Assessing Wildlife Impact on Forest Regeneration Through Drone-Based Thermal Imaging

Abstract

Assessing the extent and magnitude of wildlife impact on forest regeneration (e.g., % browsed seedlings or reduction in regeneration density) remains a central challenge. This study explores the potential of unmanned aircraft systems (UAS) to quantify wildlife impact through the integration of drone-based thermal surveys and vegetation assessments. Specifically, it evaluates whether UAS-derived wildlife density estimates can be linked to browsing intensity and regeneration structure, thereby enabling an indirect assessment of silviculturally relevant forest dynamics. By combining remotely sensed wildlife data with field-based vegetation inventories, the study aims to identify measurable relationships between structural forest characteristics and browsing effects. This approach contributes to the development of spatially efficient, objective, and reproducible monitoring methods at the forest–wildlife interface. Ultimately, the study provides a novel framework for integrating modern remote sensing technologies into wildlife–ecological monitoring and for improving adaptive, evidence-based management in forest ecosystems increasingly affected by high ungulate densities and climate-related stressors. Two silviculturally contrasting study areas were selected: a broadleaf-dominated mixed forest in Hesse, where high ungulate densities were expected, and a pine-dominated site in Brandenburg, anticipated to experience lower browsing pressure. Thermal surveys were conducted using a DJI Matrice 30T drone equipped with a high-resolution infrared camera to detect and geolocate wildlife. In parallel, browsing impact was assessed using a modified circular transect method (“Neuzeller method”). Regeneration was recorded by tree species, height class, and browsing intensity. Statistical analyses and GIS-based spatial visualizations were used to examine the relationship between estimated ungulate densities and browsing levels. Results revealed clear differences in wildlife abundance and browsing intensity between the two sites. In the Heppenheim forest, roe deer densities exceeded 40 individuals per 100 ha, correlating with high browsing pressure—particularly on ecologically and silviculturally valuable species such as sycamore maple and sessile oak. In contrast, the Rochauer Heide exhibited lower densities and a comparatively moderate browsing impact, although certain tree species still showed signs of selective pressure. This study demonstrates that drone-based wildlife monitoring offers an innovative, non-invasive means to indirectly evaluate forest structural conditions in regeneration layers. The findings highlight the relevance of UAV-supported methods for evidence-based wildlife management and the adaptive planning of silvicultural measures. The method enhances transparency and spatial resolution in forest–wildlife management and supports evidence-based decision-making in times of ecological and climatic change.

1. Introduction

Forests represent complex, dynamic ecosystems characterized by interactions between numerous biotic and abiotic factors. In Central Europe, ungulates exert a significant influence on vegetation structure and the regenerative capacity of forest stands. As an integral component of near-natural forest ecosystems, wild ungulates actively shape their habitat through species-specific behavior [1]. This influence can promote certain plant communities, but may also lead to considerable silvicultural conflicts when browsing pressure exceeds ecologically sustainable levels. The resulting impacts—particularly browsing, bark stripping, fraying, and trampling damage—are collectively referred to as “wildlife damage” and may have both ecological and economic consequences [2].

Particularly in managed forests, the impact of wild ungulates is gaining increasing relevance, as wildlife must increasingly be regarded as a site-altering factor [1]. Depending on population density, demographic structure, and food availability, the effects can be substantial, with reductions in seedling density of up to 45% and height growth losses exceeding 30 cm in heavily browsed areas [3]. Especially problematic is the selective browsing of certain tree species that are often of silvicultural interest, such as silver fir (Abies alba), sessile oak (Quercus petraea), and sycamore maple (Acer pseudoplatanus), whose development can be significantly delayed or entirely suppressed by repeated browsing. This not only alters the species composition and structure of the regeneration layer but also leads to a loss of biodiversity and adaptive capacity in the face of climatic changes [2,4]. Within the framework of functional forest ecology, ungulate browsing plays a central role in shaping regeneration patterns, structural diversity, and, consequently, the ecological resilience of forests. Systematic reviews have shown that high ungulate densities are frequently associated with increased browsing pressure, which reduces the establishment of palatable tree species and thereby shifts stand composition in favor of less preferred species [5]. These alterations not only lead to reduced regeneration density but also affect the growth performance and competitive status of juvenile trees, ultimately resulting in long-term changes to stand structure [3]. As a consequence, forest structures may become more homogeneous, resistance to disturbance may decrease, and the capacity of the ecosystem to respond resiliently to external stressors such as climatic extremes or insect outbreaks may be diminished. This highlights that the monitoring of ungulate densities must not be considered in isolation but rather within the causal chain “ungulate density → browsing intensity → regeneration success → structural diversity and resilience” [6]. The explicit integration of these relationships enables a more comprehensive analysis of the impacts of wildlife densities on forest regeneration and provides a scientific foundation for deriving effective forest management strategies, including adaptive hunting regimes and reforestation concepts. In Germany, nine ungulate species currently occur permanently in the wild, with roe deer (Capreolus capreolus) and red deer (Cervus elaphus) being the most widespread and influential species [4,7]. The damage they cause is well documented in terms of both intensity and long-term effects on forest development, and it varies considerably depending on ungulate density, habitat structure, tree species composition, and silvicultural objectives [8]. In the context of progressing climate change and the need for climate-adapted forest transformation, wildlife impact represents a critical regulatory factor for the composition of future forest stands. A deeper understanding of the interactions between wildlife and vegetation is therefore essential in order to develop integrative management strategies that reconcile the objectives of nature conservation with those of sustainable timber production. The present study addresses the significance of wildlife impact on selected tree species, considering their silvicultural characteristics, climatic suitability, and ecological relevance. Various methods have been developed to assess browsing pressure and the density of wild animal populations [9]. However, the actual population size is difficult to determine in practice [10]. In recent years, unmanned aerial systems (UAS) have established themselves as a precise and cost-effective technology for estimating wildlife populations [11,12]. While they are traditionally used for direct animal detection, this study investigates the novel potential of drones for indirectly assessing forest conditions relevant to forestry, in particular the regeneration status, based on spatial wildlife distribution patterns. Estimating wildlife densities in forest ecosystems has long presented a methodological challenge, as conventional approaches such as pellet counts, line transect surveys, or camera trapping each exhibit specific limitations. Pellet counts provide only indirect evidence of ungulate presence and relative activity but are strongly influenced by decomposition rates, weather conditions, and the temporal use of habitat by animals. Visual transect surveys are prone to observer bias and are constrained in densely forested habitats due to limited visibility. Camera traps, in turn, allow for continuous data collection but capture only point-based sections of the habitat and do not enable a spatially comprehensive coverage of larger study areas [13]. Drone-based thermal imaging systems, i.e., Uncrewed Aerial Systems (UAS), represent a methodological advancement as they allow for active, large-scale, and largely non-intrusive detection of wildlife. By employing thermal infrared (TIR) sensors, animals can be reliably detected based on their temperature contrast with the surrounding environment. Several studies have demonstrated that, with appropriate flight altitude and sensor resolution, even large ungulates can be identified within structurally complex forest environments [14]. The advantages of this technology can be summarized in three key aspects: (1) Real-time or near real-time detection: Image and video data are recorded immediately during the flight and, when combined with automated detection algorithms, can be rapidly processed into estimates of population density or spatial distribution. (2) Low disturbance levels: The use of UAS with appropriate flight protocols (e.g., sufficient altitude, low noise emission) minimizes the influence on animal behavior in their natural habitat. In contrast to ground-based methods, where human presence or trapping devices may alter animal activity, UAS-based systems allow for more realistic observations of spatial and temporal habitat use [15]. (3) High spatial coverage and reproducibility: Standardized flight routes enable large areas to be surveyed in a comparable and repeatable manner, which represents a significant methodological improvement for population and habitat monitoring. Accuracy assessments have shown that detection rates of TIR drone surveys for large ungulates in forested environments are already high, although reliable species identification is not always feasible [16]. Nevertheless, certain technological and ecological constraints must be considered when applying UAS-based thermal imaging. Dense vegetation layers can reduce the visibility of thermal signatures, while sensor resolution and flight altitude may limit species identification. Moreover, the effectiveness of detection is strongly influenced by environmental conditions, such as temperature differences between the animal body and its surroundings, as well as the thermal properties of the ground surface [17].

