Even-aged forests continue to dominate across Europe, accounting for approximately 75% of forested areas (ForestEurope, 2020). However, due to the greater resilience and resistance of uneven-aged forests to disturbances and climate change (Lafond et al., 2014; O’Hara and Ramage, 2013) their proportion is expected to increase both in Europe and globally. Thus, interest in the establishment and management of uneven-aged forests has already grown in recent years (Dãnescu et al., 2016). Structural heterogeneity of the forest is primarily the result of frequent silvicultural interventions over space and time, or from small-scale natural disturbances (Kukunda et al., 2019). Forest stand structure encompasses vertical elements (e.g., number of canopy layers), horizontal elements (e.g., spatial distribution of trees, gaps), and species (Maltamo et al., 2005).

Clear-cutting is legally prohibited in Slovenia under national forest management legislation (ZG, 1993). Forest management in the Republic of Slovenia is guided by the principles of sustainability, close-to-nature silviculture, multifunctionality, and forest management planning in order to ensure the long-term conservation of forests and their ecosystem services (ReNGP, 2007). The structural heterogeneity of Slovenian forests is primarily the result of small-scale forest management practices (ReNGP, 2007). Frequent, small-scale interventions in forests lead to the development of structurally heterogeneous stand structures or patches as well as mosaic-like stand structures with varying degrees of structural heterogeneity at the fine scale. Uneven-aged forests prevail in Slovenia (Bončina et al., 2002), with the most representative forest types occurring in mixed spruce-fir-beech forests on carbonate bedrock in the Dinaric and the Alpine eco-regions and mixed spruce-fir-beech forests on silicate bedrock in the Pohorje eco-region (Klopčič and Bončina, 2012).

At the national level, field data were collected in Slovenia until 2018 as part of the Monitoring of Forests and Forest Ecosystems (MFFE) program (Skudnik et al., 2021a; Skudnik et al., 2021b) which was implemented on permanent sample plots in a systematic 4 × 4 km grid. The plots were surveyed in 2000, 2007, 2012, 2018, and 2024. In 2020, the MFFE inventory was upgraded and converted into the National Forest Inventory (NFI). The NFI is based on an unaligned 2 × 2 km grid, with measurements being repeated on the same plots every 5 years. The NFI is designed as a panel inventory system (Skudnik et al., 2023), where the entire grid is divided into five panels and one panel is measured in each year. The former MFFE plots have been incorporated as the fifth panel in the NFI. In this way, the NFI collects data on the condition and dynamics of Slovenian forests annually, based on a representative sample with an effective spatial density of approximately 4 × 4 km. In total, approximately 750 plots are surveyed each year, and the complete NFI covers approximately 3,750 plots across the entire country (Pintar et al., 2024).

In the Slovenian NFI plots, the vertical forest stand structure is currently estimated based on expert judgment (Skudnik et al., 2022a). Although field crews undergo calibration of estimation procedures before each annual inventory, inter-observer variability can still lead to discrepancies. Therefore, the aim of this study is to investigate the potential of estimating vertical forest structure using freely available airborne low-resolution laser scanning data, which could subsequently be verified and complemented during field surveys. Within the NFI protocol (Skudnik et al., 2022a), tree heights are only measured for a subsample of trees in each plot, which means that even the available field measurements do not capture the entire vertical structure of the stand. In addition, remote sensing approaches—particularly those based on laser scanning—enable more accurate and efficient estimates of tree heights under optimal conditions compared to traditional field-based methods such as optical triangulation (Ganz et al., 2019).

The main objective of this study is to demonstrate the feasibility of distinguishing between uneven- and even-aged forest stand structures on NFI plots solely based on freely available, national airborne low-resolution laser scanning data, without the use of field-based estimates or measurements.

In the present study, we analyzed the detection of uneven-aged and even-aged forest stand structures on NFI permanent sample plots using freely available national low spatial resolution (5 points per m²) airborne laser scanning data. Understanding the spatial distribution of uneven-aged forests is particularly important as these forests are more resilient and resistant to disturbance and climate change. A comprehensive understanding of both vertical and diameter heterogeneity in uneven-aged stands is crucial for improving their adaptive capacity. While diameter heterogeneity has been intensively investigated through field-based measurements of tree diameters (Kotar, 2011; Pintar and Hladnik, 2018), this study contributes by providing a field-independent assessment of vertical structural heterogeneity based solely on ALS data. Furthermore, identifying the location and vertical structural indicators of even-aged stands is equally important as it supports informed decision making for future forest management plans to improve forest resilience and resistance under changing climatic conditions.

