Introduction

Forests and woodlands cover one-third of the global land surface (FAO and UNEP 2020), providing multiple benefits to society and contributing to sustaining human well-being (Kok et al. 2017; Haines-Young and Potschin 2018; Orsi et al. 2020). Despite the important benefits that forests grant, the provision of these services is threatened by climate change, over-harvesting and by the loss, degradation and fragmentation of forest landscapes and ecosystems (FAO 2015; Felipe-Lucia et al. 2020). In this context, sustainable forest management—management that concomitantly maintains forest biodiversity, productivity, regeneration capacity and vitality, as well as forests’ potential to fulfill a wide range of functions and services (Ministerial Conference on the Protection of Forests in Europe 2007)—has been recognized as crucial to circumvent biodiversity loss. Indeed, forest biodiversity (CBD 2006) plays a crucial role in the variety of benefits forests can supply (Mori et al. 2017; Brockerhoff et al. 2017) thanks to the interconnection of its components, including structure (Gao et al. 2014; Bohn and Huth 2017; Ćosović et al. 2020; Toivonen et al. 2022), composition (Drever et al. 2006; Gao et al. 2015) and functions (Hilmers et al. 2018). Yet, the complex nature of ecosystems also poses challenges for biodiversity assessment (McElhinny et al. 2005; Schleicher-Tappeser 2018). In fact, an agreed definition of biodiversity still does not exist (Camia et al. 2023), but as is clear from Lyashevska & Farnsworth (2012), we can refer to forest biodiversity as a measure of the total complexity within a forest habitat considering the total variety of life forms (Wilson 1988).

In such a context, estimating biodiversity is challenging, and the scientific community has been actively engaged in trying to define the different components and dimensions of biodiversity for purposes of identifying variables that can be used for estimating and reporting spatial and temporal biodiversity trends (Pereira et al. 2013).

In this context, assessing forest biodiversity requires evaluating the extent to which forests are influenced by natural processes and free from human intervention (Reif and Walentowski 2008; Winter 2012; Côté et al. 2021), also known as naturalness. Assessing forest naturalness not only helps in accurately estimating and reporting the ecological condition of forests for biodiversity protection, but also plays a crucial role in establishing objective criteria for identifying natural, old-growth forests (Buchwald 2005). In fact, old-growth forests, which exhibit structural attributes and successional processes resulting from minimal human management and disturbances, play an essential role in biodiversity conservation (Paillet et al. 2010). Additionally, their complex structures support the abundance of microhabitats (Kozák et al. 2018; Parisi et al. 2021). Identifying the characteristics of old-growth forests–here referred to as old-growth status—is crucial for searching for potential gaps in the network of protected areas (European environment agency 2014; Côté et al. 2019), particularly in Europe where human impact has historically been substantial (Potapov et al. 2017). Therefore, gaps in monitoring can lead to the ineffective implementation of protection policies (Ette et al. 2023). This may explain why, despite international conventions such as the United Nations Strategic Plan for Forests 2030 (United Nations 2019) and the EU Biodiversity Strategy for 2030 (European Commission 2020), current rates of biodiversity loss remain large (Waldron et al. 2017; Ette and Geburek 2021). Indeed, to effectively achieve forest ecosystem conservation goals, it is crucial to have technically sound and economically viable methods that include well-defined criteria and indicators to support sustainable forest management and planning at multiple levels: local, regional and national (Heink and Kowarik 2010; Maes et al. 2018; FAO 2020; Lier et al. 2022).

In this context, NFIs may play a crucial role in evaluating and estimating indicators related to forest biodiversity, naturalness and old-growth status (B-N-OG) (Corona et al. 2011; Storch et al. 2018), given the extensive data routinely collected from numerous sample plots (Heym et al. 2021). The number of studies that have assessed the utility of NFI data for monitoring biodiversity, naturalness or old-growth status is relatively small, and none has compared estimates of B-N-OG simultaneously using data for an entire country-level NFI (Winter 2008; Corona et al. 2011; Pignatti et al. 2012; Chirici et al. 2012; Storch et al. 2018; Heym et al. 2021).

It is important to underline that in this study we inherited the definitions used in quantitative social sciences (e.g., Stockemer et al. 2018): The term “indicator” is used when its calculation is based only on one variable measured or estimated, while the term “index” is used when its calculation is based on multiple indicators. So, for example, the total volume of deadwood is an indicator that can be aggregated with others to estimate a biodiversity index. The objective of this study was to test a set of indicators satisfying three criteria : (i) they must be able to estimate aggregated indexes of B-N-OG, (ii) they should be supported by substantial scientific evidence, and (iii) they can be estimated using existing NFI datasets, thereby ensuring their compatibility with available data sources (Saint-André and Hervé 2015). After a comprehensive literature review, we selected 18 indicators that were then used to estimate B-N-OG indexes using raw data from 6563 plots visited in the field in the framework of the 2005 Italian NFI (Gasparini and Tabacchi 2011). To evaluate the potential usefulness of these indicators, we estimated pairwise correlations among them and the three aggregated B-N-OG indexes, their spatial distribution and relationships between the indicators and the indexes separately for protected areas and individual forest type categories. Finally, we used the results of our analysis to propose integration of new NFI variables to facilitate and optimize B-N-OG assessments.