This study was conducted in collaboration with OGF GmbH, a forestry consulting company based in Kesselsdorf, Saxony. The company provided drone flight data and results from commissioned wildlife observation studies in two different forest areas in Germany. One site in Brandenburg has nutrient-poor pine stands and low ungulate density was expected. The second site in Hesse comprises high-yield deciduous stands where high game densities were expected. Aim of this study is to investigate whether the ungulate density determined by drone-based thermal surveys correlates with the intensity of browsing impacts on forest regeneration. Specifically, the study addresses the following research hypotheses: 1. Browsing intensity increases with decreasing regeneration height due to higher accessibility for ungulates. 2. High ungulate densities reduce species diversity in forest regeneration through selective browsing. 3. A higher number of evenly distributed sample plots improves the precision of browsing intensity estimates. 4. The ungulate density determined by drone surveying correlates with the observed browsing effects.

The empirical validation of the relationship between ungulate densities estimated by thermal UAS and field-measured browsing intensity remains limited, particularly in mixed temperate forests.

2. Materials and Methods

2.1. Study Areas

To capture different intensities of wildlife impact on forest regeneration, two contrasting forest areas were selected, representing distinct gradients in elevation, species composition, climatic conditions, and expected ungulate density. The first study area is a 1361 ha municipal forest in southern Hesse near the Baden-Württemberg border (Figure 1). It extends from contiguous stands into adjacent villages on the western slope of the Vorderer Odenwald (95–531 m a.s.l.). Mean annual temperature is 10.7 °C and precipitation ~860 mm. The terrain is steep with diverse exposures and complex geology (diorite, granodiorite, schist, granite). Soils are predominantly brown earths and pseudogleyic brown earths, often derived from decalcified loess (Hessen-Forst 2013). Favorable growth conditions result from silt- and loam-rich substrates, high pore volume, and fresh to very fresh water balance. European beech (Fagus sylvatica) dominates, with sessile oak (Quercus petraea), sycamore maple (Acer pseudoplatanus), and other broadleaves; conifers are rare, with scattered silver fir (Abies alba) and Douglas fir (Pseudotsuga menziesii). Norway spruce (Picea abies) has largely disappeared since 2018 due to disturbance. The forest is managed primarily for timber production and protection, with recreation as an additional function (Hessen-Forst 2013).

The second study area is the 573 ha Rochauer Heide in Brandenburg (Figure 2), owned and managed by the State of Brandenburg. Mean annual temperature is 10.4 °C and precipitation 708 mm. Elevations range from 120 to 160 m a.s.l. within the forest growth region “Düben-Niederlausitzer Altmoränenland.” Soils are predominantly terrestrial, moderately nutrient-rich to poor (pH 4.2–5.1), with sandy to loamy textures corresponding to trophic class III–IV. Forest stands are characterized by open structures dominated by Scots pine (Pinus sylvestris), sessile oak (Quercus petraea), and silver birch (Betula pendula). Pine occurs across all age classes and regenerates naturally, complemented by young pre-cultivation stands. Sessile oak appears in groups or as solitary trees, with low to moderate timber volumes, occasionally high; regeneration derives from artificial planting and zoochorous dispersal by Eurasian jays (Garrulus glandarius). Birch occurs mostly in scattered individuals or clumps. Minor species include northern red oak (Quercus rubra), rowan (Sorbus aucuparia), and occasional European beech (Fagus sylvatica).

2.2. Determination of the Abundance of Different Wildlife Species

Wildlife composition and relative abundance were assessed with a thermal-imaging drone survey. The Heppenheim forest was surveyed in late November 2022, targeting roe deer (Capreolus capreolus) and wild boar (Sus scrofa). The Rochauer Heide was surveyed in early February 2023, expecting roe deer, wild boar, red deer (Cervus elaphus), fallow deer (Dama dama), and potentially western capercaillie (Tetrao urogallus). All UAV flights were conducted under diffuse light conditions in the early morning hours, avoiding direct solar radiation and precipitation. Ambient temperatures during the surveys averaged 8 °C, and wind speeds remained low to ensure stable flight conditions and optimal thermal contrast between wildlife and the surrounding environment. Both surveys employed a DJI Matrice 30T with thermal sensor (640 × 512 px), zoom and wide-angle cameras, and a laser rangefinder. Continuous operation was ensured via four battery sets and mobile charging. A two-person team conducted systematic raster flights at 100–120 m altitude, covering ~70 m swaths with 50 m transects. Heat signatures were verified by zoom imagery, georeferenced via GPS and rangefinder data, and documented with simultaneous thermal, wide-angle, and zoom images. Detected individuals were marked digitally to avoid double counts. The resulting georeferenced imagery provided verifiable evidence for cartographic analysis of wildlife presence.

2.3. Assessment of Wildlife Impact on Regeneration

Field surveys were conducted in Rochauer Heide in late May 2023 and in the Heppenheim municipal forest between early and mid-June 2023. Vegetation and browsing surveys conducted in spring have become a standard in Central Europe for several methodological reasons. First, fresh browsing marks are most visible during early shoot development, as new buds and shoots are not yet obscured by understory vegetation. This allows for the clear identification of damaged terminal shoots or missing buds, thereby increasing detection accuracy [18]. Second, omission errors are reduced in spring because 1–2-year-old regeneration plants are easier to locate and less covered by competing ground vegetation. Later surveys in summer or autumn may underestimate browsing intensity due to secondary shoot growth or compensatory foliage development [19]. Third, standardized spring surveys allow for better temporal attribution of browsing events: damage occurring in winter or early spring remains visible and can be directly linked to subsequent height growth or seedling density within the same vegetation period [20]. Fourth, environmental conditions such as leaf development, light availability, and ground vegetation density are more consistent between years and sites in spring, improving reproducibility and comparability within long-term monitoring programs [21,22]. Finally, browsing assessed in spring can be more accurately related to regeneration structure—such as height class, species composition, and vitality—since growth reductions are not yet confounded by later climatic or competitive effects. This strengthens the causal interpretation between ungulate density, browsing intensity, and regeneration success, providing valuable insights for adaptive management [23]. Wildlife impact on regeneration was assessed in both areas using a browsing inventory based on the “Neuzeller method,” a circular transect sampling approach (Figure 3). Sampling points were considered valid if they were located on stocked forest soil, within stands at least 60 years old, with open or semi-open canopies, and if natural regeneration under 180 cm in height was present. At each sampling point, three regeneration subplots of 25 m² were surveyed, totaling 75 m² per point. The center of the first subplot coincided with the sampling point. A metal rod was placed at the center and fitted with a 2.52 m cord, stretched horizontally. Attached to its end was a 1.26 m wooden rod, also held horizontally, which was used to trace a full circle, yielding a 25 m² sampling area (Figure 3). While moving along the circle, all regeneration plants were identified by species and assigned to one of four height classes (0–20 cm, 20–50 cm, 50–130 cm, 130–180 cm). Each individual was examined for wildlife impact. Terminal shoot damage was recorded as either single (a single bud or leaf of the current year’s shoot was browsed) or multiple (multiple browsing events or damage to the terminal shoot in successive years). Additionally, fraying, bark stripping, and stem breakage were documented. Sampling grid spacing varied by site: 280 m in Heppenheim and 200 m in Rochauer Heide.

2.4. Data Analysis

Data from the wildlife impact survey were recorded using a custom Excel spreadsheet based on the Neuzeller method. Data restructuring and statistical analysis were carried out in RStudio (version 4.2.2) and Microsoft Excel. QGIS (version 3.36.1) was used to create spatial maps of the monitoring data, which were subsequently accessed via the mobile application QField (version 3.3.8) during fieldwork. Cartographic representations of the drone survey results were based on vector polygons of the study areas, displayed without background layers. Sampling points for the browsing inventory were placed at the center of post-processed rectangular grid cells. In cases where rectangles extended beyond the study area boundaries, data interpretation was restricted to valid points within the delineated zones. QGIS (Version 3.36.1) was also used for spatial visualization of wildlife presence, density distributions, and browsing pressure (browsing index). The browsing index was calculated using the formula by [24], in which the proportion of multiple browsing is added to half of the proportion of single browsing:

Browsing Index = %Multiple Browsing + 0.5 × %Single Browsing

A quantitative measure of browsing pressure on tree regeneration, calculated by adding the percentage of plants with multiple browsing damage to half the percentage of plants with single browsing damage. This index integrates both the frequency and intensity of browsing and serves as an indicator of wildlife impact on forest regeneration dynamics [24]. To assess differences across browsing categories and plant height classes, data were tested for normality [25] and homogeneity of variances (Levene’s test). If assumptions were violated, the Kruskal–Wallis Test [26] was applied. In cases of significant group differences, pairwise comparisons were conducted using the Wilcoxon rank-sum test. Differences in canopy closure and browsing index between the two study areas were analyzed using the Shapiro–Wilk test for normality and the F-test for homogeneity of variances. Depending on the results, either a t-test (parametric) or Mann–Whitney U-test (non-parametric) [27] was applied. To examine differences among the three regeneration plots per sampling point, normality of plant counts by height class was assessed using the Shapiro–Wilk test. The Kruskal–Wallis test was then used to identify significant differences, followed by Wilcoxon tests for pairwise comparisons, if applicable. Citavi (version 6.19.0.0) was used for reference management.