The use of tree height metrics and indicators to describe tree canopy cover and vertical heterogeneity derived from ALS data has already been demonstrated as effective for classifying stand structures or forest stands, both at the level of whole stands and individual sample plots (Pascual et al., 2008; Torresan et al., 2016; Pintar and Skudnik, 2024a; Pintar and Skudnik, 2025). In our study, both variants of the K- means clustering approach (CHM and voxels), resulted in an optimal number of six clusters, as determined using the elbow method and visual assessment of stand structure patterns within each cluster. Similar results were reported in studies of structurally heterogeneous forests in central Spain (Pascual et al., 2008), northern Italy (Torresan et al., 2016) and Slovenia (Pahernik Estate, Pohorje) (Pintar and Skudnik, 2024a), where the optimal number of clusters derived from ALS data using the K-means method was five or six, which is consistent with the results of the present analysis.

We used CHM and voxel-based metrics derived from ALS data to estimate the vertical heterogeneity of forest stand structures. The CHM method was selected due to its simplicity and ease of implementation (Kim et al., 2016; Fassnacht et al., 2024), while the voxel method was used due to its great potential to improve the accuracy of forest structure description and stand characterization analysis (Pearse et al., 2019). Both methods provided equivalent results for stand structure classification, although the differences in vertical diversity between the uneven-aged and even-aged stands were more pronounced when CHM was used. Therefore, in areas with low-density ALS data, CHM analysis appears to be a more appropriate approach for assessing vertical heterogeneity. These findings are consistent with previous research from the Pohorje region, where CHD_CHM values for mature uneven-aged stands, calculated at the level of whole stands across a 570 ha forest estate, were estimated at 1.83 and 1.86 (Pintar and Skudnik, 2024a), closely matching the value obtained in this study (1.71). However, the usefulness of voxels is confirmed by the fact that the voxel method classifies 21.5% of the 26.0% of NFI plots (Table 3) as the stand structures with the highest vertical heterogeneity (uneven-aged). In this study, the terms “uneven-aged” and “even-aged” refer to the degree of vertical canopy differentiation, validated with field measurement data. To also validate the classification results, at least 20 individual stand structures from each cluster in both analyses were also visually assessed on the NFI plots using the ALS-derived canopy height model (CHM), vertical cross-sections of the point cloud, and orthophotos. This visual inspection confirmed that the structural characteristics inferred from the canopy height diversity metrics corresponded well with the actual stand conditions observed in the field.

For higher density ALS datasets, the use of voxel-based approaches (Pearse et al., 2019) are recommended, as they allow the detection of shade-tolerant understory trees. The resilience of stands to natural disturbances is mainly influenced by the vertical heterogeneity of trees in the canopy (Senf et al., 2020), which is effectively recognized by the CHD_CHM based classification method (Pintar and Skudnik, 2024a). We can state with high confidence that voxel-based approaches will be more appropriate in future studies using high-density ALS data. These methods are particularly effective for analyzing vertical structural diversity in characteristically multi-layered forest types, such as lowland floodplain oak–hornbeam forests (Hladnik and Pintar, 2017) and selection fir–beech forests (Bončina et al., 2002). The voxel-based analyses and the corresponding results presented in this study provide a valuable basis for further research into the vertical structure of forests using freely available national ALS datasets, which are expected to be recorded at higher point densities in the future.

In the Alpine and Pohorje ecological regions, ALS data were acquired in 2014 and 2015, while the field measurements on permanent sample plots were conducted in 2018. Despite the slightly longer time interval between the two data sets, the field measurements are still suitable for validating stand structure classification based on ALS data, as this type of comparison has already been applied at the national level (Skudnik et al., 2022b). This is also supported by the fact that harvesting occurred on 29% of all plots in the Alpine and Pohorje ecological regions between 2012 and 2018—a period that is 2–3 years longer than the period between the ALS survey (2014–2015) and the National Forest Inventory (2018). On the vast majority of plots where harvesting occurred, only individual trees were removed, which does not significantly affect the comparability of the data sets. Only 2% of the plots were more than two thirds of the trees harvested. The validation is also not significantly affected by tree height growth, as spruce trees at approximately 900 m a.s.l. in this area exhibit an annual height increment of 28–40 cm (Kotar, 2011).