Materials

Study area

Our study is conducted over the entirety of Italy which is divided into 20 Administrative Regions and 107 provinces. Forests in Italy cover nearly 11 million ha—more than one-third of the total Italian territory (INFC 2007a, b)—and are mainly found in mountainous, hilly and interior regions. Italian forests are composed mostly of deciduous species (68%), mainly oaks (Quercus. cerris L., Q. pubescens W., Q. ilex L., Q. petraea (M.) L., Q. robur L.), European beech (Fagus sylvatica L.) and chestnut (Castanea sativa M.). While dominant conifers are Norway spruce (Picea abies K.) and pines, the latter are mostly located in mountainous (e.g., Pinus sylvestris L. and P. nigra A.) and coastal areas (e.g., P. pinea L., P. pinaster A. and P. halepensis Mill.). Overall, due to its latitudinal and altitudinal variability, Italy is a very biodiverse country, hosting half of the European forest types (Barbati et al. 2014).

Italian national forest inventory

The Italian NFI was designed to satisfy the reporting requirements set by the United Nations Framework Convention on Climate Change. NFI estimates play a crucial role in various national and international reporting processes, including Italy's national forest report, Kyoto Protocol, national natural capital report, European report on sustainable forest management and the FAO’s global Forest Resource Assessment. The 2005 Italian NFI has three phases and is based on an unaligned systematic sampling design (Fattorini et al. 2006) featuring a field plot randomly located in each cell of a 1 km × 1 km systematic grid (INFC 2007a, b). In the first phase, 301,000 sample points are photo-interpreted to identify their land use/land cover classes consistent with the first level of the CORINE Land Cover nomenclature system (Bossard et al. 2000) and with the FAO forest definition (Gasparini and Tabacchi 2011; FAO 2020). In the second phase, a subsample of approximately 30,000 forest and other wooded land sample points was selected from the first-phase set and visited in the field to assess forest categories and other qualitative information. Lastly, a third-phase sample of approximately 7,000 sample points was selected from among the 30,000 second-phase points. These points became the centers of 13-m radius NFI plots that were visited in the field to acquire quantitative data. The resulting database was used with statistically rigorous, unbiased design-based estimators to produce official estimates at both regional and national levels for approximately 50 qualitative and quantitative forest variables, mainly developed to estimate wood resources and the amount of carbon stocked and removed from the atmosphere.

In this study we used the data collected in the field during the third phase of the 2005 NFI cycle. After removing data for inaccessible plots and plots in temporarily unstocked areas, data for 6563 plots classified into 20 forest categories remained (Fig. 1). In Appendix, section A, Table 2 we report and explain the available raw 2005 Italian NFI data that we used.

Fig.1
figure 1

Spatial distribution and relative frequency of Italian NFI (INFC 2007a, b) forest type categories

Full size image

Copernicus high-resolution layers: imperviousness degree

To assess forest naturalness, along with other single-variable indicators, we used an indicator of hemeroby that is based on the geographical distance of an NFI plot from the nearest impervious map unit as mapped by Sentinel-2 imagery in the Copernicus High-Resolution Layer, denoted as Imperviousness Degree (HRL IMD) for the year 2006 (Congedo et al. 2016; European Environment Agency 2018), (https://land.copernicus.eu/pan-european/high-resolution-layers/imperviousness, last accessed on September 13, 2022). The HRL IMD consists of a 20-m resolution pan-European raster layer produced using a semi-automated classification, based on calibrated NDVI (Kuc and Chormański 2019). The environmental indicator mapped by HRL IMD is defined as “human-produced surfaces that are essentially impenetrable by rainfall” (Moglen and Kim 2007; Strand 2022). The HRL IMD map was reclassified into a Boolean impervious/non-impervious mask, and the distance from the nearest impervious map unit was calculated for each NFI plot.

Italian network of protected areas

To study the potential relationship between the selected B-N-OG indexes and ecological and management conditions, we included the Italian protected areas in our analysis. We considered both the official dataset of protected areas from the Italian Ministry of Environment (EUAP—Elenco Ufficiale delle Aree Protette) and the Natura 2000 network.

The EUAP includes 24 National Parks, 144 State Natural Reserves, 134 Regional National Parks, 365 Regional Natural Reserves and 171 Other National Protected Areas (Fig. 2, panel a) for a total of 2,878,963 ha, in total approximately 11% of national territory (Boitani et al. 2003). The Natura 2000 consists of 2,314 sites (Fig. 2, panel b) established based on the "Habitats" and "Birds" European Directives (Council Directive 92/43/EEC and Directive 2009/147/EC, respectively), covering 5,844,915 ha, or approximately 19% of the national territory (Ministry of Environment and Energy Security 2022).

