1 Introduction

The effective meeting of conservation objectives is a highly sought after, and ongoing goal within conservation organisations. Meeting objectives effectively allows these groups to achieve more and better preserve species and habitats. However, to do this, knowledge of the target species or habitat is required, as well as the monitoring of progress towards conservation objectives, so that management plans can be adjusted based on whether the conservation action is having the desired effect, and at what rate progress is being made towards conservation objectives (Berger-Tal and Lahoz-Monfort 2018). However, it is becomingly increasingly recognised that habitats and sites are highly dynamic (Zeller et al. 2020), with constant changes occurring that could affect the progress of any management goals. This requires management plans and targets to be constantly adjusted (Estes Jr. et al. 2021), which in turn requires frequent monitoring of progress to make informed conservation decisions, as well as adjust management objectives based on new information (Pullin et al. 2013; Kapos et al. 2009).

The monitoring and management of scrub vegetation is a particularly challenging aspect of site management. As an important habitat type, scrub often acts as an ecotone between woodland and more open habitat, and contains diverse communities and species encompassing a wide range of vegetation (El Balti 2021; Gimingham, Chapman, and Webb 1979). Scrub can also increase the biodiversity of a site, often having more species variety than woodland, and can act as a suitable alternative habitat if a species is displaced from its natural habitat (Keith et al. 2014; McArthur and Kitchen 2007). However, too much scrub can start encroaching on existing woodland and grassland habitats and, if not managed, the scrub can succeed into woodland, which would not be suitable habitat for many grassland and heathland species. It is important to know how scrub levels are changing on a site over time as a result of scrub management, to know whether more or less scrub management is required, as well as to ensure that current scrub management methods are having an effect. Because of this, a lot of UK site management goes into scrub control and the managing of scrub levels, and many site designations, such as sites of special scientific interest (SSSIs) and special areas of conservation (SACs) have conditions related to maintaining and managing suitable amounts of scrub on a site.

Current ground-based assessments of scrub, are usually based on subjective assessments of scrub cover by surveyors, typically in the form of structured walks as recommended by Common Standards Monitoring Guidance (Joint Nature Conservation Committee 2019). While this does allow for a broad assessment of changes in scrub levels, it does not provide an accurate measure of change across the whole site. Depending on when along the transect the scrub assessments are made, inaccurate assessments could be made about whether scrub has increased or decreased across the site as a whole. An example of this could be the areas where the scrub assessments are carried out having less scrub, leading to a conclusion of lower scrub levels despite overall scrub levels across the site having increased. The measured levels of scrub cover are also often subjective, with the assessor making a broad estimate based on what they can see. More frequent surveys may exacerbate this problem, as a surveyor may be less likely to notice slight changes in the vegetation on-site if they are there frequently, and increased familiarity with the site could lead to a lack of objectivity. On-foot surveying can also be time-consuming and costly, especially for organisations that manage large numbers of sites.

Scrub is loosely defined, encompassing a wide range of communities and compositions (El Balti 2021; Dierßen 2008), and is characterised based on its state as a transitional habitat between ground vegetation and woodland, defined as ‘all stages from the scattered bushes to closed-canopy vegetation… usually less than 5 m tall’ (Mortimer et al. 2000). Because of this, the spatial attributes of vegetation, primarily its height, can be used to distinguish scrub from woodland and ground vegetation. Single vegetation layers, including scrub, in semi natural systems, can be isolated from each other through the categorisation of the surface layer based on its height. This approach has been used successfully when calculating the biomass of shrub-grassland habitats (Cunliffe, Brazier, and Anderson 2016), and previous studies have used the structural component of vegetation height as a sole defining characteristic of scrub (Vafidis et al. 2021) using the typical range of 1–5 m (Mortimer et al. 2000). In this manner, the scrub in an area can be characterised and quantified by assessing the structural component of vegetation height.

Unmanned aerial vehicles (UAVs), or drones, have been identified as a potential alternative to ground-based scrub monitoring through the use of high-resolution imagery. Drones have been shown to provide more accurate data than traditional ground-based methods with regards to habitat mapping and land cover analysis, primarily in the form of quantification of vegetation structure (Wich and Piel 2021; Ancin-Murguzur et al. 2019), having higher resolution than satellite imagery (Inoue 2020) that often cannot identify small vegetation patches due to a lack of image resolution (Marston et al. 2017). While drones have been used to measure spatial and temporal changes in vegetation structure and habitat, most current studies rely on expensive equipment or complex methodologies such as LiDAR (Resop, Lehmann, and Hession 2021; Hyyppä et al. 2020), multispectral sensors (Villoslada et al. 2020) or complex machine learning (Cruz et al. 2023; Detka et al. 2023), which may be unfeasible in terms of cost or training requirements.