3. Results

3.1. Results of Aerial Surveys

Results from the drone-based wildlife surveys are presented within a grid of polygon tiles. The area of each complete tile was 7.84 hectares in the Heppenheim study site and 4.0 hectares in the Rochauer Heide, reflecting differences in raster grid spacing.

3.1.1. Abundance and Density of Wildlife Species in the Heppenheim Study Area

The initial survey of the Heppenheim municipal forest was conducted over four consecutive days. Each day was characterized by heavy cloud cover, with periods of low-lying dense clouds. Flights were carried out from pre-dawn twilight through to dusk, capturing diurnal and crepuscular animal activity. Figure 4 shows the drone flight paths overlaid with georeferenced detections of various wildlife species. Flights were executed systematically in tight, serpentine transects. In total, 549 roe deer (Capreolus capreolus), 57 wild boar (Sus scrofa), nine hares (Lepus europaeus), six red foxes (Vulpes vulpes), and two raccoons (Procyon lotor) were detected. The latter three species are listed under “Other” in

Table 1. Detected wildlife in Heppenheim study area.

Species	Absolute Count (1st Survey) [n/1362 ha]	Average Count (1st Survey) [n/100 ha]	Absolute Count (2nd Survey) [n/1362 ha]	Average Count (2nd Survey) [n/100 ha]
Roe Deer	549.0	40.3	490.0	35.9
Wild Boar	57.0	4.2	100.0	7.3
Hares	11.0	0.8	13.0	0.9
Red Fox	6.0	0.4	1.0	0.07
Others	2.0	0.1	3.0	0.2

. Species abundances are reported in absolute counts [n] and as densities per 100 hectares. Roe deer were by far the most abundant species, with a mean density of 40.3 individuals per 100 ha.

A second survey in January 2024 recorded 490 roe deer, 100 wild boar, 13 hares, one fox, and three raccoons (

Table 2. Recorded game species from the first aerial survey in the Rochauer Heide study area.

Species	Absolute (Flight 1) [n/572 ha]	Average (Flight 1) [n/100 ha]	Absolute (Flight 2) [n/572 ha]	Average (Flight 2) [n/100 ha]
Roe deer	19.0	3.31	23	4.02
Red deer	2.0	0.35	2	0.35
Wild boar	7.0	1.22	28	4.89
Predators	2.0	0.35	5	0.87
Capercaillie	2.0	0.35	0	0.0
Hares	0.0	0.0	3	0.53

). This represents an 11% decline in roe deer and a twofold increase in wild boar compared to the first survey. Variable weather with sunny intervals likely reduced thermal contrast, leading to underestimation of roe deer.

Roe deer occurred across most of the area, with distinct clusters in the northeast, southeast, southwest, and central-west. Several polygons contained >8 individuals, while only few polygons were without detections. Wild boar were observed as scattered individuals and two larger sounders, one at the northern edge. Their distribution suggests morning movements into cover or resting in dense vegetation. Hares were largely restricted to the northeast, mostly as solitary animals. Foxes and raccoons appeared only sporadically; raccoons were located in tree cavities, foxes in open ground.

The spatial distribution of five wildlife species in Heppenheim is shown in Figure 5. Grid cells display roe deer abundance (color scale) and occurrences of other species (dots). Roe deer were present in nearly all cells, with highest densities (>10 individuals) in central and southern areas, moderate densities (4–10) in eastern, southeastern, and western regions, and low densities (1–3) elsewhere. Empty cells were rare and confined to the margins. Wild boar occurred sporadically, mainly in the south and southwest. Hares were concentrated in the south, mostly as solitary individuals. Foxes were widely but sparsely distributed, without clear clustering. Raccoons were detected only occasionally. Overall, roe deer dominated the community with distinct hotspots, while other species appeared at low frequency and with localized distribution.

3.1.2. Abundance and Density of Different Wildlife Species in the Study Area Rochauer Heide

The aerial survey in the Rochauer Heide (Figure 6) was completed within 1.5 days under dense cloud cover, ensuring strong thermal contrast and reliable detection of small species such as capercaillie. Flights followed systematic S-shaped transects. In total, 2 red deer, 19 roe deer, 7 wild boar, 1 raccoon, 1 red fox, and 2 capercaillies were detected (

Table 3. Total and affected number of plants per hectare and their proportions by tree species in Heppenheim.

Species	Total Regeneration [n/ha]	Percentage of Total Regeneration [%]	Regeneration Affected [n/ha]	Percentage Affected [%]
F. sylvatica	3953	46.8	2006	50.7
A. spec.	2018	23.9	1187	58.8
Q. petraea	1234	14.6	132	10.7
F. excelsior	833	9.9	443	53.2
P. avium	287	3.4	69	24.0
J. regia	30	0.4	0	0.0
A. alba	26	0.3	19	73.1
P. menziesii	22	0.3	12	54.5
L. decidua	22	0.3	22	100.0
C. betulus	11	0.1	7	63.6
P. abies	7	0.1	0	0.0

). Figure 6 shows flight paths and detection sites.

Figure 7 illustrates roe deer distribution in Rochauer Heide. Densities were relatively uniform, with most detections comprising pairs and a group of four at the southern boundary. Few animals occurred near the federal road. Red deer, predators, and capercaillie were recorded only sporadically; one of two red deer detections was just outside the study area, where four additional individuals were observed resting beyond the boundary. Wild boar were located in dense pine and spruce regeneration. Given the evergreen cover, undetected individuals of wild boar or other species cannot be excluded.

A second survey in March 2024 was evaluated only by figures, without cartographic analysis. Detected were 2 red deer, 23 roe deer, 28 wild boar, 3 hares, 2 foxes, and 3 raccoons. Roe deer density increased slightly (~+1/100 ha), while red deer remained unchanged. A large wild boar herd was recorded, and hare detections rose compared to the first flight. The observed increase in wild boar density between the two survey periods can likely be attributed to seasonal mobility and reproduction, as wild boar populations typically expand during spring and early summer due to farrowing, while favorable thermal conditions and increased food availability enhance their activity and detectability in open areas.

3.2. Evaluation of the Impact of Game

To evaluate wildlife impact, key stand parameters were assessed: total inventoried area, regeneration stages, upper canopy composition, and areas protected by fencing or individual measures. Following the “Neuzeller method,” three regeneration plots were surveyed per sampling point to test for differences among plots. Detailed regeneration data include tree species composition by height class and plant density per hectare, complemented by browsing indicators such as the number of affected versus unaffected plants, frequency of terminal shoot browsing, and evidence of fraying or bark stripping.

3.2.1. Tree Species Distribution in the Upper Stand

In Heppenheim (Figure 8), European beech (Fagus sylvatica) dominates the upper canopy with 67.4%. Sessile oak (Quercus petraea, 9.0%), Douglas fir (Pseudotsuga menziesii, 7.2%), and sycamore (Acer pseudoplatanus, 4.1%) occur at lower proportions.

Other deciduous species, including bird cherry, silver fir, red oak, hornbeam, English oak, walnut, small-leaved lime, black locust, and chestnut, are present only sporadically as single trees or small groups. Spruce (Picea abies) has been largely lost to recent calamities and persists only in isolated remnants.