Fig. 2
figure 2

Italian network of protected areas. Panel a (left) reports the Italian protected areas according to the EUAP; on panel b (right) different colors represent the Natura 2000 network in relation to the year of institution. On the background, the Copernicus High-Resolution Layer “Tree cover density” (ESA, 2018)

Full size image

The two networks have substantial overlap. Almost 40% of the protected areas were included in both the EUAP and Natura 2000 areas. For this reason, we considered only the classification protected versus non-protected areas, independently of whether a plot was in EUAP or in Natura2000 area. Of the 6,563 NFI plots, 67% were outside protected areas, while the remaining 33% were included in EUAP or Natura 2000 areas.

Methods

Based on a comprehensive literature search of scientific publications and reports, we defined 18 indicators (Sect. "Plot-level indicators used to assess B-N-OG"), each of which was further classified as relevant or not relevant for the three biodiversity indexes, B, N and OG. The indicators were then calculated for each of the 6,563 NFI plots (Sect. "Statistical analysis"). To facilitate aggregation for estimating the indexes, we normalized the indicators to the range 0–1 by applying a min–max normalization, the same approach used by Storch et al. (2018). The three B-N-OG aggregated indexes were then estimated for each plot as the average of the selected indicators.

Finally, the resulting database was analyzed to investigate multiple relationships: i) pairwise correlations among the different indicators, ii) spatial distributions to highlight geographical trends, iii) distributions for individual forest type categories and management types to better understand different levels of forest biodiversity, naturalness and old-growth status in distinct ecosystems and iv) the effects of the network of protected areas.

Plot-level indicators used to assess B-N-OG

In the following sections we describe the indicators selected and their relevance for B-N-OG assessment. We classified the indicators into five categories: (i) composition, (ii) structure, (iii) regeneration, (iv) deadwood and (v) hemeroby (Table 1).

Table 1 Single-variable indicators and how they were combined to create biodiversity, naturalness and old-growth status (B-N-OG indexes), with main references
Full size table

(i) Composition

First, we calculated the number of tree species, a very straightforward and frequently used indicator of biodiversity (Van Den Meersschaut and Vandekerkhove 2000; Sullivan et al. 2001; Crist et al. 2003).

The distribution of tree species was then characterized using the Shannon and the Evenness indexes; both calculated based on the basal area. The Shannon index, commonly used in ecological studies, served as a direct measure of compositional diversity (Varga et al. 2005; Lexerød and Eid 2006; Ozdemir et al. 2008; Arekhi et al. 2017). The Evenness index, which describes the strength of interactions within communities and functional trait diversity, was calculated as the ratio of the observed Evenness index to its maximum value with the same number of species (Pielou 1969).

(ii) Structure

For structural diversity we first considered growing stock volume (GSV) as reported by the NFI (Tabacchi et al., 2011). Then we considered the variability of tree height as estimated by the standard deviation of height as a measure of vertical forest layering (Zenner 2000). In addition, Stand Density Index (SDI) (Reineke 1933) was used to describe the density of living trees on each NFI plot. SDI is characterized in terms of quadratic mean diameter and number of trees per hectare by calculating the number of stems per hectare in these stands related to 25 cm mean diameter. SDI is independent of site conditions and tree age (Zeide 2005), the latter not commonly recorded during NFI field measurements.

The presence of large trees is often considered a relevant indicator for B-N-OG assessment (Basile et al. 2020; Asbeck et al. 2022; Marziliano et al. 2021; Larrieu et al. 2021, 2022; Santopuoli et al. 2022). Tree dimensions can be considered a proxy for the age of the trees, and the presence of old trees is a direct measure of OG. Large trees are also more frequently habitat trees, and the tree-related microhabitats they host are of primary concern for forest biodiversity because they can harbor many endangered specialized species of flora and fauna. To estimate tree size, we considered both maximum tree height and maximum DBH for all living trees. We also considered a frequency index at the plot level which determines the occurrence of living trees with DBH≥40 cm. While the threshold for identifying large trees could be differ by country (Gilhen-Baker et al. 2022), the Italian Environment Ministry proposed a 40-cm threshold for living trees in the Mediterranean area; we used this threshold for current study (Blasi et al. 2010).

Lastly, the Gini coefficient, widely used in forestry, was used to analyze diversity in the distribution of tree sizes (Valbuena et al. 2016; Meyer et al. 2021; Motta et al. 2022; Hirschmugl et al. 2023)

(iii) Forest regeneration

Forest regeneration has been recognized as an indicator of forest biodiversity within the guidelines of sustainable forestry issued through the Ministerial Conferences on the Protection of Forests in Europe and the Montréal Process (Ministerial Conference on the Protection of Forests in Europe 2007; Montréal Process 2009). For this study, regeneration was estimated using three indicators based on the methods and definitions used in the Italian NFI. The first indicator of regeneration refers to the number of small trees with heights between 50 and 130 cm; the second indicator is for small trees taller than 130 cm but with maximum DBH of 2.4 cm, and finally the third indicator was similar to the second but with a maximum DBH of 4.4 cm.