In 2021, a study was carried out using photogrammetric point cloud modelling of UAV-derived imagery to identify scrub cover based on height at two sites; Daneway Banks and Flatholm Island (Vafidis et al. 2021). The study found that consumer-grade UAVs can be used to generate structure-from-motion (SfM) point clouds derived from aerial images to generate 3D models at a very high resolution (Vafidis et al. 2021). This method allows accurate data on scrub levels to be collected without the use of expensive sensors or machine learning methods.

This method of scrub assessment appears to have high replicability and ease of accessibility, making it potentially of use to conservation practitioners. This method could hypothetically be used over multiple years, allowing temporal changes in scrub cover to be easily determined, as well as specific changes in small scrub stands to be identified. It is also hypothesised that this method could allow for directional change in scrub levels (i.e., whether scrub is being cleared, or succeeding into woodland) to be assessed through monitoring changes in grassland and woodland levels in addition to scrub.

In this study, we assess whether the method previously outlined in Vafidis et al. (2021) for scrub assessment, can be used to measure temporal changes in scrub, including changes in specific scrub stands, as well as identify directional change in scrub through the assessment of grassland and woodland at two UK nature reserves.

2 Materials and Methods

3 Results

3.1 Daneway Banks

All 30 points manually measured and confirmed as scrub and ground vegetation during the ground verification carried out during the 2017 and 2021 data collection periods were found to match the height bands assigned to them during the orthomosaic creation, confirming that no erroneous classification had taken place and that any consistent inaccuracies within the data overestimated vegetation height by no more than 1.6 cm, and underestimated vegetation height by no more than 4 cm.

The total area of the Daneway Banks Nature Reserve is 16.927 ha (Natural England 2012) or 169,270 m². For 2015, the on-site scrub was comprised of 1533 separate stands, with a size range between 0.03 and 2170.35 m². The total area of scrub on the site was 24,763.7 m², comprising 14.63% of the total site (Figure 1).

In 2017, there was a total of 2555 separate stands of scrub, with a size range between 0.01 and 2288.92 m². The total area of scrub on-site was 27,968 m², making up 16.52% of the total site, a difference of 1.89% and an increase of 12.92% from 2015.

In 2021, there were 2420 separate scrub stands, with a size range between 0.01 and 1556.78 m². The total area of scrub on-site was 25,200.1 m², making up 14.89% of the total size, a difference of 1.63% and a decrease of 9.87% from 2017, and an increase of 1.78% from 2015.

The total area for scrub growth and loss between 2015 and 2021 were as follows:

• Scrub Growth: 9051.04 m² (5.35% of total area).
• Scrub Loss: 15,875.87 m² (9.38% of total area) (Figure 2).

When looking at the direction of change regarding scrub growth between 2015 and 2021, 57.19% (3.06% of total site area) of new scrub in 2021 was grass or bare ground in 2015, and 42.81% (2.29% of total study area) of new scrub was trees in 2015. In terms of scrub loss between 2015 and 2021, 73.73% (6.92% of total study area) of scrub lost between 2015 and 2021 became grass or bare ground, and 26.27% (2.46% of total study area) of scrub lost between 2015 and 2021 became trees (Figure 3).

3.2 Flatholm Island

When determining the study area for Flatholm Island, cliff, and beach habitats, as well as buildings and other functional spaces, were excluded from the study area. Additionally, the northern tip of the island was excluded from the study area, as the UAV flights did not cover it, and the northern part of the island is already heavily managed and contains almost no scrub. The final study area was 167,156.45 m², as opposed to the total terrestrial area of Flatholm, which is approximately 350,000 m².

All 30 points manually measured and confirmed as scrub and ground vegetation during the ground verification carried out during the 2019 and 2021 data collection periods were found to match the height bands assigned to them during the orthomosaic creation, confirming that no erroneous classification had taken place and that any consistent inaccuracies within the data overestimated vegetation height by no more than 0.9 cm, and underestimated vegetation height by no more than 1.5 cm.

