Since 2004, we have selected four study sites in the Dinaric Alps by following strict criteria (Table 1): Lom (LOM), Janj (JAN), and Perućica (PER) in Bosnia-Herzegovina and Biogradska Gora (BGO) in Montenegro (Figure 1; Table 2). We have focused our selection on mixed, uneven-aged montane forests belonging to the Abieti-Fagetum forest type (Figure 2). During the last 20 years, more than 40 researchers from 7 countries have been studying forest dynamics and ecological processes at different scales and according to a multidisciplinary approach (Table 3).

LOM and JAN are Forest Reserves established by the State Institute for Protection of Cultural Monuments and Natural Rarities in Bosnia and Herzegovina (Maunaga et al., 2001). LOM reserve was set in 1956 (Pintarić, 1999; Motta et al., 2011). Initially, the Reserve was split between a core area of 55.8 ha under a strict protection and a buffer zone of 242 ha with low-intensity management. More recently (2021), strict protection was extended to the whole LOM reserve. JAN reserve was established in 1954 (Pintarić, 1999; Keren et al., 2017). As LOM, JAN was initially split between a strict protection core area (57.2 ha) surrounded by a buffer zone (237.8 ha) that were merged in 2021. In the same year, JAN was named a UNESCO World Heritage Site as an extension to the Ancient and Primeval Beech Forests of the Carpathians and Other Regions of Europe, and a new 295 ha buffer zone was delineated.

PER and BGO are among Europe’s most significant and best-preserved OGF (Sabatini et al., 2018). At both sites, the OGF covers more than 1,000 ha, including different forest types along an altitudinal gradient. The Abieti-Fagetum is the most represented one, covering more than 500–600 ha at both sites. The first official “protection” measures for these forests date back to 1893 for PER (Fukarek and Stefanović, 1958; Pintarić, 1978) and 1878 for BGO (Motta et al., 2015a; Stupar and Milanović, 2017), respectively. In both cases, the primary protection goal was hunting and game protection. BGO was declared a National Park (Biogradska Gora National Park) in 1952, while PER acquired the same status in 1962 (Sutjeska National Park).

All the sites are characterized by high fertility (Drinic, 1957), are among the highest carbon reserves in Europe (Keith et al., n.d.), and host some of the tallest European native trees, reaching more than 60 m height in PER and BGO and more than 55 m in LOM and JAN (Leibundgut, 1976; Diaci, 1999).

We collected one core from the living trees with DBH >10 cm from all the trees in the IPP of LOM and from 75% of the trees, randomly selected, in the IPP of PER. The cores were collected at 50 cm height from the ground to better estimate tree age and reconstruct the stand age structure. They were also used to analyze growth patterns and reconstruct disturbance history at the stand level. In addition, in each research sites, 12–30 dominant trees randomly selected in the study areas were cored (two cores for each tree) at DBH to analyze climate growth relationships (Castagneri et al., 2014; Bosela et al., 2017; Martinez del Castillo et al., 2022) and to estimate the maximum age reached by dominant trees of each species. In BGO, due to the restrictions posed by the National Park Administration, only standing or lying dead trees were cored, resulting in an under-representation of dominant Norway spruces.

All cores were fixed to wooden supports in the laboratory and prepared with a razor until an optimal surface resolution was achieved. Tree ring widths were measured to the nearest 0.01 mm using the LINTAB device and TSAP-Win software package (Rinntech, Heidelberg, Germany). The tree-ring series were visually and statistically cross-dated by CooRecorder software (Larsson and Larsson, 2018).

To describe past disturbance events, we analyzed the growth pattern in cores from permanent plots to identify releases from suppression. The detection of growth releases was obtained with Jolt software (version 6.01P by R. L. Holmes, University of Arizona, Tucson, Arizona), using the running mean radial-averaging method (Rubino and McCarthy, 2004; Fraver and White, 2005) and taking into account abrupt growth increment over 166% of the previous years in a window of 10 years. Only the growth increments that last at least 4 years were classified as releases from suppression (Schweingruber et al., 1990). A chronology of the releases found in the analyzed trees, all the species together, was constructed for each site. As the time to respond to stand opening is highly variable among species and individuals, the results were grouped at 10-year intervals (Altman, 2020; Izworska et al., 2022). The percentage of trees showing a growth release in a given decade was then plotted against time (Motta et al., 1999) starting from 1700 when at least 15 trees were available in both sites.