(iv) Deadwood

The amount of deadwood volume plays a crucial role for assessing B-N-OG because it influences the presence of bryophytes, lichens and saproxylic beetles and is also used for assessing N and OG (DeWalt et al. 2003; Lassauce et al. 2011; Parisi et al. 2016, 2019, 2020a; Czerepko et al. 2021). In this study we also considered the ratio of dead-to-living biomass because it contributes to understanding ecosystem dynamics (Olson 1963; Harmon et al. 2001; Hahn and Christensen 2005; Wirth et al. 2009a, b). The assessment of deadwood volume includes use of a classification into five classes (Maser et al. 1979) that offer valuable information for assessing the development stage of forest stands (Bertini et al. 2010; Parisi et al. 2019, 2020a). In addition, it is associated with the presence of diverse organisms that utilize trees at different decay stages (Franklin 1981; Gibbons and Lindenmayer 2002; DeWalt et al. 2003; Vuidot et al. 2011), because it is useful for assessing the impact of deadwood on soil carbon storage (Błońska et al. 2023). For these reasons, we included two indicators, one for stumps and one for standing dead trees, each calculated as a ratio between the volume in more decayed classes divided by the volume in the less decayed classes.

(v) Hemeroby

Hemeroby can be considered as the opposite of naturalness and is used to measure the impact of humans on the ecosystems (Chirici et al. 2012). As an indicator of hemeroby for this study, we calculated the distance of each NFI plot from the nearest impervious map unit (urban areas, artificial surfaces) as mapped by the European Earth Observation Copernicus program. Firstly, we constructed a Boolean impervious/non-impervious map reclassifying the Copernicus High-Resolution Layer—Imperviousness Density map (European Environment Agency 2018) for the year 2006 at 20 m resolution. Then, we constructed a map of the distance from the artificial surface class of the Boolean map using a maximum distance of 10 km and extracted a distance for each NFI plot. The values of the indicator were standardized on the basis of a threshold of 10 km.

Statistical analysis

The values of the 18 standardized indicators were estimated for each of the 6,535 NFI plots and then averaged at national, provincial and regional levels, as well as for protected areas and for each of the 20 forest type categories outlined in the Italian NFI. The distribution of the 18 selected indicators is reported in Fig. 3.

Fig. 3
figure 3

Distribution of the 18 selected indicators

Full size image

The indicators were averaged for the five categories (composition, structure, regeneration, deadwood, hemeroby) and on the basis of the three B-N-OG indexes.

Multiple statistical analyses were conducted using the database. First, we estimated correlations for all pairs of the 18 indicators from Table 1 (Wei et al. 2017). Second, we conducted a principal component analysis (PCA), a multivariate technique that reduces the data’s dimensionality while preserving their covariances. When applied to our data (18 indicators for 6535 observations), PCA finds the eigenvalues and eigenvectors of the covariance matrix of each indicator. The purpose of the PCA was to better understand the relationships among the indicators.

Combining the correlation analysis and the PCA, we estimated the pairwise correlations between the indicators to assess their interrelationships (highlighting overlaps and gaps) and to ensure a rigorous quantitative approach for analyzing their redundancy when the indicators are aggregated to estimate the B-N-OG indexes. The pairwise correlations (r) among B-N-OG indexes were then analyzed using the Pearson correlation method. Analyzing the correlation between B-N-OG indexes serves the purpose of understanding underlying ecological processes and dynamics within forest ecosystems. For instance, positive correlations between B-N-OG may suggest shared ecological traits or processes that led the indexes to increase, while negative correlations may indicate trade-offs between indexes. The same approach was also used to analyze the relationships between the five different categories (composition, structure, regeneration, deadwood and hemeroby).

Finally, a one-way analysis of variance (ANOVA) was conducted to analyze the relationship between B-N-OG indexes and the classes of specific variables available in external datasets. Specifically, the one-way ANOVA was used to assess statistically significant differences among the means of B-N-OG indexes (i) between areas inside and outside the boundaries of protected areas (considering both EUAP and Natura 2000 networks) and (ii) between different forest type categories. To better understand the relationships between the three aggregated B-N-OG indexes and the forest type categories we performed a ranking analysis. All the analyses were performed using R software (Chambers 2008; Wei et al. 2017).

Results

Analysis of single-variable indicators

The pairwise correlations between the 18 indicators are presented in Fig. 4, while the results of the PCA are shown in Fig. 5. Further details are available in Appendix, section A, Table 3 and 4.