In the Rochauer Heide (Figure 9), Scots pine (Pinus sylvestris) dominates the canopy with ~74%, followed by sessile oak (Quercus petraea, 15%), occurring both in stand form and as scattered individuals. Silver birch (Betula pendula, 5.5%), European larch (Larix decidua, 1.2%), and Norway spruce (Picea abies, 0.6%) are present only sporadically, while all other species occur locally in minor proportions.

Roe deer (Capreolus capreolus) are selective mixed feeders that prefer easily digestible, nutrient-rich browse such as buds and young shoots of broadleaved species like oak (Quercus spp.), maple (Acer spp.), and ash (Fraxinus excelsior) [28,29]. In Heppenheim, the high proportion of palatable deciduous species provides abundant forage and likely attracts higher browsing activity, whereas in the Rochauer Heide, the dominance of Scots pine and nutrient-poor soils limits the availability of preferred browse, resulting in lower browsing intensity.

3.2.2. Canopy Closure of the Upper Stand

Canopy closure determines light availability for natural regeneration and differs markedly between the two sites due to species dominance. The boxplot illustrates the variation in canopy closure between the two study areas, Heppenheim (green) and Rochauer Heide (blue). The y-axis represents the proportion of ground area covered by tree crowns, while the x-axis distinguishes the study sites. The boxes show the interquartile range (IQR), representing the middle 50% of observations, with the horizontal line inside each box indicating the median canopy closure. The whiskers extend to the minimum and maximum non-outlier values, and isolated points denote outliers. The mean canopy closure is slightly higher in Heppenheim (0.81) than in Rochauer Heide (0.77), suggesting denser and more heterogeneous canopy structures in the former. In contrast, the narrower range observed in Rochauer Heide reflects a more uniform forest structure with relatively consistent crown cover. In Heppenheim, European beech (Fagus sylvatica) accounts for ~67% of the canopy, while Scots pine (Pinus sylvestris) dominates in Rochauer Heide (~74%). Canopy closure values differ significantly (p = 0.0001; Figure 10), with mean values of 0.81 in Heppenheim and 0.77 in Rochauer Heide. The interquartile range is narrower in Rochauer Heide, apart from a few outliers (Figure 10).

3.2.3. Differences Between the Rejuvenation Circles

Each sampling point was subdivided into three regeneration plots following the Neuzeller method. Values refer to total regeneration per point; categories of wildlife impact were not distinguished. Data distributions were non-normal. No significant differences were detected among height classes in either study area. Species-wise analysis likewise revealed no significant differences, except for birch (Betula pendula) in Rochauer Heide. Here, the Kruskal–Wallis test indicated significant differences between plots (p = 0.0010), confirmed by Wilcoxon rank-sum tests between plot 1 and plot 2 (p = 0.0066) and plot 1 and plot 3 (p = 0.0010).

3.2.4. Assessment of the State of Regeneration

The comparative analysis of natural regeneration at the two study sites, Heppenheim and Rochauer Heide, reveals significant differences in the quality and degree of browsing damage affecting young trees, despite both locations exhibiting full regeneration coverage (Figure 11). Total regeneration density was high at both sites, with 8453 individuals per hectare in Heppenheim and 7144 individuals per hectare in Rochauer Heide, corresponding to 100% regeneration share. However, the proportion of undamaged regeneration varied markedly. In Rochauer Heide, 91.6% of the regeneration was unaffected, compared to only 53.8% in Heppenheim. This disparity indicates a substantially higher level of browsing and mechanical impact at the Heppenheim site. The share of damaged regeneration in Heppenheim reached 46.2%, while in Rochauer Heide, it remained as low as 8.4%. These figures highlight a critical difference in browsing pressure and suggest either higher ungulate densities or less effective wildlife management practices in Heppenheim. Terminal shoot browsing provides further insight into browsing intensity. In Heppenheim, 15.5% of individuals exhibited evidence of single terminal shoot browsing, and an additional 30.1% had been browsed multiple times. In stark contrast, only 2.2% of individuals in Rochauer Heide were affected, both in terms of single and repeated browsing. Repeated browsing of terminal shoots is particularly detrimental, as it inhibits vertical growth and may compromise long-term stand structure and competitiveness. Mechanical damage caused by fraying and rubbing was also more prevalent in Heppenheim (5.9%) than in Rochauer Heide (0.6%). Bark stripping was negligible at both sites, with no recorded damage in Rochauer Heide and only minimal presence in Heppenheim (0.4%).

In summary, while both sites demonstrate sufficient regeneration density, the qualitative integrity of regeneration in Rochauer Heide is markedly superior. The minimal level of browsing and mechanical damage suggests a balanced interaction between forest regeneration and local wildlife populations. In contrast, the high levels of damage observed in Heppenheim indicate an overabundant browsing pressure, which poses a significant threat to forest development and biodiversity maintenance. These findings underscore the urgent need for targeted wildlife management interventions in Heppenheim to reduce browsing intensity and promote sustainable forest regeneration.

Natural regeneration was strongly dominated by European beech (Fagus sylvatica) with 3953 individuals ha⁻¹, accounting for 46.8% of the total regeneration, followed by Maple species (Acer spp.) (2018 ha⁻¹; 23.9%) and Sessile oak (Quercus petraea) (1234 ha⁻¹; 14.6%). European ash (Fraxinus excelsior) (833 ha⁻¹; 9.9%) and Wild cherry (Prunus avium) (287 ha⁻¹; 3.4%) were less abundant, whereas all other species contributed less than 1% to total regeneration (Table 1). Browsing impact varied substantially among species. Complete browsing was recorded for European larch (Larix decidua) (100%), while high proportions were also observed in Silver fir (Abies alba) (73.1%) and European hornbeam (Carpinus betulus) (63.6%). Intermediate browsing intensities occurred in European beech (F. sylvatica) (50.7%), Maple (Acer spp.) (58.8%), and European ash (53.2%). In contrast, Sessil oak (Q. petraea) (10.7%) and Wild cherry (P. avium) (24.0%) showed relatively low browsing rates. No browsing was detected for Juglans regia and Picea abies. Overall, the data indicate a strong dominance of F. sylvatica in regeneration, while high browsing pressure on several coniferous and minor broadleaved species suggests a potential shift in future stand composition.

The following figures present a detailed analysis of the main tree species in the regeneration layer. The x-axis depicts four height classes (Class 1: 0–20 cm, Class 2: 20–50 cm, Class 3: 50–130 cm, Class 4: 130–180 cm), while the y-axis shows the average number of regeneration plants per category. For each class, total regeneration (“R. total”) is shown alongside regeneration unaffected by browsing (“R. unaffected”), regeneration with single (“T. single”) or multiple terminal shoot browsing events (“T. multiple”), and regeneration affected by fraying or breakage (“Fraying”). Rounded averages and their proportions of the total regeneration within each height class are indicated above the bars. For European beech (Figure 12), plant density declined with increasing height. In class 1, 1234 individuals ha⁻¹ were recorded, of which most were unaffected by browsing. In classes 2 and 3, browsing pressure increased, with multiple browsing events occurring about twice as often as single events. In class 4, 557 individuals ha⁻¹ remained, with the highest proportion of unaffected plants and only marginal fraying. The analyses were conducted using Shapiro–Wilk tests to assess data normality, followed by independent-sample t-tests for normally distributed variables and Mann–Whitney U tests for non-parametric comparisons between study areas. This addition clarifies the analytical basis of the visualized trends and strengthens the methodological transparency and scientific rigor of the results presentation.

Sessile oak (Quercus petraea), wild cherry (Prunus avium), and common ash (Fraxinus excelsior) were concentrated almost exclusively in the lowest height class, whereas European beech (Fagus sylvatica) and sycamore maple (Acer pseudoplatanus) showed higher overall densities and a more even vertical distribution. Beech accounted for 46.7% of all individuals, followed by sycamore maple (23.9%), sessile oak (14.6%), and ash (9.8%). Sycamore maple exhibited a comparable decline, from 1028 individuals ha⁻¹ in class 1 to only 43 in class 4. Browsing intensity increased from 40% in class 1 to over 75% in classes 2 and 3, where multiple browsing dominated. Even in class 4, 38% of the remaining individuals were browsed, indicating high susceptibility to ungulate pressure. Sessile oak (Figure 13) showed the steepest decline. Of 1227 individuals ha⁻¹ in class 1, about 90% were unaffected. In all higher classes, densities were minimal or absent, suggesting strong browsing impact combined with drought-related mortality. Common ash followed a similar trajectory (Figure 13). In class 1, 616 individuals ha⁻¹ were recorded, with 43% affected by browsing. In class 2, browsing reached 85%, with multiple browsing three times more frequent than single events. Only 14 individuals ha⁻¹ remained in class 3, evenly split between browsed and unaffected plants. No individuals occurred in class 4.

