three main tree species was recorded within the gap. Quercus seedlings taller than 20 cm were most frequent in the open stand, followed by the gap edge. They required more light compared to shorter Quercus seedlings. Quercus competitors, F. Ornus and O. carpinifolia, indicated similar patterns for taller seedlings. The density decline towards the gap edge and gap center was more pronounced for both competitors, but especially for O. carpinifolia, suggesting a slightly higher shade tolerance in both species. These patters were consistent on the herb and Quercus subplots. indicated similar patterns for taller seedlings. The density decline towards the gap edge and gap center was more pronounced for both competitors, but especially for O. carpinifolia, suggesting a slightly higher shade tolerance in both species. These patters were consistent on the herb and Quercus subplots. The density of shorter Quercus seedlings was additionally positively associated with soil depth and LIV for continentality and negatively associated with LIV for nutrients, coverage of S. autumnalis and distance to Quercus seed trees, as seen in Table 2. The GLMM model for the subsample with measured diffuse light and topsoil moisture indicated largely the same pattern with a negative association of Quercus density with distance to seed trees and diffuse light intensity, with the exception of a positive association with coverage of F. ornus. The GLMM model for taller Quercus seedlings indicated a positive association with LIV for light, LIV for soil moisture, and distance to Quercus seed trees, and a negative association with rockiness. The model for the subsample indicated a negative association with rockiness and coverage of F. ornus.
The density of shorter F. ornus seedlings was positively associated with soil reaction and negatively associated with LIV for nutrients, continentality and light. The density of taller F. ornus seedlings indicated a positive association with LIV for soil reaction, coverage of S. autumnalis and LFH and a negative association with LIV for continentality, soil moisture, and flat relief. The O. carpinifolia short seedling sample was too small to be modelled. The model for taller O. carpinifolia seedlings indicated a negative association between density and LIVs for continentality, light and soil moisture, as well as for rockiness. The density of shorter Quercus seedlings was additionally positively associated with soil depth and LIV for continentality and negatively associated with LIV for nutrients, coverage of S. autumnalis and distance to Quercus seed trees, as seen in Table 2. The GLMM model for the subsample with measured diffuse light and topsoil moisture indicated largely the same pattern with a negative association of Quercus density with distance to seed trees and diffuse light intensity, with the exception of a positive association with coverage of F. ornus. The GLMM model for taller Quercus seedlings indicated a positive association with LIV for light, LIV for soil moisture, and distance to Quercus seed trees, and a negative association with rockiness. The model for the subsample indicated a negative association with rockiness and coverage of F. ornus.
The density of shorter F. ornus seedlings was positively associated with soil reaction and negatively associated with LIV for nutrients, continentality and light. The density of taller F. ornus seedlings indicated a positive association with LIV for soil reaction, coverage of S. autumnalis and LFH and a negative association with LIV for continentality, soil moisture, and flat relief. The O. carpinifolia short seedling sample was too small to be modelled. The model for taller O. carpinifolia seedlings indicated a negative association between density and LIVs for continentality, light and soil moisture, as well as for rockiness. Table 2. Results of the generalized mixed-effects model (GLMMs) analysis of natural regeneration density according to tree species predicted by factors (n = 477). Research plots were treated as random effects. Labels following factor names denote the factor levels for comparison, while the numbers in brackets within the results denote the standard error (SE).