Fig. 4
figure 4

Correlation matrix among Italian NFI’s derived indicators. Statistically insignificant correlations (p < 0.01) are left blank

Full size image
Fig. 5
figure 5

Results of the PCA showing the first two components. The plots in the main graph are colored based on the 20 forest categories (same colors and codes of the map in Fig. 1). In the smaller graph the same plot is instead colored on the basis of the belonging to a protected area (green) or not (red). (Color figure online)

Full size image

For a large proportion of the cases, the relationships between the indicators were statistically meaningful even though the correlations were quite variable. Considered in terms of absolute values, the max_H, GSV and the L_trees had the largest average correlations, r = 0.27 and r = 0.23, respectively. The PCA confirmed the same results with GSV, v_H, SDI, max_DBH and max_H having the largest eigenvalues with the first two principal components together explaining only the 38% of the total variability.

On the opposite side, HEM (as the distance from artificial surfaces) was less correlated with the other indicators (average r = 0.06), with greatest (positive) correlation with the GSV. Small correlations were also found for the three regeneration indicators with average r ranging between 0.08 and 0.1. Once again, PCA confirmed these results with HEM, REG_1, REG_2 and REG_3 having the smallest eigenvalues.

The largest correlation was between GSV and max_H (r = 0.79) and max_DBH (r = 0.77).

In general, the average agreement between the seven indicators belonging to the “structure” category is large (r = 0.48) and always positive. The relationship between the three indicators belonging to the “composition” category is even greater in absolute values (0.74) because the relationships between E and the other two indicators, n_SP and S, were r = − 0.52 and r = − 0.85, respectively. A different behavior is shown by the deadwood category indicators with only small agreement among them with average absolute value of only r = 0.08. The main indicator of this category (DWD) is positively related to GSV (r = 0.35), and consequently also with the max_H (r = 0.31) and max_DBH (r = 0.24), as well as with L_trees (r = 0.22) and SDI (r = 0.19). The average correlation for the three regeneration category indicators is always positive (r = 0.32), although correlations with indicators external to this category were small. Correlations between GSV and the B-N-OG indexes were small, ranging from r = 0.12 to r = 0.16. Finally, HEM had only small correlations with all the other indicators with the greatest values for GSV (r = 0.10) and n_SP (r = − 0.125). The correlations and the statistical significances among indexes for the five categories is reported in Fig. 6. More details are available in Appendix, section A, Table 5 and 6.

Fig. 6
figure 6

Correlations and statistical significances among B-N-OG indexes components. Statistically insignificant correlations (p < 0.01) are left blank

Full size image

The results confirm a positive and consistent agreement of structure with deadwood and composition (r = 0.2 and r = 0.14, respectively), while the relationship between composition and deadwood is weaker (r = 0.1). Regeneration has weak negative relationships with structure and deadwood and no meaningful relationship with composition. Finally, hemeroby shows a weak positive relationship with structure, and negative relationship with the other categories of indicators.

Analysis of aggregated B-N-OG indexes

Following are the results of aggregating the indicators to estimate the indexes. Plot-level B estimates ranged between 0.031 and 0.357 with mean of 0.149 and standard deviation (sd) of 0.042; N estimates ranged between 0 and 0.435 with mean of 0.060 and sd of 0.048, while OG estimates ranged between 0 and 0.441 with mean of 0.088 and sd of 0.054) (Fig. 7). More details on B-N-OG estimates at plot-level are available in Appendix, section B.

Fig. 7
figure 7

Histogram distributions and correlations between the biodiversity, naturalness and old-growth status (B-N-OG) indexes

Full size image

The correlations between the three aggregated B-N-OG indicators are also reported in Fig. 7. The data demonstrated a stronger relationship between plot-level B and OG (r = 0.81), than between B and N (r = 0.15) or between N and OG (r = 0.25).

Mean estimates of the three aggregated B-N-OG indexes demonstrated relevant diversity across Italy. Greatest B and OG means were in the northern regions, i.e., Trentino Alto Adige, Veneto and Friuli Venezia Giulia (B 0.178, 0.169, 0.166 and OG 0.140, 0.110, 0.109, respectively) (Fig. 8). On the other hand, the southern regions, such as Calabria, Puglia and Sicilia, had the greatest N mean estimates (i.e., 0.089, 0.084 and 0.082, respectively).

Fig. 8
figure 8

Regional, NFI plot-level means for biodiversity, naturalness and old-growth status (B-N-OG) indexes

Full size image

Further detail of B-N-OG at national and regional levels is available in Appendix, section B.

Figure 9 shows the relationships between regional forest area—as reported in official NFI statistics for 2005 (INFC 2007a, b)—and the relative mean B-N-OG estimates.

Fig. 9
figure 9

B-N-OG mean estimates in Italian regions and forest area (INFC 2007a)

Full size image

As reported in Fig. 10, the greatest estimates of the mean for the three indexes were for the mountainous coniferous forest type categories (dominated by fir, spruce, stone pine and larch), while the smallest mean estimates were for broadleaved artificial plantations including poplar plantations, especially regarding N.