Wild cherry declined sharply from 264 individuals ha⁻¹ in class 1 to only one in class 4. In classes 2 and 3, nearly all plants were browsed, mostly multiple times. In summary, all species declined with increasing height. Beech maintained the highest regeneration potential despite intensive browsing in intermediate classes. Sycamore maple showed stronger browsing susceptibility, while sessile oak, ash, and wild cherry suffered drastic losses after the first height class. For oak, additional environmental stressors likely exacerbated mortality. In the Brandenburg study area, sessile oak (Quercus petraea) clearly dominated the regeneration layer, accounting for 57.2% of all individuals (4089 plants ha⁻¹). Field observations confirmed a widespread and vigorous natural regeneration of sessile oak, predominantly of zoochorous origin. Scots pine (Pinus sylvestris) was the second most common species, with 1318 plants ha⁻¹ (18.4%), despite being the most abundant species in the overstory. Other species, including rowan (Sorbus aucuparia), silver birch (Betula pendula), European beech (Fagus sylvatica), and northern red oak (Quercus rubra), occurred only sporadically and each contributed less than 10% to total regeneration, thus representing secondary species within the regeneration layer. Figure 14 displays regeneration density per hectare by tree species and height class. Across the first three height classes, sessile oak consistently exceeded all other species, with particularly high densities in the lowest class. Approximately 1100 individuals ha⁻¹ were still present in the two intermediate classes, whereas only ~314 individuals ha⁻¹ remained in the tallest class, reflecting a marked decline. In some stands, however, oak individuals had already surpassed the defined height threshold for regeneration measurement. Scots pine, rowan, and silver birch showed increasing mean densities across the lower height classes, with rowan maintaining this trend into the tallest class. By contrast, Scots pine and silver birch declined in the upper classes. Other species such as European beech, northern red oak, pedunculate oak (Quercus robur), and Norway spruce (Picea abies) displayed variable but inconsistent patterns across height classes, from which no clear trend could be inferred based on descriptive data alone. To evaluate the effects of ungulate browsing and potential species-specific differences in browsing pressure, the dominant and ecologically relevant secondary tree species are examined in greater detail in the following analysis.

For Sessile oak (Q. petraea) showed highest densities in the lowest height class (Figure 15), with numbers declining sharply in taller classes. Browsing pressure increased with height, peaking at 15.6% repeated browsing in class 3. Fraying and breakage remained rare (<1.2%). Scots pine (P. sylvestris) displayed the opposite pattern, with increasing densities from class 1 to 3, before declining in class 4. Browsing was generally low (<1.2%), but fraying accounted for 3.2% in the tallest class. European beech (F. sylvatica) was scarce, averaging ≤ 94 plants ha⁻¹ in the first three classes and only 20 in class 4. Browsing was negligible except in class 3, where 12% of individuals were affected. No fraying was recorded. Rowan (S. aucuparia) increased steadily across height classes, reaching 281 individuals ha⁻¹ in class 4. Browsing rose correspondingly, with the highest impact (10.8% multiple browsing) observed in the tallest class. Birch (Betula spp.) followed a similar trend to Scots pine, increasing up to class 3 and declining thereafter. Browsing was minimal (<3%), with fraying and bark-stripping detected only in class 4 (1.8%).

Overall, browsing intensity peaked in the 50–130 cm height class, with sessile oak most strongly affected. In contrast, Scots pine, rowan, and birch showed higher potential to escape browsing pressure by advancing into taller classes. Across species, the proportion of individuals damaged by browsing or fraying generally remained below 10%.

3.2.5. Comparison of Wildlife Impact in the Study Areas Heppenheim and Rochauer Heide

Both study areas were surveyed using a standardized wildlife impact inventory, limited to areas without fencing or other wildlife protection measures. In Heppenheim, 114 sample plots covering a total area of 8550 m² were recorded, while in the Rochauer Heide, 105 sample plots totaling 7875 m² were included. Plots were excluded where regeneration exceeded 180 cm in height or where structural anomalies (e.g., plantations, insufficient canopy openings) were present. In both regions, the surveys exclusively represent natural regeneration conditions without protective interventions.

• Quantitative Wildlife Impact

In Heppenheim, 46.2% of all recorded seedlings were affected by ungulate browsing, with 30.1% heavily (multiple times) and 15.5% lightly (once) browsed. Fraying was detected in 0.6% of individuals. In contrast, browsing intensity in Rochauer Heide was substantially lower at 8.4% (5.9% heavy, 2.2% light), with fraying affecting only 0.4% of plants. Bark stripping was absent in both areas. Finding: The impact of browsing in Heppenheim was more than fivefold higher than in Rochauer Heide, likely reflecting differences in ungulate density, hunting pressure, forest structure, and habitat composition.

For Species Composition and Abundance in Heppenheim was found out regeneration was dominated by European beech (Fagus sylvatica, 46.8%), followed by sycamore maple (Acer pseudoplatanus, 23.9%), sessile oak (Quercus petraea, 14.6%), and common ash (Fraxinus excelsior, 9.9%). In Rochauer Heide, sessile oak prevailed (57.2%), accompanied by Scots pine (Pinus sylvestris, 18.4%), rowan (Sorbus aucuparia, 9.6%), and silver birch (Betula pendula, 5.0%). European beech was scarce (3.6%). Finding: Species composition mirrors site conditions and overstory structure. Beech dominance in Heppenheim reflects mesic colline deciduous forests, whereas oak–pine regeneration in Rochauer Heide corresponds to drier, sandy habitats.

For Species-Specific Wildlife Impact in Heppenheim, European beech (50.7%) and sycamore maple (58.8%) were most affected, primarily by repeated browsing. Common ash also showed marked damage in intermediate height classes. Sessile oak was less impacted (10.7%) but occurred mainly in the lowest height class. In Rochauer Heide, sessile oak showed the highest browsing intensity (10.8%), particularly in height classes 2 and 3 (up to 24% combined). In contrast, Scots pine (1.4%), birch (2.5%), and rowan (11.0%) exhibited much lower browsing levels and established successfully despite occasional impacts. Finding: Preferred hardwoods in Heppenheim experienced high browsing pressure, while in Rochauer Heide even sensitive species such as oak were able to establish, though browsing increased at critical developmental stages. In both study areas, browsing pressure peaked in height classes 2 (20–50 cm) and 3 (50–130 cm). In Heppenheim, maple, ash, and beech exhibited multiple browsing rates of 50%–75% in these classes, coupled with strong declines in regeneration density at greater heights. Beech, although affected, demonstrated relative resilience. In Rochauer Heide, sessile oak showed increasing browsing pressure with height, while Scots pine, birch, and rowan maintained or increased densities in higher classes and were only marginally browsed. Finding: Browsing pressure concentrated on height strata most accessible to ungulates, with clear interspecific differences in susceptibility and resilience. Middle height classes consistently proved most vulnerable.

2.

Browsing Index

To quantify spatial variation in browsing intensity, a browsing index was calculated based on the categories multiple browsing and single browsing. The index ranges from 0 to 1 and is visualized using a red color gradient, with higher values indicating greater browsing pressure. Grey polygons denote sampling units without data, including truncated units at the study area margins, while white polygons represent low values (0.0–0.1). In Heppenheim, the index was calculated for European beech, Sycamore maple, Alder, European ash, Sessile oak, and Wild cherry. In Rochauer Heide, the analysis included Sessile oak (Quercus petraea), Scots pine (Pinus sylvestris), rowan (Sorbus aucuparia), silver birch (Betula pendula), and European beech (Fagus sylvatica). Differences between study areas were statistically significant (Figure 16). In Heppenheim, the browsing index ranged from 0 to 1, with a mean of 0.386 and an interquartile range of 0.251–0.555, indicating both higher and more variable browsing pressure. By contrast, Rochauer Heide showed a lower mean index (0.095) and reduced variability (IQR: 0.025–0.137).