For age estimation, tree age from cross-dated cores was taken as the number of rings between the pith and the cambium (age at the coring height) when cores included the pith. In the other cases, if possible, the pith location was geometrically estimated (Motta and Nola, 2001). The missing radius was measured as the distance from the estimated pith location and the innermost ring. Then, missing rings to the pith were estimated by counting the number of rings in a segment equivalent to the length of the missing radius in the innermost part of the core. This number was added to the number of rings counted on the core to obtain the estimated tree age at the coring height. When it was impossible to estimate the age, e.g., for the decay or because the missing part could not be estimated reliably, we used the number of rings counted, indicating the presence of a missing part not counted (e.g., >480 years).

Between-plot and between-forest variability was explored using a multivariate approach based on PCA free ordination analysis employing a single matrix (142 plots and 20 structural variables). Forest structure variables for living trees (basal area, density, volume, quadratic mean diameter, standard deviation of diameter, Shannon diameter diversity), coarse woody debris (overall CWD volume, volume of stumps, logs, snags having a diameter >47.5 cm), and regeneration (overall density of saplings and divided by species) were used to assess the between-forest and the between-plots variability. Species composition was measured as a proportion of single species (beech, silver fir, Norway spruce, and others) on the plot basal area. A particular focus on large (DBH >80 cm) and medium-size (DBH >50 cm) trees was adopted by measuring the densities of these two diameter classes. PCA was performed to explore the correlation structure among living trees, regeneration, and CWD attributes. The average structures of old-growth forests were indicated by centroids, and convex hulls indicated their range of variability. PCA was performed using the PC-ORD 7.10 (McCune and Mefford 1999) statistical package. The axes’ statistical significance was tested using the Monte Carlo permutation method based on 10,000 runs with randomized data.

The density of living trees with a diameter greater than or equal to 7.5 cm was between 412 and 516 trees per hectare in BGO and JAN, respectively (Table 4). The basal area and volume ranged from 45.3 to 66.7 m² ha⁻¹ and from 747.3 to 1201.9 m³ ha⁻¹ in LOM and JAN, respectively, with JAN having the highest values and LOM having the lowest. It is worth noting that the average volume data from LOM is influenced by the karstic morphology with sinkholes that limit the occupancy of the forest cover. In fact, the 1 hectare permanent plot in LOM, situated on a relatively uniform slope, has a volume of 1,116 m³, which is the same order as the other three sites (Motta et al., 2011). The beech accounts for 19.6 to 38.3% of the basal area in JAN and BGO, respectively. Silver fir is the dominant species at all sites, ranging from 46.6% in LOM to 59.8% in PER. The incidence of Norway spruce is higher in the northern sites (LOM and JAN) than in the southern ones (PER and BGO). Additionally, all sites have an impressive amount of CWD, ranging from 305.0 m³ ha⁻¹ in LOM to 410.7 m³ ha⁻¹ in BGO. Despite having high densities and volumes, all sites are relatively rich in regeneration, with LOM having the highest density with 4,772 individuals per hectare and PER having the lowest with 2,312 individuals per hectare. The carbon stocks range from 398.0 MgC ha⁻¹ in LOM to 484.0 MgC ha⁻¹ in JAN. The diameter distribution (Figure 3) has the structure of the uneven-aged forests and has a rotated sigmoid shape (Motta et al., 2015b) which is the typical distribution of the old-growth stage (Goff and West, 1975; Lorimer et al., 2001). The stature (i.e., the average of the five highest trees) a fertility index for uneven-aged and multilayered forests (Susmel, 1980) ranges between 55.8 m in BGO and 46.8 m in LOM (see Supplementary material for further details).

The in-depth study of the two IPP in LOM and PER has shown a density of large trees (DBH >1 m) of 19 trees per hectare in LOM (11 silver firs and 8 Norway spruce) and 14 trees in PER (11 silver firs and 3 Norway spruce). The largest beech has a DBH of 82 cm and 90 cm in LOM and PER, respectively (see Supplementary material for further details).

We detected 564 releases in the IPP of LOM (493 trees analyzed) and 469 releases in PER (296 trees analyzed). The mean percentage of release per tree and the percentage of trees with release are slightly lower in LOM (about one release/tree and 64% of trees with release) than in PER (about 1.5 release/tree and 80% of trees with release).

Eighty-seven percentage of the decades in LOM and 77% of the decades in PER have less than 10% of the trees showing a release (Figure 4). At both sites, 13% of the decades have between 11 and 20% of trees with releases. In LOM, there is a low and regular distribution of the percentage of released trees, with a maximum percentage of about 13%. In PER, the pattern is characterized by slightly higher inter-decadal variability. However, the percentage of trees with release is over 20% (maximum 26%) only in two decades: 1880 and 1960. These conditions of low and quite regular % of releases for each decade have been relatively stable in both forests for the last three centuries.