Fig. 10
figure 10

Biodiversity, naturalness and old-growth status (B-N-OG) mean estimates (with standard deviation bars) over plots in the Italian NFI plots among the 20 Italian NFI forest categories. The Italian NFI forest categories are sorted by the overall B-N-OG mean, in descending order

Full size image

The ANOVA suggested that the main effect of species groups, conifers or broadleaves, on B-N-OG means was statistically significant (p < 0.001). Further information on ANOVA results is reported in Appendix, section D.

After ranking the forest type categories on the basis of the mean estimates of the three indexes, we found a strong relationship not just between B and OG (r = 0.929) but also between N and B (r = 0.693) and N and OG (r = 0.665). The greatest rank order was always obtained for beech, fir, spruce and larch forests (Fig. 11).

Fig. 11
figure 11

B-N-OG pairwise ranking in forest categories

Full size image

Relationship with protected areas

More than one-third of the plots are located inside protected areas (2154 plots, 32.8% of the total). Specifically, the Natura 2000 network comprises 1924 plots, while EUAP comprises 1062 plots with 832 plots located in both EUAP and Natura 2000 protected areas. Considering that each Natura 2000 site could host more than one habitat, most NFI plots were located in habitats “Quercus ilex and Quercus rotundifolia forests” (9340) and "Apennine beech forests with Taxus and Ilex" (type 9210) with 925 and 830 plots, respectively (Biondi et al. 2010; Commissione Europea 2013). Further information and habitat descriptions are available in Appendix, section C.

Mean B estimates for inside and outside protected areas were not statistically significantly different (p = 0.483). On the other hand, the differences between mean N and OG inside and outside EUAP protected areas were statistically significant (p < 0.001 and p = 0.005, respectively). Hence, mean N and OG over plots inside protected areas were larger than outside protected areas (N 0.077 vs. 0.052, OG 0.091 vs. 0.087).

Figure 12 shows the distribution of B, N and OG inside (blue) and outside (red) protected areas.

Fig.12
figure 12

Biodiversity, naturalness and old-growth status (B-N-OG) distribution inside and outside protected area

Full size image

Among Natura 2000 habitat macroclasses (Biondi et al. 2010), temperate mountains coniferous forests registered the greatest B-N-OG estimates (0.170, 0.101, and 0.132, respectively), while deciduous Mediterranean forests and sclerophyllous Mediterranean forests registered the smallest estimates for all the three indexes (Fig. 13). Moreover, mean B estimates for inside and outside protected areas were not statistically different (p = 0.280). Conversely, the differences between mean N estimates and between mean OG estimates inside and outside protected areas were statistically significant (p < 0.001 and p = 0.006, respectively).

Fig. 13
figure 13

Biodiversity, naturalness and old-growth status (B-N-OG) mean estimates over plots, per ‘Natura 2000 habitats’ macroclasses. The classes are sorted by the overall B-N-OG mean per habitat, in descending order

Full size image

Discussion

Threats to forest habitats are likely to increase, mostly due to human-related activities such as climate change, urbanization and habitat loss and degradation (Seto et al. 2012; Aronson et al. 2014; Curtis et al. 2018; Kondratyeva et al. 2020). In this scenario, multiple international processes (Montréal Process 2009; Maes et al. 2018; Convention on Biological Diversity 2019; Forest Europe 2020; FISE 2021) require consistent tools to assess and report on B-N-OG. These requirements could be fulfilled by implementation of a sound, multipurpose, forest monitoring system based on data acquired from multiple sources (Chirici et al. 2011, 2012; Rondeux et al. 2012; Reise et al. 2019). In such a context, a relevant question is whether data routinely acquired by NFIs can contribute to such monitoring systems. For this study, we tested a set of indicators for assessing forest B-N-OG indexes using data acquired by the Italian 2005 NFI.

Single-variable indicators

When analyzed in terms of single-variable indicators, on the basis of both PCA and correlation analysis, our results confirm the relevance of the GSV variable but at a greater level than previous studies such as Storch et al. (2018). GSV is, in fact, strongly correlated with both the B index (r = 0.68) and the OG index (r = 0.88). This is easily explicable from an ecological perspective view, at least in forest conditions typical of Italy where the accumulation of GSV is typical of less disturbed forest ecosystems that can accumulate a greater level of biodiversity and can evolve to conditions that have greater OG. This is confirmed by the finding that forest type categories with greater estimates of the aggregated B-N-OG indexes (beech, spruce, silver fir, larch and stone pine forests) are those dominating mountain areas in Italy. These areas are less disturbed by forest logging and by fires, while deciduous broadleaved forests dominated by oaks (mainly Q. cerris and Q. pubescens) together with Mediterranean forests are more disturbed by forest loggings of the coppice system and by fires.

Further, deadwood volume showed a strong, positive correlation with the presence of large trees, GSV and structural indicators in general. While data on deadwood are extremely relevant for assessing B-N-OG indexes from an ecological perspective, the Italian dataset was strongly characterized by the dominance of forests with very small deadwood values (Fig. 3). For this reason, more advanced indicators based on the analysis of just one component of deadwood volume were only weakly related to the overall aggregated indexes. Because we cannot generalize these results, these indicators should be reconsidered when analyzing regions characterized by greater values of deadwood volume.