In Heppenheim, elevated browsing index values were concentrated in the central and northeastern parts of the upper study area, with additional hotspots in the eastern and western sections of the lower area (Figure 17). In Rochauer Heide, browsing indices were generally low, often close to zero in the northeastern and central regions (Figure 18). Slightly higher values occurred in the southern and southeastern sectors below the federal road and along the western boundary. The relatively uniform distribution of browsing indices in the northern part of Rochauer Heide likely reflects consistent silvicultural and wildlife management practices within the forester’s district. By contrast, elevated values at the southern and western peripheries probably result from differing management regimes in adjacent forest areas and potential ungulate immigration. Analysis of aerial survey data suggests that roe deer were randomly distributed across the landscape, with no discernible spatial pattern.

The data reveal a markedly higher browsing impact in Heppenheim compared to Rochauer Heide. In Heppenheim, browsing imposed strong selective pressure on preferred deciduous species, resulting in structural homogenization and favoring browsing-tolerant taxa such as beech. By contrast, regeneration in Rochauer Heide exhibited a more balanced dynamic: although browsing intensity increased in mid-height classes, species such as oak, birch, and pine successfully established and progressed into higher strata. Browsing pressure peaked in the 50–130 cm height zone, corresponding to the typical foraging range of roe deer. Consequently, regeneration was shaped less by overall suppression than by vertical and species-specific browsing patterns, reflecting differences in species tolerance and ungulate preference. Intensive browsing in Heppenheim constrained the establishment of target species, whereas Rochauer Heide maintained greater potential for structurally and compositionally diverse regeneration. These findings underscore the need for site-adapted wildlife management and silvicultural strategies to sustain forest resilience under continuous browsing pressure.

4. Discussion

In this study, two forest areas with contrasting regional characteristics were surveyed using drone-based aerial wildlife detection. Simultaneously, ungulate browsing impact on natural regeneration was assessed. The primary objective was to quantify ungulate presence and density, and to evaluate whether browsing pressure provides reliable inferences on regeneration dynamics. The following discussion critically examines the applied methodologies of drone-based wildlife monitoring and browsing assessment, their respective outcomes, and the relationship between ungulate occurrence and browsing intensity.

4.1. Thermal Drones as a Suitable Tool for Wildlife Surveys

Unmanned aircraft vehicles (UAVs) have been employed for wildlife detection for several years [30,31,32]. Surveys can be conducted with thermal infrared (IR), video, or high-resolution still cameras [33]. For example, in bird colony counts, such sensors are operated during autonomous flights, enabling subsequent orthomosaic analysis with dedicated algorithms [25]. Recent years have seen rapid technological advances in UAV systems, with each new generation offering improved performance, such as greater battery capacity, weather resistance, camera resolution, and georeferencing accuracy. The drone model used in this study, the DJI M30T, is currently the most advanced option in DJI’s enterprise series in terms of technical specifications. Its high-sensitivity thermal sensor enables reliable wildlife detection from flight altitudes of 100–120 m, covering a ground field of view of approximately 70 m. As [33] anticipated, such technological developments have substantially increased the efficiency, accuracy, and cost-effectiveness of UAV-based surveys.

4.1.1. Detection of Animals

Israel [32] found that thermal drones perform particularly well under cool, overcast conditions without direct sunlight. Low flight altitudes further improve the detectability of small animals such as fawns. Strong solar radiation can cause heterogeneous heating of forest surfaces [34], leading to secondary heat signals and reduced contrast between target and background [35]. Our surveys confirmed the reliable detection of small animals in both study areas.

4.1.2. Roe Deer Density and Browsing Pressure in Heppenheim

In Rochauer Heide, capercaillies, raccoons in tree canopies, and foxes were recorded, whereas in Heppenheim (Figure 19) carnivores and hares were observed. Both surveys were conducted under heavy cloud cover, which minimized potential detection errors caused by background heat signals. Errors were therefore more likely attributable to human oversight or dense vegetation. Reference [36] reported an average detection rate of 88% for roe deer and artificial heat sources under leaf-off conditions, with 100% detection in deciduous stands.

4.1.3. Population Structure and Regeneration Dynamics in Rochauer Heide

In contrast, lower detection rates occurred in dense coniferous regeneration, particularly spruce [36]. The present surveys, conducted in late November (Heppenheim) and mid-February (Rochauer Heide), likewise took place outside the growing season, when leafless trees reduced visual obstruction and thereby enhanced detection accuracy (Figure 20).

False positives or overestimations—such as double counts—remain a methodological concern [37]. The DJI M30T reduces this risk by enabling “pinpoint” marking with a laser rangefinder, which carries forward to subsequent flight paths. If only one animal was previously marked but two heat sources appear, the operator can verify the presence of an additional individual. Reference [35] observed higher roe deer detection rates at night, likely reflecting increased nocturnal activity. Similarly, [12] reported greater white-tailed deer counts during evening twilight than in early morning. Reference [38] showed that daytime flights are also feasible under overcast conditions. Consistent with these findings, our Rochauer Heide flights were conducted under heavy cloud cover, yielding stable detection rates for roe and red deer. By contrast, the second flight in Heppenheim took place under partially sunny skies, which may account for the comparatively lower roe deer count. Despite considerable technological progress, the accuracy of drone-based thermal detection of wildlife remains limited by the current state of sensor technology and by environmental factors affecting infrared signal quality. In particular, canopy closure represents a key factor that significantly influences detection probability. A dense forest canopy can partially absorb or scatter the heat emitted by animals, thereby weakening their thermal signatures and reducing detection accuracy. Reference [39] demonstrated that the reliability of species identification decreases markedly under closed canopies, as the temperature contrast between animal bodies and their surroundings diminishes with increasing vegetation density and greater sensor distance. Similarly, Berezina [40] noted that high leaf density and atmospheric humidity can affect the propagation and reflection of thermal radiation, thereby reducing detection performance under canopy conditions. Nevertheless, drone-based thermal imaging currently represents the most precise and least invasive method for detecting wildlife in forested habitats. Modern high-resolution thermal sensors operating in the 8–14 µm spectral range enable reliable detection of medium- to large-sized mammals, particularly when the camera is oriented perpendicularly and flight altitudes range between 60 and 120 m [39]. A recent systematic review by [41] further confirmed that, despite remaining uncertainties caused by vegetation occlusion, UAS-based systems outperform traditional ground-based survey methods in terms of spatial coverage, detection reliability, and minimal disturbance. In summary, while a high degree of canopy closure continues to present a technological challenge that can limit detection probabilities—particularly for smaller species—UAS-based thermal imaging currently represents the highest achievable level of accuracy for wildlife detection under real forest conditions. Future advances in sensor technology, such as improved thermal sensitivity, multispectral data fusion, and AI-assisted image analysis, are expected to further reduce these limitations. Until then, drone-based thermal remote sensing provides the most methodologically reliable foundation for estimating wildlife densities and monitoring animal behavior in structurally complex forest ecosystems.

4.2. Wildlife Disturbance Due to Drone Flights

The DJI M30T generates considerable rotor noise at flight altitudes of 100–120 m, particularly under windy conditions [36]. Such noise can disturb wildlife, especially when drones are operated recreationally with erratic flight patterns [42]. Nevertheless, [36] reported no drone-induced behavioral changes in roe deer and only minimal responses in red and fallow deer. Reference [43] recommend a minimum flight altitude of 90 m for deer species when using drones of comparable size to the DJI M30T. In the Rochauer Heide study area, forests extend in all directions. Survey launches required vehicle access, which may have disturbed red deer prior to flights [4]. Observed red deer retreated from approaching vehicles during detection, suggesting that pre-flight disturbance was caused by vehicle and generator noise rather than drone activity (Figure 21).

In Heppenheim, only roe deer and wild boar were expected. The hilly terrain facilitated efficient survey coverage across slopes, reducing the need for repositioning and thereby minimizing disturbance. According to [44], deer exhibit lower sensitivity to disturbance outside their active phases at dusk and night, indicating that daytime flights may be more suitable.

4.3. Hovering and Pausing Capability

A key advantage of UAVs, particularly quadcopters, is their ability to hover and pause during flight [33,36]. Although greater altitude and speed increase survey coverage, image resolution must remain adequate [33]. In addition, the capacity to rotate and maneuver in three dimensions enables precise inspection of heat signals, offering a clear advantage over fixed-wing or manned aircraft.