The age of dominant trees by species and site is summarized in Table 5. The three species showed similar maximum ages (except for Norway spruce in BGO affected by sampling limitations), and the age of oldest trees are near to the maximum biological age of the species (Castagneri et al., 2013; Piovesan and Biondi, 2020). Maximum age is achieved by silver fir in PER (487 years) and by beech in BIO (486 years). For Norway spruce, we cannot estimate the age of the oldest trees, neither in LOM nor in PER, due to the decay, but the maximum age is of the same order (>454 years in PER). As the estimated age is related to the coring height (130 cm above the ground), we can suppose that, taking into account the years employed by the tree to reach the sampling height, the oldest dominant trees should have an age >500 years. The two 1 ha sampling plots (see Supplementary material for further details) have a density of 15 trees ha⁻¹ having >400 years at the sampling height (50 cm) in LOM (2 beeches, 7 silver firs, and 7 Norway spruces) and 11 trees ha⁻¹ in PER (4 beeches, 5 silver firs, and 2 Norway spruces).

The forest structure of the four old-growth forests of the Dinaric Alps suggests some similarities and differences among sites. The difference is expressed by the distance among their centroids in the ordination environment and the similarities are expressed by the overlap among convex hulls polylines indicating the maximum surface area occupied by plots belonging to the same site: blue JAN, yellow LOM, grey BGO, and green PER (Figure 5). Considering the whole surface area occupied by the convex hulls, we observed that overlapping sites occupied 61.1% of this surface area and 15.3% was the proportion occupied by all the 4 sites. Instead, only 38.9% of this surface area was occupied by non-overlapping convex hulls. PER and BGO emerged as being quite similar regarding forest structure expressed by many structural data but particularly by a higher quantity of CWD and a lower proportion of Norway spruce. LOM and JAN were characterized by denser (De) and diverse (HD) stands with a higher density of regeneration (R-tot) and a higher volume of stumps (most of the stumps are of natural origin but in JAN and LOM there are also a few signs of human activity at the borders between the former core area and the buffer zone). JAN resulted separated from LOM along a gradient of overall basal area (BA), volume (Vol) and a lower proportion of beech. Therefore, the structural variability is PER > BGO > JAN > LOM. The overlapping is very high between PER and BGO and decreases following a latitudinal gradient towards JAN and LOM. The surface area proportion occupied by the convex hulls of the 4 study sites was 12.3% (LOM), 22.1% (JAN), 26.3% (PER), and 39.4% (BGO).

All the selected study sites fit in the strict old-growth structural criteria reported in Table 1. The four sites have the highest volume of living trees for the mixed montane forests of the central-southern part of Europe (Table 6). The high quantity of living biomass is typical of the old-growth stage. However, this characteristic is not an indicator of naturalness itself since high levels of biomass can also be found in managed forests or in forests that are in development stages (e.g., mature or transition) earlier than the old-growth one (Castagneri et al., 2008; Molina-Valero et al., 2021). The four sites have a stratified vertical structure, an uneven-aged age structure with low relationships between age and diameter (Motta et al., 2011), a rotated sigmoid diameter distribution (Motta et al., 2015b), and a clumped horizontal structure (Lingua et al., 2011). All these structural characteristics are peculiar of the old-growth developmental stage. The tree species approach their maximum age expectancy not with sporadic individuals but with many dominant trees (15 ha⁻¹ older than 400 years in LOM and 11 ha⁻¹ PER) and are richer in large and old trees (Table 5) compared to other OGF of the Carpathians (Pavlin et al., 2021).

All the study sites have impressive amounts of CWD ranging between 305.0 and 410.7 m³ ha⁻¹ (Table 4). The high quantity of CWD is peculiar to the old-growth stage and is the legacy of the long-term small-scale disturbance history that has characterized the dynamics in the last centuries (Lombardi et al., 2012; Mikolas et al., 2021; Motta et al., 2023). Besides, all sites have a high density of large (>50 cm DBH) snags (ranging between 18 and 27 ha⁻¹) and logs (between 33 and 46 ha⁻¹ with largest diameter >50 cm) (Table 4).

The high regeneration density is related to the fact that all the dominant species are shade-tolerant and can regenerate and establish below the forest cover and to the small-scale disturbances that create gaps that allow the regeneration to establish and grow continuously in the stand. The canopy gap fraction in the studied sites is 19.3% in LOM, 17.2% in PER, and, in extended gaps fraction, 41.4% in LOM, and 39.7% in PER (Nagel and Svoboda, 2008; Bottero et al., 2011).

Tree rings provided detailed insight into temporal patterns of forest disturbance history in LOM and PER. In both forests, most decades in the last three centuries experienced less than 10% of canopy opening signs (release from suppression). We have observed in one decade a maximum of 16% of trees showing a release in LOM and 26% in PER, with only two decades with >20% of trees showing a release in PER. Thus, the dynamics of these stands have been strongly controlled by small-scale disturbances that have maintained a typical small-gap structure for a relatively long time. According to this small level of disturbances, these stands can be classified at the limit of the range of dynamics, from disturbance-structured to near steady-state (Antos and Parish, 2002; Halpin and Lorimer, 2016).