Variable
Quercus

Discussion
In the black pine plantations of the Slovenian Karst, a significant density of Quercus (2700 ha −1 ) and noble broadleaves (1500 ha −1 ) was found, mostly in form of advance regeneration. Similar Quercus seedling densities in Pinus plantations have been reported in some studies [12,13]. Higher densities have been reported in studies from Balkan mountains [11] and from Poland [14], while in some other studies the densities were lower, likely due to more extreme sites and higher stand densities [15].
Since 2011 was an abundant mast year for Quercus [30], we expected many one-year-old Quercus seedlings. However, we did not record any one-year-old seedlings from any of the main tree species in 2012, which is probably due to the dry spring and summer of that year. The period from May to September in the research area had less precipitation than the 1971-2000 average [20]. This finding is consistent with the results of other authors who conclude that tree regeneration in Mediterranean forests is restrained by drought and therefore canopy provides a significant nurse effect for seedlings [31,32].
Quercus establishment was better within closed stands, on deeper silicate soils, with less S. autumnalis coverage and in proximity of seed trees. The importance of soil depth, ground vegetation competition and seed trees for Quercus establishment has been emphasized in several studies [11,13,33]. In addition to stronger seed rain under forest stands and less competition from ground vegetation, small rodents and birds, especially jays (Garrulus glandarius L.), are also important factors as vectors for acorns. It appears that they prefer forest stands rather than gaps [13,15].
Taller Quercus seedlings were more abundant in open stands and at gap edges, on more humid soils and on less rocky sites. Due to the deep root system of Quercus, competition with the grass S. autumnalis did not represent a major hindrance to already established Quercus seedlings. Many other researchers have stressed the relative abundance of Quercus advance regeneration in stands and problems with establishment in gap centers [11][12][13][14][15]33].
It seems that taller Quercus seedlings are more light demanding than competing F. ornus and O. carpinifolia; they may also better tolerate the influence of the continental climate within gaps (i.e., low relative air humidity, large daily and annual variation in temperature), especially Q. cerris and Q. pubescens [25]. However, due to the low Quercus density, it seems that gaps in this study were too large and were not made gradually following established Quercus advance regeneration. In the sub-Mediterranean region, special care is needed when regenerating stands, as gaps can also be problematic due to summer droughts [32]. Unlike other studies of topsoil moisture distribution within gaps [34,35], measurements in this study indicated that seedlings in gaps were more exposed to drought during summer months compared to closed stand (comp. [36]), despite the absence of interception by crowns and water consumption by tree roots.
An important difference between Quercus and its two main competitors is that it thrives better on deeper and more acidic soils. While the proportion of Quercus declined with height class, the proportion of F. ornus was more stable and O. carpinifolia increased in number. The latter is probably partly due to its higher shade tolerance and lower mortality, and partly also due to the successional development of stands and improved soil conditions. The ratio of Quercus to competing tree species is strongly in favor of the latter; therefore, tending and mixture regulation in favor of Quercus is needed [12].
In addition to soil conditions, overbrowsing by ungulates represents another factor that is not directly influenced by silviculture. Browsing of regeneration by ungulates exceeded the recommendations for sustainable natural regeneration [37], and in particular hampered the development of Quercus. The association of Quercus subplots with thorny species (Rosa sp., Rubus sp.) may also indicate an escape strategy for this species [38]. Thus, successful recruitment of seedlings would require a significant decrease in game density.

Conclusions
This study did not cover all gap sizes. The task of future research is to consider a larger range of gap sizes, which will be available with the further development of gradual conversion. The guidelines presented are therefore largely based on past Quercus regeneration studies [12,[39][40][41] and observations and should be followed with some caution. Contrary to the common opinion that Quercus seedling potential is low in P. nigra plantations, this study indicated that gradual conversion by means of natural regeneration is feasible. This is especially true for less degraded sites with deeper, silicate soils. Quercus establishment is successful under closed stands (e.g., 15-20% of diffuse light). Regular thinning of P. nigra stands is important to favor stability and facilitate recruitment of advanced regeneration. Later, with the development of Quercus, small gaps (e.g., 20-30% of diffuse light) should be created and progressively expanded. Since Quercus seedlings can survive up to 25 years below P. nigra canopies, the interventions can be gradual. Low advance regeneration density and overbrowsing require strict protection of seedlings from harvesting and browsing damage as well as regulation of tree competitors. If tree commercial quality is a concern, following the thicket phase, the gap size should reach at least 0.5 ha due to the heliotropic growth of Quercus. It seems likely that at current game densities active gradual transformation of P. nigra plantations is not possible. Therefore, in order to promote stable mixed future forests, in addition to the described silvicultural methods, it is also necessary to regulate ungulate populations. List of sampled tree species and shrubs on vegetation subplots. Frequency represents the proportion of subplots where a species was present; regeneration means overall regeneration structure.