Surveys were conducted manually by a two-person team consisting of a pilot and a camera operator. Dense forest cover can obstruct drone signals, making manual operation particularly advantageous. Narrower flight paths increase detection density but may reduce accuracy [36]. The camera operator monitored the thermal feed for subtle signals, guided the pilot, and ensured that heat sources were visually confirmed and documented. This division of labor reduces fatigue and improves concentration [45], thereby enhancing detection accuracy in complex environments (Figure 22 and Figure 23).

4.4. Interpretation of Aerial Survey Results

In the Heppenheim study area, both drone surveys confirmed the presence of roe deer and wild boar, along with other mammal species. The deciduous-dominated urban forest was surveyed in November and January, when the canopy was leafless. Weather conditions differed: the first flight was conducted under dense cloud cover, the second under intermittent solar radiation. To ensure reliable thermal detection, north-facing slopes were preferentially surveyed during brighter periods. According to the local forester, no changes in hunting practices occurred between flights that could explain the observed decline in roe deer numbers [45]. Harvest planning in community-managed hunting districts has relied on paired exclosure control methods, but diverging assessments by managers and hunters led to the commissioning of aerial surveys. While hunters estimated 5–10 roe deer per 100 ha and the municipality assumed ~20 per 100 ha, drone-based counts yielded ~40 per 100 ha, which hunters contested [44]. Despite technical and observer-related uncertainties, these counts represent minimum estimates. Unlike other studies e.g., [35,37], no correction factor was applied. Given the low proportion of conifer regeneration, the risk of undetected signatures was considered minor [36], although cloud cover, ground warming, or foliage may still have reduced detection probability.

In Rochauer Heide, the state forest enterprise of Brandenburg (LFB Lübben) has implemented a hunting regime for more than 15 years aimed at promoting natural regeneration without protective measures, regulating ungulate densities, and advancing near-natural forest stands. These measures were accompanied by the project “Target-Oriented Hunting in Forests, (ZIORJA). In the absence of reliable density data, population adequacy was primarily assessed via regeneration status. Drone surveys confirmed the presence of red deer, roe deer, and wild boar. The results suggest that this strategy effectively maintains low and balanced population densities. Differences between flights were negligible and within expected short-term variability [12,36]. Alongside various deciduous species, Scots pine regeneration was abundant; although less dense than spruce, it may still have reduced detectability. Overall survey conditions were optimal, enabling detection even of small mammals. To contextualize the results, home-range sizes of roe and red deer must be considered. For roe deer, Ellnberg [46] equated home range with territory, ranging from 7 to 108 ha (median 57 ha), though size varies with individual behavior. Later studies reported negative correlations with forest cover (~15 ha for females, ~30 ha for males: [47] or seasonal variability (49–109 ha in males: [48]. For red deer, average home ranges of 79 ha for males and smaller for females were found in resource-rich areas [49], while much larger ranges (500–2,000 ha) were observed in Saxony [49]. Applied to the polygon grid, roe deer home ranges in Rochauer Heide would cover ~4–8.5 polygons, and in Heppenheim ~2–4 or more. Red deer ranges would span ~20 polygons or the entire Rochauer Heide area. Given the snapshot nature of drone surveys, spatial variation in local densities is expected, as illustrated by the heterogeneous distribution maps in Heppenheim. At high densities, roe deer often occur in the absence of predators, limited interspecific competition, or insufficient hunting pressure [46]. Over time, such populations are regulated by intraspecific competition and carrying capacity, with reduced reproduction once thresholds are reached. In Heppenheim, hunters increasingly reported does without fawns [47]. Drone surveys, however, provided only minimum counts without sex or age differentiation. With densities of 35–40 roe deer per 100 ha, overlapping ranges and competition are likely, varying by sex, age, and reproductive status but generally increasing with density [47].

In Rochauer Heide, roe deer density was 3.3–4 individuals per 100 ha, suggesting low competition for food and cover. Browsing assessments confirmed extensive regeneration, indicating sufficient forage and densities below carrying capacity. No data on body weight or fawn-per-doe ratios were available, precluding comparison with [47]; however, roe deer home ranges in this continuous forest landscape are plausibly smaller than in agricultural or mixed-use systems [47], implying minimal competition. Red deer density was 0.35 per 100 ha (two individuals in total). While Thomae [50] suggested overlapping ranges at such densities, other studies [50,51] indicate otherwise, with supplementary feeding reducing range size. Despite potential overlap, the detected red deer density remains low, and the two individuals may have moved beyond the study area shortly after surveys. Given the available forage and cover, higher densities would likely have resulted in additional detections.

4.5. Assessment of Wildlife Impact

The wildlife impact assessment method, adapted from the “Neuzelle Method,” was applied operationally for the first time in this study. This novel implementation closes a methodological gap in forestry practice and provides a new basis for evaluating regeneration processes under browsing pressure. The following sections examine its suitability, the occurrence of tree species, and the resulting patterns of species diversity and wildlife impact.

4.6. Suitability of the Wildlife Impact Assessment Method

Wildlife impact was assessed using a procedure adapted from the “Neuzelle Method,” originally developed for forest conditions in Brandenburg. Several modifications were introduced in this study. The method integrates sample points into the existing grid of the Brandenburg National Forest Inventory (LWI; 200 × 200 m), which records three sample plots per point, with the largest radius being 12.98 m. In the original procedure, the first regeneration circle is located at the northern edge of the LWI plot. Here, the circle center was shifted to the central sampling point, as the objective was not to collect stand structural parameters but to assess browsing pressure. Unlike conventional browsing assessments that record seedlings only above 10–20 cm, this study included all regeneration from 0 cm upwards to capture browsing effects already at the seedling stage. Browsing was further classified into single and multiple events, enabling detection of cumulative browsing pressure over several years. These adjustments address a common limitation of large-scale inventories, where browsing is recorded only on existing regeneration, offering no insight into suppressed or absent species. A central hypothesis was whether using three regeneration circles per sample point influences browsing assessment. Statistical analyses revealed no significant differences between circles—neither across all species nor within individual taxa. The only exception occurred in birch regeneration in Rochauer Heide, but this is likely coincidental, as it was restricted to one species and lacks comparative support. The Neuzelle Method was developed for homogeneous Scots pine stands in Brandenburg, where relatively uniform light conditions favor even regeneration. This fits well with Rochauer Heide, where canopy closure is consistent across the site. In contrast, Heppenheim exhibits higher canopy closure. Here, the method’s strength lies in recording regeneration from the seedling stage upward, revealing that while seedlings establish below 20 cm, few individuals advance into higher strata—indicating suppression or mortality before reaching later stages.

4.7. Ungulate Species and Their Impact on Tree Regeneration

Drone surveys in both study areas confirmed the presence of roe deer, wild boar, and hares; in Rochauer Heide, red deer were additionally detected. All these species are capable of causing economic damage in forestry. Roe deer are widespread and among the most relevant browsers in Central Europe. As selective feeders, they preferentially consume terminal shoots, leaves, and buds of young deciduous and coniferous trees. Although browsing constitutes their primary impact, rubbing and fraying also occur. Their foraging selectivity often promotes competitive advantages for less preferred species while suppressing rare or silviculturally valuable taxa. Red deer are associated primarily with bark stripping, particularly on young trees. Human disturbances may trigger herd formation, leading to severe localized damage. As intermediate feeders with a tendency toward grazing, they also cause browsing, rubbing, and fraying. Wild boar, as omnivores with a broad dietary spectrum, damage regeneration mainly through rooting and seed predation. While soil aeration may offer short-term benefits, rooting generally reduces natural and artificial regeneration success. Hares can inflict heavy browsing damage in plantations and small enrichment plantings, often resulting in high mortality or regeneration failure. Effective protection typically requires hare-proof fencing. When browsing damage is observed, species attribution is not always possible. Roe deer browse up to ~120 cm and red deer up to ~150 cm, with higher pressure expected under prolonged snow cover. Additional indicators such as tracks, droppings, and bite marks are essential for differentiation [52]. Browsing, fraying, and bark stripping occur year-round but vary seasonally: summer vegetation may reduce browsing and bark stripping, whereas fraying and rubbing peak during antler development. Tree species differ markedly in attractiveness. Silver fir, oak, maple, and noble hardwoods such as rowan, wild service tree, ash, whitebeam, and wild cherry are particularly susceptible. Beech is generally less palatable, though often bark-stripped in summer when its bark peels easily. Among conifers, Scots pine and European larch are most frequently frayed, while spruce and fir are less affected. Spruce and fir, however, are highly susceptible to bark stripping in winter, whereas wild cherry and rowan are among the broadleaves most at risk of fraying [53].