The validation of stand classification using field measurement data was successful, as statistically significant differences in stand variables (derived from field measurements) were found between stand structure types classified from ALS data based on vertical heterogeneity and dominant height. The variables used in the analyses are also used in other national forest inventories in Europe. Therefore, the proposed methods can also be applied on a larger scale, which is particularly important given the increasing global need for a “living forest inventory” (Coops et al., 2022) that can capture forest dynamics in near real time. Europe is still dominated by even-aged forests (ForestEurope, 2020). In times of climate change, the aim is to improve the resilience and resistance of these forests, which can be achieved by increasing forest heterogeneity (O’Hara and Ramage, 2013; Senf et al., 2020). Therefore, an increase in the proportion of uneven-aged forests is expected in Europe and beyond (Dãnescu et al., 2016; Dãnescu et al., 2017).

All methods and analyses presented in this study are transferable and can be applied in future studies across other Slovenian forests as well as well as in forests outside the Slovenian borders. They are applicable in both uneven-aged and even-aged forests. In the latter, it will be easier to monitor a possible increase in vertical heterogeneity, which is particularly important in times of climate change, as uneven-aged forests are much more resilient. The applicability of the proposed methods largely depends on the availability of ALS data, which has become increasingly available across European countries in recent years (Pintar and Skudnik, 2025; Pintar and Skudnik, 2024b; Pintar, 2025). When applying these methods in other regions, including other Slovenian forests or international contexts, it is important to check the key parameters (the radius used for treetop identification, the height class interval for calculating canopy closure, and the CHD index) and adjust them if necessary to take local forest stand conditions into account. The parameter values used in this study (e.g., the 3 m window radius for local maxima detection, 5 m height class width, and 5 m canopy-closure threshold) were selected to match the prevailing crown architecture of the study area, where trees typically have tall, narrow crowns with moderate vertical stratification. These values should not be considered universal. In forests where crowns are wider, more irregular, or strongly overlapping—such as many broad-leaved or tropical forest types—optimal parameterization may differ to accurately capture individual treetop positions and canopy layering. The classification framework remains transferable, but parameter values should be adjusted to canopy architecture to ensure that canopy height diversity reflects structural differences rather than species-specific crown form. For example, in predominantly coniferous stands, where tree crowns are narrower and more steeply tapered, a smaller search radius may be suitable for detecting treetops, while in stands with broadleaves, a larger radius may be needed to accurately identify local maxima.

Although the classification approach based on canopy height diversity successfully differentiated vertically heterogeneous from more homogeneous stand structures, some limitations should be acknowledged. The CHD metrics used in this study describe the vertical distribution of canopy elements; therefore, their interpretability depends on the extent to which different canopy layers are detectable in the ALS point cloud. In stands where stratification is strong and continuous, canopy height diversity values clearly reflect these differences. In contrast, in stands where stratification is less pronounced, spatially patchy, or transitions between layers are gradual, structural gradients may be more subtle, and the resulting classification boundaries less distinct. For this reason, the classification results should be interpreted as a continuum of structural conditions rather than discrete stand types. Future applications may benefit from integrating canopy height diversity with additional descriptors of horizontal tree arrangement or stand density, which could further refine the detection of intermediate structural classes. Furthermore, separating stands prior to classification based on their species composition (e.g., predominantly coniferous, predominantly broadleaved, and mixed stands with varying degrees of mixture) could improve parameter selection, such as the search radius used for local maxima detection. This could be efficiently achieved using high-resolution satellite or airborne multispectral imagery to pre-classify plots before structure classification. The results presented here apply specifically to mixed spruce–fir–beech forests in the Alpine and Pohorje ecological regions. The canopy height diversity metrics quantify vertical stand structure and are therefore not directly affected by factors such as elevation or parent rock; however, these site conditions may influence long-term stand development and should be considered when applying the classification framework in different environmental conditions.