A model of the extent and distribution of woody linear features in rural Great Britain

Abstract Hedges and lines of trees (woody linear features) are important boundaries that connect and enclose habitats, buffer the effects of land management, and enhance biodiversity in increasingly impoverished landscapes. Despite their acknowledged importance in the wider countryside, they are usually not considered in models of landscape function due to their linear nature and the difficulties of acquiring relevant data about their character, extent, and location. We present a model which uses national datasets to describe the distribution of woody linear features along boundaries in Great Britain. The method can be applied for other boundary types and in other locations around the world across a range of spatial scales where different types of linear feature can be separated using characteristics such as height or width. Satellite‐derived Land Cover Map 2007 (LCM2007) provided the spatial framework for locating linear features and was used to screen out areas unsuitable for their occurrence, that is, offshore, urban, and forest areas. Similarly, Ordnance Survey Land‐Form PANORAMA®, a digital terrain model, was used to screen out where they do not occur. The presence of woody linear features on boundaries was modelled using attributes from a canopy height dataset obtained by subtracting a digital terrain map (DTM) from a digital surface model (DSM). The performance of the model was evaluated against existing woody linear feature data in Countryside Survey across a range of scales. The results indicate that, despite some underestimation, this simple approach may provide valuable information on the extents and locations of woody linear features in the countryside at both local and national scales.