4.8. Height Differences in Natural Regeneration and Their Modification by Ungulate Browsing

The distribution of height classes in natural regeneration and associated browsing pressure provide insights into species performance. Terminal bud removal is among the most severe effects of ungulate browsing, causing forking and stem deformities in broadleaves and reducing timber quality. Repeated browsing can reduce height growth by more than 25% and biomass by up to 40%, particularly in maple and oak seedlings [54]. Broadleaves on nutrient- and water-rich soils may compensate single events, whereas conifers show limited recovery from repeated browsing [4]. Canopy closure additionally restricts light-demanding species such as oak [55]. The critical vulnerability phase corresponds to the period when regeneration remains within browsing height. Kupferschmift [56] describe this “outgrowth period,” noting that repeated browsing depletes reserves, with height growth losses >25% potentially lethal. On poor sites, compensation is limited, and seedlings may be entirely consumed, leading to underestimation of browsing intensity. In Heppenheim, regeneration was dominated by beech (Fagus sylvatica, ~50%), half of which was browsed. Maple (Acer pseudoplatanus) accounted for 25% (60% browsed), ash (Fraxinus excelsior) for >50%, and cherry (Prunus avium) for 24%. Browsing peaked in height classes 2–3. Beech, usually less affected where abundant [57], showed unexpectedly high rates, with >66% of individuals in classes 2–3 browsed, 55% repeatedly. In class 4, >80% remained unaffected, confirming reduced vulnerability above browsing height. Fraying was rare but more frequent in taller plants. Sycamore maple, highly susceptible [58], showed nearly 80% browsing in classes 2–3. Its limited compensatory ability [59] and repeated browsing [60] explain its decline in Heppenheim. Sessile oak (Quercus petraea) is equally susceptible [59,61]. While class 1 contained many unaffected seedlings, higher classes were nearly absent, indicating mortality mainly due to browsing. Secondary factors include shading, drought, and seed predation [62,63]. Ash regeneration was limited to the first three classes, with >40% browsed in class 1 and only a few individuals surviving in class 3. Its high susceptibility [1] and delayed negative response [64] suggest strong browsing-induced decline. Wild cherry, subject to roe deer browsing [65], occurred at low densities and almost disappeared beyond class 2, reflecting selective browsing [22,52,66,67,68,69,70]. Drought and shade tolerance [62] did not mitigate this effect. Other minor species (Abies alba, Pseudotsuga menziesii, Larix decidua, Carpinus betulus, Acer platanoides) occurred rarely (<26/ha) and likely suffered high mortality from seedling browsing [59,71], consistent with their classification as highly susceptible.

5. Conclusions

The aim of this study was to assess whether modern thermal drone surveys can be used to estimate browsing impact by ungulates. To this end, several drone-based wildlife surveys were conducted in two different managed forests. The resulting imagery enabled the identification of differences in wildlife presence and their temporal consistency. These findings were related to the results of parallel assessments of browsing impact, thereby contributing to the broader forest–wildlife interaction discourse. The objective of this thesis was to examine whether aerial surveys using a modern thermal imaging drone can enable an estimation of wildlife impact.

To this end, multiple wildlife population surveys were conducted in two different managed forests using drone flights. The resulting thermal images allowed for the identification of wildlife presence and its spatial and temporal consistency. The results of an additional field-based wildlife impact assessment—showing 46% of regeneration browsed in Heppenheim vs. 8% in Rochauer Heide—were then evaluated.

5.1. Drone Surveys for Estimating Wildlife Populations

The drone model and flight methodology applied in this study enabled reliable detection of various wildlife species and revealed clear differences between the two areas. Accounting for weather conditions, even small animals were detected. Detection proved effective in both the open, thinned pine stands of Rochauer Heide and the leafless, branch-rich beech stands of Heppenheim. With increasing flight experience, species recognition and photographic documentation were consistently feasible. Although detection errors due to stand structure cannot be excluded, detection rates are presumed to approximate actual population sizes. All surveys were conducted during daylight, when wildlife typically remains in resting phases, thereby facilitating counts. Depending on topography, density, and weather, daily coverage of 300–800 ha per system was feasible. The technology provided accurate censuses, and georeferenced imagery allowed precise spatial localization of observations. However, a direct quantitative relationship between wildlife density and browsing pressure could not be established. Wildlife impacts differed measurably between sites despite verified density estimates, indicating that browsing pressure can be coarsely inferred from density. Thus, drone-based surveys enable approximate assessment of browsing intensity or browsing indices, but precise interpretation requires additional site-specific information.

5.2. Browsing Impact in the Study Areas

To assess ungulate impacts, the wildlife impact zone (0–180 cm) was divided into four height classes. For each class, the number of individuals and the type of damage—browsing, fraying, and bark stripping—were recorded. The survey also provided information on overstory and light availability, enabling a differentiated evaluation of browsing effects across species and height classes. Differences in browsing intensity between the two study areas can be attributed to variations in wildlife density, consistent with descriptions in the literature. The first hypothesis, that wildlife impacts vary across height classes, was partially confirmed. Most browsing occurred between 20 and 130 cm, with species-specific preferences partly consistent with literature. Under high wildlife densities, certain species were browsed so heavily that they were absent from the upper classes. The second hypothesis, that regeneration diversity depends on wildlife presence, was confirmed. In Heppenheim, high browsing pressure led to the survival of almost exclusively European beech (Fagus sylvatica) in the tallest class, while other main species declined due to mortality. In Rochauer Heide, browsing intensity was low across all species, allowing multiple species to outgrow the browsing zone and maintain diversity. The third hypothesis, that plot-level differences influence results, was rejected. Variations between plots were mostly non-significant, indicating even distribution of wildlife impact and no localized browsing preferences. The fourth hypothesis, positing a correlation between wildlife density and browsing intensity, was confirmed. The Hessian site showed high densities and high browsing pressure, while the Brandenburg site exhibited both low densities and low impact. These findings align with the first two hypotheses, though unmeasured factors may also influence regeneration outcomes.

5.3. Summery

This thesis investigates the use of modern thermal imaging drones (UAS—Unmanned Aircraft Systems) for estimating wildlife density and assessing wildlife browsing impact in managed forests. Two ecologically contrasting forest areas in Germany—Rochauer Heide (Brandenburg) and the Heppenheim city forest (Hesse)—were selected to —explore whether drone-based detection of ungulate presence can be used to draw conclusions about their impact on forest regeneration. Systematic drone surveys were conducted using a DJI M30T equipped with thermal and optical imaging technology. These flights successfully detected multiple wildlife species, including roe deer (Capreolus capreolus), wild boar (Sus scrofa), and red deer (Cervus elaphus), under various habitat and weather conditions. The gathered georeferenced imagery allowed for the mapping and quantification of wildlife presence with high spatial resolution. In parallel, field-based assessments of wildlife impact were carried out using a modified circular transect method (“Neuzeller Verfahren”) to document the number, height class, and browsing damage (including single/repeated browsing, fraying, and bark stripping) of regeneration-layer tree individuals. The calculated browsing index was then compared to observed wildlife densities. The findings demonstrate that browsing pressure varied significantly between height classes and tree species. In Heppenheim, high browsing intensity, especially between 20 and 130 cm, led to reduced regeneration diversity, with European beech (Fagus sylvatica) being the only species reaching upper height classes. In contrast, Rochauer Heide showed low browsing pressure and higher species diversity in regeneration. Statistical analyses confirmed that wildlife browsing patterns correlate with measured wildlife densities. The study also found no significant influence of sample plot quantity on browsing assessment outcomes. These results support the hypothesis that drone-derived wildlife density can be used as a proxy for browsing pressure, especially when combined with ground-based assessments. However, additional ecological and site-specific factors must be considered for accurate interpretation. This work demonstrates the potential of thermal drone technology as a reliable, non-invasive tool for wildlife monitoring and contributes to improving adaptive wildlife and forest management strategies.