Using lidar to assess the development of structural diversity in forests undergoing passive rewilding in temperate Northern Europe

Study sites

Four localities where chosen for the examination and comparison of forest structure by lidar (Figs. 1 and 2). The sites are situated within a 900-km² area delimited by the UTM zone 32N coordinates: Northering: 6190000–6220000 and Eastering: 550000–580000. The main locality was Vorsø, an island in Horsens Fjord in eastern Jutland (Denmark). Vorsø was chosen because of the unique and well-documented vegetation history (Halberg & Gregersen, 2010). We divided Vorsø into seven areas, hereafter named zones, according to the history of the island. The zones are named with the letter V for Vorsø, followed by Y (young), M (mid-aged) or O (old) corresponding to the site history (vegetation age) of the zone. The numbers (1, 2,..) differentiates zones and areas of similar age. Areas influenced by humans, such as buildings, roads, and old gardens, were excluded from the seven zones of Vorsø.

A foundation bought and protected the island in 1929; leaving the three existing forest patches (V_O1, V_O2 and V_O3) and the main part of the farmland (V_M1 and V_M2) for passive rewilding. The passive rewilding process of the three forests were initiated by the destruction of the drainage and the eradication of bigger stands of non-indigenous tree species. In 1979 farming ceased in the remaining farmland (V_Y1 and V_Y2). No management, planting or seeding actions have been performed during the years of protection, although cormorants (Phalacrocorax carbo) was regulated before 1979 to keep numbers down. The main influencing factors since protection have been tree-killing excrements from the cormorant colonies (heavy impact on V_O2, and minor to medium impact on V_O1 and V_M1), windfall (mainly V_O1), roe deer (Capreolus capreolus), Dutch elm disease, along with heavy seed supply of sycamore maple to V_M2 from V_O2 and V_O3.

At the time of our survey the dominant tree species were oak (Quercus robur), maple (Acer pseudoplatanus), ash (Fraxinus excelsior), beech (Fagus sylvatica) and elm (Ulmus glabra, heavily reduced by Dutch elm disease), as well as alder (Alnus glutinosa) in wet areas. Frequent occurring shrub species were elderberry (Sambucus nigra), blackthorn (Prunus spinosa), raspberry (Rubus idaeus), single-seeded hawthorn (Crataegus monogyna), cherry plum (Prunus cerasifera), dog rose (Rosa canina) and willow species (Salix cinerea, Salix caprea).

Three managed forests nearby Vorsø representing typical management scenarios and hence relevant for comparison to the Vorsø passive rewilding case were pointed out:

• –

  Tranbjerg (hereafter Tran_Y3), a 4.51 ha young managed stand planted in 1983 on former farmland. The main tree species is oak (Quercus robur). Drainage is unchanged. There has been no fencing. Managers performed weeding after planting, and thinning of the stand has been carried out a couple of times.

• –

  Fløjstrup (hereafter Fløj_Y4), a 0.92 ha young managed stand established in 1983 by natural regeneration after a clear-cut. The main tree species is sycamore maple (Acer pseudoplatanus) mixed with beech (Fagus sylvatica) and ash trees (Fraxinus excelsior). There is only surface drainage and the stand was too wet for forest machinery to operate. The stand was fenced during regeneration and thinning has been carried out 3–4 times to control numbers of trees and remove some tree species.

• –

  Stensballe (hereafter Sten_M/O), a 20.42 ha semi-old stand of mainly broadleaved trees under commercial forest management. The main species is beech (Fagus sylvatica). Mature trees are selectively harvested and the upbringing of new trees is secured by natural regeneration, leading to a mixed stand of age classes.