Results for regeneration are similar to the results for deadwood, because only very limited regeneration was found for the plots (Fig. 3). Nevertheless, these indicators showed a general negative correlation with structure indicators—especially GSV, max DBH and height, and the presence of large trees—suggesting that greater forest structures could hinder certain stages of regeneration (class III, above all).

As expected, forest structure (i.e., variability of tree height and Gini coefficient) and composition indicators (i.e., Evenness) showed an overall negative pattern of correlation, apart from the number of species which was positively correlated with SDI and variability of tree height.

As underlined in several other studies (Chirici et al. 2011; Galluzzi et al. 2019), the composition indicators estimated using NFI data consider only tree species, a very small component of the overall α-biodiversity. Therefore, the results of this study should be interpreted with caution.

Additionally, the hemeroby indicator (i.e., calculated on the basis of the distance from anthropic disturbance) displayed small correlations with other indicators. Nevertheless, the correlations are still meaningful; for example, GSV is greater when the distance from urban areas increases because such areas are probably more difficult to access and thus less easily exploitable for forest operations.

Importantly, when we analyzed the PCA results (Fig. 5), we found an interesting possibility for clustering into subpopulations among plots on the basis of their forest type categories. A comprehensive quantitative cluster analysis is beyond the scopes of this study, but we tried to characterize the subpopulations using three ellipses in Fig. 5 that we labeled on the basis of the European forest type categories from Barbati et al. (2014). Mountain forests are thus dominated by Alpine and Appennine forests with beech, spruce, silver fir and larch, Mediterranean forests are those with holm and cork oaks together with Mediterranean pines, and finally thermophilus forests are instead dominated by deciduous broadleaved (mainly downy and Turkey oaks).

Aggregated B-N-OG indexes

When compared in an aggregated way, the B and OG indexes demonstrated strong correlation, while naturalness was mostly uncorrelated with either of them. This is, of course, easily explained because of the selection of the input indicators. However, when geographical regions and forest type categories are ranked based on the three indexes, interesting differences emerged between B and OG, too. The indexes have consistent estimates when compared across the different types of management types. High forests have the greatest estimates followed by uneven-aged systems then by even-aged forests and finally by coppice forests. An example for the B index is reported in Fig. 14.

Fig. 14
figure 14

Example of distribution of the biodiversity index in the different sylviculture systems

Full size image

From a geographical perspective, our results showed that the northern part of Italy, particularly Trentino Alto Adige, Veneto and Friuli Venezia Giulia, exhibited the greatest B-OG estimates, most probably for three reasons: (i) mountainous topography, (ii) dominant forest type categories with greater B-OG estimates and (iii) large amounts of deadwood and GSV (Giannetti et al. 2022; Vangi et al. 2021). Smaller levels of anthropogenic disturbance (Riitano et al. 2016), compared to northern regions, are related to the largest mean N estimates obtained in southern regions such as Calabria, Puglia and Sicilia. Forest fragments with large N estimates were mainly found in protected areas or remote mountainous regions which experienced fewer human disturbances compared to forests nearby to population centers (e.g., Lombardi et al. 2012; Motta et al. 2006). Moreover, previous studies confirmed that habitat quality increases where human population become sparser (Newbold et al. 2015), such as from lowland to mountainous areas. The combined presence of protected areas in mountainous areas (Sallustio et al. 2015, 2017) and different forest management approaches used in the different forest type categories can indeed explain this trend in Italy.

It is interesting to note that despite artificial broadleaved plantations (including poplar) exhibiting the smallest B-N-OG estimates among forest type categories, artificial coniferous plantations displayed a different trend (Fig. 10). Indeed, several studies in Italy (De Meo et al. 2017b, a, 2022) demonstrated that artificial coniferous plantations could host large amounts of deadwood, compared to the average deadwood volume per hectare highlighted in the Italian NFI (Pignatti et al. 2009). Similarly, coniferous forests—particularly fir forests—exhibited some of the greatest mean estimates of B and OG. This finding aligns with previous research indicating that Alpine coniferous forests, including fir forests, have large volumes of deadwood compared to other European forest types (Barbati et al. 2014; Puletti et al. 2019). Here, the effects of windthrow events caused by climatic factors are compounded by limited harvesting in high-altitude forests or less accessible areas (Pignatti et al. 2009). Moreover, multiple studies conducted in Italy (Ciancio and Nocentini 2004; Travaglini et al. 2012; Parisi et al. 2016) highlighted the complexity of fir forests, emphasizing their potential to reach old-growth conditions, even when mixed with broadleaved species. Additionally, larch and stone pine forests recorded the largest N mean estimates over plots, closely followed by fir forests, perhaps given their capacity to reach the upper altitudinal limit of the forest (Didier 2001; Blasi C. and Biondi E. 2017). Notably, forests dominated by broadleaf species are predominantly managed using the coppice system (Fabbio 2016; Gasparini and Tabacchi 2011), such as deciduous oaks (Quercus spp., 33%), European hophornbeam (Ostrya carpinifolia Scop., 17%), beech (Fagus sylvatica L., 13%), sweet chestnut (Castanea sativa Miller, 16%), which are usually grown as pure stands, and the evergreen holly oak (Quercus ilex L., 10%) forests. This approach holds great significance in Italy, because coppices account for 35% of the overall forest area and are the main source of wood production (INFC 2007a). Thus, our results indicated that these forest type categories tend to have smaller B-N-OG estimates, most probably because of the strong negative impact of the coppice system and the easy accessibility.