In the 2019 dataset, the on-site scrub was comprised of 1116 separate stands, with a size range between 0.007 and 2073.34 m². The total area of scrub on the site was 17,018.2 m², comprising 10.18% of the total site (Figure 4).

In 2021, there were 1400 separate scrub stands, with a size range between 0.006 and 752.89 m². The total area of scrub on-site was 14,560.2 m², making up 8.71% of the total size, a difference of 1.47% and a decrease of 14.44% from 2019.

The total area for scrub growth and loss between 2019 and 2021 were as follows:

• Scrub Growth: 6955.53 m² (4.16% of total area).
• Scrub Loss: 16,935.77 m² (10.13% of total area) (Figure 5).

When looking at the direction of change regarding scrub growth between 2019 and 2021, 88.89% (3.7% of total site area) of new scrub in 2021 was grass or bare ground in 2019, and 11.11% (0.46% of total study area) of new scrub was trees in 2019. In terms of scrub loss between 2019 and 2021, 96.42% (9.77% of total study area) of scrub lost between 2019 and 2021 became grass or bare ground, and 3.58% (0.36% of total study area) of scrub lost between 2019 and 2021 became trees (Figure 6).

4 Discussion

This study shows how height maps generated from UAV-derived aerial images using the method established in Vafidis et al. (2021) can be used to measure temporal and directional change in scrub levels in an accurate, quantified way.

While photogrammetry-based methods have been used to assess and quantify vegetation structure (Alonzo et al. 2020; Brüllhardt et al. 2020; DiGiacomo et al. 2020), most monitoring of temporal change in vegetation height has been carried out using satellite imagery, with a cell size ranging from 30 m to 1 km (Ghafoor et al. 2022; Li et al. 2022; Zhang et al. 2021; Munsi, Areendran, and Joshi 2012), with most drone-based monitoring of vegetation height changes using Lidar (De Almeida et al. 2020; Zhou et al. 2020). Very few studies have measured temporal change in vegetation structure using photogrammetry (Nuijten et al. 2021; Watanabe and Kawahara 2016), despite photogrammetry having equivalent accuracy to LiDAR when used to measure vegetation structure (Filippelli, Lefsky, and Rocca 2019). This means that while LiDAR would still be suitable in situations where penetration of the canopy is necessary, photogrammetry as a tool for measuring temporal and directional change in vegetation height works as a cheaper and more accessible alternative to LiDAR-based methods in situations where canopy penetration is not necessary, while providing more detail than satellite imagery.

While Flatholm has no specific targets in terms of scrub levels, keeping the site a suitable breeding colony for lesser black-backed gulls is a conservation priority, with an aim to reduce scrub across the southern half of the island. The directional change at Flatholm Island showed that the majority of scrub loss (96.42%) was due to scrub clearance. In contrast to this, 26.27% of scrub loss at Daneway Banks was due to succession of the scrub into woodland. While this would still be seen as scrub loss, the priority goals of the site are based around the maintaining of grassland habitat and the prevention of succession of grassland, with overall scrub levels kept below 10% (pers. comm) due to the importance of the grassland habitat for bird and invertebrate species. This means that a decrease in scrub due to woodland succession would work against overall site goals, despite still contributing to the specific scrub cover objectives. Because of this, it may be more suitable to view changes in scrub levels as a shift towards either vegetation succession or reduction, with reduction being the management goal. This allows an increase in scrub levels to be beneficial if it is due to tree clearance, as it progresses the site towards a higher level of open habitat. When scrub changes were quantified as a shift towards either vegetation reduction or succession, both sites showed a slight shift towards reduction, with 58.046% of total changes in scrub cover at Daneway Banks (Figure 7), and 60.2% of changes in scrub cover at Flatholm Island favouring a shift towards vegetation reduction (Figure 8).

This shift towards vegetation reduction and the establishment of more open habitat is desirable for both sites, and could be increased if less new scrub was growing in areas that were previously open grassland, meaning that scrub management targets could be met by preventing the growth of new scrub in grassland areas, as opposed to the clearance of existing scrub.