The methodological approach used here has the potential to reconstruct the disturbance history carefully, but conversely, this can be done at a relatively small scale, in this case one hectare (Nagel et al., 2014). Thus, these results have to be taken with caution as, if we had expanded the structural analysis at the landscape scale, we could have detected some legacies of past infrequent intermediate severity-disturbances from fire (PER), wind (PER, BGO, LOM) and from small landslides and snow avalanches (BGO), large enough to favor the regeneration of more light-demanding tree species like maple and other broadleaves (Nagel et al., 2014; Motta et al., 2015a). Besides, more intensive past land use by the local population has been observed locally and at the regional scale during the middle ages (Finsinger et al., 2022; Cagliero et al., 2022b). The extensive overlap between convex hulls (Figure 5) confirms the commonality of structures and processes already observed in the structural data and the disturbance regime.

The current forest cover of OGF in Europe is residual (or even absent for many forest types, particularly in temperate and Mediterranean forests), since most European forests have been intensively managed and, even if withdrawn from regular management, show structural legacies of the past land use. After the cessation of the management, some structural elements can recover faster, and, in contrast, others recover only in the long term. For example, large dead trees, old trees, rotated sigmoid diameter distribution, gap fraction, development of microhabitats (von Oheimb et al., 2005; Paillet et al., 2010; Meyer and Schmidt, 2011; Albrich et al., 2021; Kõrkjas et al., 2021) and the time framework of the restoration (Frelich et al., 2005; Nagel et al., 2013) is often measured not by decades but by centuries (Motta et al., 2015b; Albrich et al., 2021; Meyer et al., 2021; Martin-Benito et al., 2022; Larrieu et al., 2023).

The criteria adopted by the recent EU Guidelines (European Union Commission, 2023a) “Defining, Mapping, Monitoring and Strictly Protecting EU Primary and Old-Growth Forests” are relatively broad and not well-defined. Consequently, most of the selected and mapped OGF according to the EU guidelines could have a low rate of old-growthness and could not adequately represent the final stage of the forest dynamics. The EU approach considers as OGF those stands “relatively old and relatively undisturbed by humans.” As such, it identifies a critical amount of forests mostly represented by “future potential old-growth” (Spies, 2004), but, even if it is an appropriate approach to protect the residual biodiversity related to late seral developmental stages, has limits both for strict protection of current OGF and for those scientific goals that need an absolute reference of the final stage of the forest dynamic processes. The priority of biodiversity conservation would be to accurately select the “current” old-growth stands, analyze their distribution, and protect them at the continental scale. These forests represent a residual fraction of the continental forest cover and deserve a different recognition and designation among the massive selection made by the EU Guidelines. A fundamental issue is related to the scientific purposes connected to the old-growth designation. Indeed, for scientific research, only OGF selected with strict criteria can be reliably used as a reference or benchmark (Table 7). Another issue to be taken into account when research and benchmarking are involved is the need to assure long-term potential conservation of the whole range of variability (Hansen et al., 1991) inherent to the complete forest dynamics, including stand-replacing disturbances and early seral stages. For this purpose, it would be necessary to define a minimal critical size of old-growth sites (Landres et al., 1999; Veblen, 2003; Peck et al., 2015). The Abieti-fagetum is not characterized by high severity and large disturbance events (Senf and Seidl, 2021), and in the studied Dinaric forests, we have not observed stand-replacing disturbances larger than 1 ha, even at the landscape scale (Garbarino et al., 2012; Motta et al., 2015a). However, considering a comprehensive variability covering the European range of the montane forest types, we have to take into account the occurrence in this forest type of high-severity stand-replacing events larger than 15 ha (Maroschek et al., 2023). In a temperate primary forest, depending on the longevity of the trees and on the persistence of each developmental stage, the percentage of the forest covered with old-growth stage can vary approximately from 25 to 60% (Swanson et al., 1994; Oliver and Larson, 1996). As a consequence, we can hypothesize, in a prudential way and according to the information currently available, a size of at least 25–30 ha to host the current and future entire range of variability of the different developmental stages of the forest dynamics (Landres et al., 1999; Keane et al., 2009) and to guarantee long-term conservation of the whole biodiversity associated to the forest type. A more detailed study will be necessary to explore the range of variability in the European forest types as natural disturbance regimes and interactions among disturbances at different spatial and temporal scales are poorly understood (Landres et al., 1999). Besides, future climate change impacts can significantly impact the current framework (Seidl et al., 2017; Thom et al., 2017).