Lidar-derived metrics

Lidar-data was acquired with an airborne Optech Altm Gemini Laser Altimeter system in March 2007 (kortforsyningen.dk, 2015) with a footprint of 40 cm. The point cloud density was one return per 2.2 m² on bare ground and included up to two returns per pulse with a vertical and horizontal accuracy of 0.67 and 0.06 m, respectively. We derived 25 metrics from the height-normalized point cloud after deleting overlapping returns. The 25 metrics were found in existing lidar literature and selected on a criteria of being used in comparison to manual ground surveys in previous studies of for being alike such metrics (Table S1). We performed calculations and processing in the ArcGIS (version 10.2.2.3552; Esri ArcMap, Redlands, California, USA) software and LasTOOLS (Version 141117; Rapidlasso, GmbH, Gilching, Germany). Basic grid cells measured 10 × 10 m ensuring sufficient returns per cell for statistical treatment.

We furthermore computed 25 metrics of horizontal heterogeneity based on the first 25 metrics, by calculating the standard deviation of the values for each cell (10 ×10 m) and the eight surrounding cells and ascribing this SD-value to each cell. Thereafter the cells were aggregated into bigger cells of 30 × 30 m, by taking the mean of the SD-values of nine adjacent cells to reduce pseudo-replication by adjacent cells influencing each other in the SD calculation.

Statistical analysis

We extracted the values from each 10 × 10 m cell of the 25 metrics for vertical structure and each 30 × 30 m cell of the 25 metrics for horizontal heterogeneity and exported it to the R (R i386 3.1.2Ink; R Core Team, 2014) statistical software. We scaled and centered the values and completed one PCA (hereafter PCA_10m) based on values from the 10 × 10-m grid cells (n = 7,319), and another one (hereafter PCA_30m) based on the 30 × 30-m grid cells (n = 1,002).

Based on a threshold of about 90% explained variance, the first five principal axes were tested for significant differences in medians and variances among the three managed forests and seven rewilded zones of Vorsø focusing on the even-aged stands (Table 1), using Kruskal–Wallis as a non-parametric test for medians and Levene’s test for variance. In addition, we included extreme PCA axes values as a measure of unique vegetation structure in the analyses.

Rewilding of abandoned farmland

The youngest zones on Vorsø and the youngest managed forests (Tran_Y3 and Fløj_Y4) have developed completely differently, which we conclude must be due to the different afforestation techniques used, as well as the related management (fencing and weeding in Tran_Y3 and Fløj_Y4). The two latter are evolving homogeneously in line with previous investigations of intensively managed forests (Fuller, Foster & McLachlan, 1998; Nilsson, Hedin & Niklasson, 2001). The larger median values for the managed stands on PCA_10m axis 1 clearly indicate that the dense establishment led to a higher growth rate, even compared to V_M1, which has had 50 more years to grow tall vegetation. A recent study modelling post-disturbance spontaneous regeneration in the Bavarian Forest National Park, Germany, showed uniform development of mean tree height across five sites over 80 years, despite differences in their starting points (Hill et al., 2017). In contrast, the stand heights investigated in this paper develop differently. Since soil and climatic conditions are similar within the study area, we assign those structural differences to the method of establishment.

The relatively similar vegetation development in the two young Vorsø zones (V_Y1 and V_Y2) indicates predictability of underlying controlling ecological factors, also found in other studies (Bullock, Moy & Pywell, 2002; Prach & Pyšek, 2001), despite the complex interactions between the factors influencing development. The similar development suggests that spontaneous regeneration might be predictable, when conditions are well known as assumed by ongoing vegetation development modelling, for example, (Hill et al., 2017).