Scrub levels at Daneway Banks increased from 14.63% to 16.52% from 2015 to 2017, before dropping down to 14.89% from 2017 to 2021. While the decrease in scrub levels from 2017 to 2021 shows that progress towards the management target of no more than 10% scrub, or 16.927 m², is ongoing, scrub levels have not dropped below the 2015 levels of 14.63%, and so any management plans will need to be adjusted, with more surveys carried out to ensure that the decline in scrub levels continued. Up to date maps of scrub cover, as well as maps of scrub cover changes could be used to plan the removal of previously-neglected areas of scrub. For example, the removal of large scrub stands along the northern and southern borders of the site would reduce scrub levels down to 16,749 m², or 9.89% of the site, meeting site management goals, at which point any new scrub stands could be identified and cleared to maintain existing scrub levels (Figure 9).

Flatholm Island scrub levels showed a marked decrease in scrub levels between 2019 and 2021, showing that their current scrub management is sufficient for reducing scrub levels. Assuming that the target scrub levels are similar to Daneway (no more than 10%) then they have achieved this goal and the site management should be changed from decreasing scrub levels to maintaining current levels of scrub. Maps of temporal change in scrub levels could be used to maintain current scrub levels through the identification and removal of new scrub stands as they appear.

However, scrub management goals for both sites are based on ideal levels of scrub estimated using on-foot surveys, which are highly subjective and often inaccurate (Katzner et al. 2011), and a goal of scrub loss may not be contributing to the overall goal of grassland habitat due to scrub succession into woodland. This means that scrub levels of less than 10% at Daneway Banks may not be a suitable target, as it is based on a subjective, ground-based assessment of what 10% scrub would look like. Because of this, it may be more suitable to establish new management targets based on photogrammetric point cloud data rather than basing conservation goals on figures collected via inaccurate means. This could then allow for new management goals to be set and monitored more closely with annual surveys.

Although vegetation height has been shown to be an appropriate structural component for the characterisation of scrub (Vafidis et al. 2021; Cunliffe, Brazier, and Anderson 2016), the lack of any other compositional information on the site vegetation is a limitation of the method, particularly in situations where the identification of particular species is required, such as during the identification and removal of invasive species, or the preservation of endangered species. The Orthomosaic maps generated using this method have been shown to be suitable for species identification (Durgan et al. 2020), although ground-based assessment of species is shown to be more suitable for species with very similar characteristics, or species with specific ground-based characteristics that are relied on for classification (Hardin et al. 2007). Drones equipped with LiDAR or hyperspectral sensors have been shown to be more effective at identifying individual plant species (Sankey et al. 2018), but these are expensive and may be outside of the budget of most conservation organisations. Whether or not species composition and identification is carried out using the generated orthomosaic, ground assessments or using more advanced sensors will therefore depend on the cost, exact species of note, and how easily identifiable they are.

Another potential limitation is that although the ground verification confirmed that no erroneous classification had taken place and that all manually verified scrub and ground vegetation points were accurately classified, an exact comparison between the height values determined via the ground verification and the drone-derived height values could have been carried out to determine the exact error values, as it would help to identify any individual errors or outliers (as opposed to the current method of ground verification which only accounted for consistent errors present across the entire dataset), and should be added to any future methodologies. Ground verification points for woodland points could also be obtained in future studies, assuming suitable equipment and training is available to measure the hight of woodland points safely and accurately.

In conclusion, this study demonstrates how height maps of scrub obtained using UAV-derived imagery can be used to measure temporal and directional changes in scrub levels, as well as identify whether scrub changes are contributing to overall site management goals. This method provides a cheaper and more accessible alternative to LiDAR-based methods that conservation organisations can carry out with a consumer-grade drone, facilitating easier monitoring of vegetation structure and scrub levels, especially when combined with more thorough, pre-existing ground verification methods.

Author Contributions

Matthew Jordan: conceptualization (equal), data curation (equal), formal analysis (lead), investigation (lead), methodology (lead), writing – original draft (lead), writing – review and editing (lead). Jim Vafidis: conceptualization (equal), funding acquisition (lead), project administration (lead), supervision (lead). Mark Steer: data curation (equal), project administration (equal), supervision (equal), validation (lead). Kathy Fawcett: project administration (equal), supervision (equal). Kathy Meakin: project administration (equal), resources (equal), supervision (equal). Gareth Parry: resources (equal). Matthew Brown: resources (equal).

Conflicts of Interest

The authors declare no conflicts of interest.