These results can support future forest management strategies for coppice forests, used in Italy for fuel-wood and biomass production. Due to its consistent impact on B-N-OG, future application of the coppice system should be reconsidered. If these results are confirmed, the conversion of coppice to high forests should be preferred.

Consistently with other studies (Winter et al. 2014), no large differences in the index estimates can be found between forests within or outside protected areas. This can be explained by considering that forest management and conservation rules can be applied in very different ways within the protected areas. Once again, it is important to underline that protected areas, and more specifically large National Parks, are predominantly located in highland or mountainous regions and have a historical association with depopulation (Romano 1995; Romano et al. 2021). NFI plots within these forests, adjacent to protected areas, may exhibit specific structural traits that are directly influenced by the N and OG status of the protected areas (Barredo et al. 2021). Furthermore, the significant differences found for mean N and OG over plots in protected forest areas—compared to plots located outside protected areas—could be also related to the large amount of deadwood (Parisi et al. 2022). Moreover, creation of protected areas is relatively new in ecological timescales and thus consolidated results of their protection could not be visible yet. Finally previous studies have already underlined the strong variability of environmental conditions in national parks due to their phytogeographic, ecological and physiognomic heterogeneity (Capotorti et al. 2012).

Conclusions

The results of our test case demonstrate a useful assessment of forest biodiversity, naturalness and old-growth-status (B-N-OG) based on NFI data at multiple scales, from national to local levels. Moreover, the proposed set of indicators incorporated additional variables that are not related to traditional inventory purposes (Corona 2016), thus aiming toward multipurpose resource surveys. The set of the indicators we used demonstrated that PCA can identify groups of the main dominant types of forest ecosystems in Italy. We found this to be an encouraging result for future applications in larger pan-European monitoring framework.

The use of B-N-OG indexes presented here could enhance design-based assessment at the national level, enlarging the added value of NFI for supporting conservation strategies and sustainable forest management planning with trend analysis over time (Ćosović et al. 2020).

Our results also highlighted several considerations regarding the potential of NFI data in B-N-OG assessment at the European level. Firstly, NFIs should consider the future possibility of including field assessment of specific variables more directly related to biodiversity indicators: (i) the presence/absence census of all vegetation components, thus including herbs and lichens in order to be able to report a more complete assessment of α-diversity and (ii) tree-related microhabitats that can more accurately complement deadwood assessment for biodiversity monitoring purposes. Secondly, more resources for the harmonization/standardization of NFIs in Europe should be invested by the European Union and by the member countries; otherwise, differences in definitions and methods will hinder implementation of a complete pan-European forest monitoring system. Thirdly, because some basic historical forest variables such as the GSV that are traditionally designed for assessing the wood production of forests, were also important for monitoring B-N-OG, a more relevant role of remote sensing for such purposes can be hypothesized, because GSV can be quite easily mapped with remotely sensed data (Camia et al. 2023). In fact modern methods in combination with large datasets of multisource remotely sensed data can be used for accurately monitoring indicators (such as GSV) and forest disturbances trends in time.

Future recommendations can be envisaged for additional progress in the studies on forest biodiversity, naturalness and old-growth status, facilitating a more robust monitoring and comparison of results in Europe.

  • More investments in NFIs to improve the consistency and comparability of B-N-OG indicators across different countries.

  • More objective approaches for selecting the indicators. The aggregation of the indicators to create B-N-OG indexes is still very subjective. The method used in these cases is based on the opinion of experts and the review of previous studies, but the scientific background and the ideas of scientists related to such complex and debated disciplines may strongly and adversely affect the final decision on which indicators should be considered. For this study, we introduced the use of the well-known principal component analysis that can be used to support a more objective selection and aggregation of the indicators, eventually also on a weighted basis.

  • More time-trend analysis. As correctly demonstrated in Reise et al. (2019), future studies should concentrate more on the analysis of temporal trends in the indexes using the NFI data acquired at multiple dates. Unfortunately, at the date when this study was initiated, data for NFI2015 in Italy were not yet available (Burrascano et al., 2023).

  • More use of stratification. We think that all the indexes proposed should be applied with caution if not stratified based on potential conditions. In fact, areas affected by strong environmental limiting conditions (a typical example is for the island of Sardinia with extreme heat and drought) may result in small B-N-OG estimates. This can be easily implemented through stratification.