The two fields abandoned in 1929 (V_M1 and V_M2) also predominantly shared the same conditions. However, they developed quite distinct vegetation structure, identified by significantly different median and variance values in nine of 10 cases (PCA_10m) (Table 2). Different seed input is concluded as being the main reason for this divergence (Halberg & Gregersen, 2010). The prevailing western wind provided an effective vector for seeds of the fast-growing Acer pseudoplatanus trees from V_O2 and V_O3 to V_M2, causing the divergence between V_M2 and V_M2, which underlines the broadly acknowledged importance of seed banks and seed supplies (Benayas, Bullock & Newton, 2008; De Steven, 1991; Olsson, 1987). The ongoing diverging development of vegetation through 78 years in the two abandoned field sites was hence largely predictable when general successional ecology was considered and are in line with a recent study concluding that old-growth spatial pattern is determined by the initial patterns of regeneration (Hill et al., 2017). Thus, the combination of the recorded site history and the statistical result presented in the two above pairwise comparisons, despite different outcomes, underpin that designing spontaneous regeneration projects with specific aims for the vegetation structural outcome is possible when considering ecological theory and site-specific conditions carefully. For instance, a heavy seed supply into an abandoned farmland site might lead to structural development similar to managed stands (Sten_M/O in this case), at least for the first 78 years. We believe that site managers should make use of these basic rules in forest ecology when designing conservation projects.

In general, passive regeneration on Vorsø is progressing at a slow pace compared to other studies (Cramer, Hobbs & Standish, 2008; Verburg & Overmars, 2009). After 28 and 78 years, respectively, three spontaneous regenerated zones of Vorsø still hold shrub-dominated low vegetation and open areas. The timespan of 20–60 years for the development of stands of mid-successional tree species on abandoned farm land, concluded in previous studies (Cramer, Hobbs & Standish, 2008; Verburg & Overmars, 2009), fail to comply with most spontaneous regenerated areas on Vorsø, which are too dominated by gaps and low and sparse vegetation to fulfill this criteria. Successional arresting factors such as (high water level, drought and competition from herbs and grasses) are all present on Vorsø; however, as former farmland it should be categorized as having moderate conditions for colonization of woody vegetation, for example, compared to abandoned mining sites (Macdonald et al., 2015) or other areas exposed to heavy disturbances. In this light, we suggest a prolongation of the recognized timespan of natural afforestation under moderate conditions, such that the theory can comply with moderate conditions with a low degree of seed input from trees.

For biodiversity, the slow development might be beneficial, especially if early successional stages are underrepresented in a landscape level (Angelstam, 1998; Brawn, Robinson & Thompson, 2001). This is also supported by a study reviewing early-successional forest ecosystems, arguing that the prevailing focus on recovery of the closed-canopy stage should be shifted to greater focus on the qualities of the early-successional stages (Swanson et al., 2011). Besides, naturally developed gaps are found more valuable to biodiversity than artificial gaps created by thinning operations (Seidel, Ehbrecht & Puettmann, 2016). Thus, the spontaneous regeneration on Vorsø is generally in line with desired vegetation development according to international research, and of presumable interest to other afforestation projects in abandoned farmland sites in the temperate zone.

Rewilding of existing forests

This study evaluates rewilding of existing forests by comparing sites of unmanaged forest stands, which has previously been managed, to one stand (Sten_M/O) still under silvicultural management in terms of tree harvest, drainage and pest control. The results indicate a relatively high degree of vegetation structure heterogeneity in the V_O1 rewilded Vorsø forest. The stand is dominated by mature oak and ash trees, and shows the lowest median value on the axis 3 of PCA_10m (Fig. 3C), meaning that it has the most open canopy layer among the older stands, following disturbances by cormorant colonies, windfall and Dutch elm disease (Halberg & Gregersen, 2010). The reason for greater windfall impact here, for example, compared to the nearby managed Sten_M/O forest, is the increased water level, caused by drainage destruction, which prevents the roots from growing deep into the ground and causing existing roots to rot (Halberg & Gregersen, 2010). When we combine our results to the site history, our findings are in line with a Canadian study reporting that stands which have undergone canopy disturbance can evolve a high degree of spatial pattering (Zenner, 2005). According to international research, the impact of disturbances to biodiversity through unique forest structure is well documented (Berg et al., 1994; Navarro & Pereira, 2012; Nilsson, Hedin & Niklasson, 2001). We argue that the slow, moderate and persistent dynamics that have occurred in this rewilded stand (V_O1) have produced such vegetative attributes crucial to biodiversity (Fig. S2), and interpret it as being in line with the mediate disturbance hypothesis.

In contrast, V_O2 was subject to heavy and devastating disturbance from a packed colony of great cormorants in the 1980s, which entirely terminated the vegetation of tall-stemmed trees (Halberg & Gregersen, 2010). Instead of creating a unique structural composition, it rather pushed back the succession to the shrub or early stand stage (Fig. 2E). Our analyses cannot detect increased heterogeneity in this rewilded forest compared to a managed forest, and are therefore contrary to the findings in a review of post-disturbance vegetation structure, which state that regenerated vegetation is more heterogeneous than typical closed-canopy forest (Swanson et al., 2011). The divergence to our findings could be due to more homogenous stand replacement following the heavy impact of cormorant colonies compared to the regenerated vegetation following, for example, wildfires or windstorms.

Evaluation of the presented method

The capability of distinguishing forest areas by this method seems to be very sensitive, as it finds the overall similar vegetation development of the two young spontaneously regenerated zones (V_Y1 and V_Y2) to be different on six out of 10 parameters. The test of PCA_30m axes does not find any significant differences between these two former fields. Previous research shows that a scale of minimum 500 m² is necessary to adequately detect defining structural attributes of different forest categories (Zenner, 2005). This is in line with our findings, where the 900-m² grid cells confirm the similarity of the young Vorsø stands better than the 100-m² grid cells, although the methods are different.

The few significant tests of V_O3 (0.38 ha; four cells in PCA_30m) and Fløj_Y4 (0.92 ha; 20 cells in PCA_30m) reveal that we reached the lower threshold of cell numbers. The 36 cells of V_O3 in PCA_10m apparently satisfy the statistical analysis to achieve trustworthy results. Increased point density of point clouds allows researchers to reduce cell size to one-m² or even less (Müller & Brandl, 2009), providing 100 cells per 100 m² instead of only one, as in this paper, though still in consideration of the above discussed scale issue. However, the high degree of significant results in the pairwise tests in combination with corresponding point densities in the literature, convince us of the sufficient point density in this study (in general between 0.5 and 0.9 per m²). A study from Colorado, USA, (Hall et al., 2005) calculated useful estimates of stand height, total above-ground biomass, foliage biomass and basal area from an average point density of 1.23 returns per m². A study from the UK (Boyd & Hill, 2007) analyzed lidar intensity values captured over an area of woodland based on an average point cloud density of one return per 4.83 m², and an Italian study (Mura et al., 2015) estimated indices of structural diversity by using an average density of 1.5 pulses per m².

The PCA assessment ensures that the information in the point cloud retrieved by the calculated metrics is analyzed by non-correlated principal components, hereby avoiding that the metrics contribute with correlated information. Assumable, single lidar-derived metrics could explain the main part of the information instead of PCA axes. However, the wide number of metrics contributing to the eigenvectors in this study (Tables S5 and S6) also clarifies that the PCA axes express a complex interaction of metrics, which could be particular useful when dealing with low-density point clouds. A previous investigation succeeded in using PCA for choosing the three most significant metrics among nine metrics for modelling above-ground biomass across three forest study sites (Li, Andersen & McGaughey, 2008). We reason that the PCA axes are better suited for pairwise comparisons of forest sites than specific metrics, and that it is a different task to model concrete forest measurements, e.g., above-ground biomass, compared to designating key structural attributes suitable for differentiation of vegetation structure between stands. This is supported by the fact that two stands holding similar biomass can have very different structural composition. More research is needed to explore which methods are best suited for evaluating vegetation structural quality of forest sites using wall-to-wall lidar data.

As a disadvantage of the suggested simple method, we should mention the difficulties of ecological meaningful axes interpretation in some cases. In addition, it is a drawback that the PCA axes values cannot be directly compared to equivalent studies of other sites. The creation of a multi-site likewise analysis could form the foundation of a common reference dataset of PCA axes values, at least for delimited homogenous areas, for example, the northern temperate broadleaved tree species dominated zone. However, this is beyond the scope of this study.