are regularly cut and laid, they can be defined as "managed hedges" (hereafter referred to as hedges) and are particularly widespread and ecologically important landscape features in farmed habitats (Baudry, Bunce, & Burel, 2000). Hedges were originally used to define or enclose fields making them stock-proof, and standards or lines of trees within them were important to demarcate ownership boundaries. More recently, with the availability of relatively lowcost and low-maintenance fencing, land owners are putting much less effort into establishing and maintaining hedges (Antoine, 2001).
However, a recent review investigating the potential importance of hedges to a range of ecosystem services (ES) at landscape scales indicated that they are not merely artifacts of previous management systems but may play a vital role in delivering services (Wolton, Pollard, Goodwin, & Norton, 2014) even in quite unexpected ways.
For example, they have been shown to reduce the incidence of bovine tuberculosis in British cattle herds in high-prevalence regions (Winkler & Mathews, 2015).
The multiple roles which hedges play in the supply of ES include (1) provision: food (sloes, berries, fungi, etc.) and firewood (Wolton, Pollard, et al., 2014); (2) regulation: modification of the microclimate in and around field systems, reduction of soil erosion by wind (Sanchez, Lassaletta, McCollin, & Bunce, 2010), carbon capture and storage in growing woody material and in litter (e.g., extensive linear networks, such as the bocage networks in France, contain considerable sequestered carbon (Robertson, Marshall, Slingsby, & Newman, 2012), restriction of the movement of agricultural livestock, and retention of water and sediment through their role as barriers to soil erosion and in the absorption and storage of water (Gascuel-Odoux et al., 2011;Jongman & Bunce, 2009;Thomas, Ghazavi, Merot, & Granier, 2012;Van der Zanden, Verburg, & Mücher, 2013). In addition, certain species are also associated with key regulatory functions (see below); (3) supporting: soil creation, water and nutrient cycling, and species distribution networks ); (4) cultural: esthetics-hedges are included in definitions of English National Character Areas (Natural England 2014); (5) recreation-hedges support game species such as pheasant and attract wildlife, birds in particular, for enthusiasts to watch and enjoy (Hinsley & Bellamy, 2000); and (6) ownershipmarking boundaries between different groups and owners. Hedges are recognized as being particularly important for biodiversity, and their value as semi-natural habitats spanning increasingly ecologically impoverished agricultural landscapes is widely recognized (see Dainese, Montecchiari, Sitzia, Sigura, &Marini, 2016 andMorelli, 2013). Both the herbaceous flora which grows under and beside the woody shrubs (Roy & de Blois, 2008;Smart, Bunce, Firbank, & Coward, 2002) and the woody vegetation which forms the hedge provide important species and structural heterogeneity as well as providing connectivity between semi-natural habitat components (Batary, Kovacs-Hostyanszki, Fischer, Tscharntke, & Holzschuh, 2012;Roy & de Blois, 2008;Russ, Briffa, & Montgomery, 2003;Staley et al., 2012). By providing a refuge for a wide range of taxa effectively eliminated from fields as a result of agricultural improvement (Smart et al., 2006), woody linear features help to maintain functioning agro-ecosystems in which predators of crop pests, pollinators, and pollen-producing species all play their roles (Pocock, Evans, & Memmott, 2012;Barr and Petit, 2001;Baudry et al., 2000).
Despite the role that hedges may play in the delivery of services in the wider countryside, work investigating ES delivery at landscape scales (e.g., Burkhard et al. 2014) tends to ignore the contribution of hedgerows (and other linear features). Although boundary and linear features are defined as a Broad Habitat (part of a framework classification for 37 habitat types across the whole of the UK by JNCC, see Jackson, 2000), most researchers focus on the areal features within a landscape rather than on their borders and perimeters; consequently, there is a lack of spatial data detailing the types and locations of linear features across broad spatial scales.
The effective management of our natural resources for the future is dependent upon data describing its extent and condition (MEA, 2005). It can be monitored at any number of scales, but to understand resource management at a national level, it is important to have access to national data such as those used in the National Ecosystem Assessment (UK National Ecosystem Assessment 2011). Attempts to quantify the extent of boundary linear features at national scales are rare. One method, used in the Countryside Survey (CS), is stratified random sampling which used field survey to provide national statistics of the extent of the different linear features. CS used detailed field mapping of the extent and condition of linear features in nationally representative sample of 1-km squares Petit, Stuart, Gillespie, & Barr, 2003). Repeat surveys of the same squares make it possible to understand patterns of change in length and condition of hedges and lines of trees with recent results indicating declines in managed hedgerows as they decay into lines of trees . While estimates based on the same approach over time provide useful indices of change for policy makers and essential information for reporting, they do not provide valuable location-specific information, except for in the actual squares in which CS takes place (those data remain confidential).
Field mapping of hedgerows at a national scale would be both expensive and time-consuming; a potential alternative is the use of remote-sensed or satellite data (Kerr & Ostrovsky, 2003). However, the spatial resolution of large-scale remote-sensed data makes an as-  (Jackson, 2000). Previous attempts to map hedges from satellite imagery have led to generalized maps, for example, the French national hedgerow density maps, as developed by the L'Inventaire Forestier National (IFN) or more detailed regional hedgerow mapping (Vannier & Hubert-Moy, 2008). Standard aerial photography and LiDAR (Light Detection And Ranging) offer better solutions (Black, Green, Mullooley, & Poveda, 2010), but for any region of moderate size are currently made difficult owing to the expense of data capture and magnitude of material to be stored and analyzed.

| 8895
SCHOLEFIELD Et aL.  (Smith & Fuller, 2001) to field boundaries. Great Britain (GB) is used as a case study to produce a predictive model of both woody linear features and other linear features which is then validated against existing Countryside Survey data at 1-km square, land class (Bunce, Barr, Gillespie, & Howard, 1996) and national scales, although this approach could easily be applied elsewhere provided a linear network and a canopy height dataset are available. The model uses two key national datasets: (1) the LCM2007 spatial framework-based on that of the Ordnance Survey MasterMap (OSMM) topography layer which provides robust polygon boundaries for GB; and (2) the NEXTMap® Britain DSM series (hereafter referred to as NEXTMap), which provides digital terrain mapping for the UK land surface, indicating the height of features and land parcels above ground height. NEXTMap data are at relatively coarse resolution (5 m), but coverage for GB is comprehensive. and for the total length of woody linear features.
The datasets and modeling approaches are described in detail below. The model was validated against CS data at different spatial scales to provide information on its performance, as described.

| Data inputs and feature attribution
A linear spatial framework formed the basis of the model. The Ordnance Survey Mastermap (OSMM) topography layer provides a detailed cartographic view of the landscape including individual buildings, point features, transport infrastructure, field boundaries, and areas of land. OSMM polygon objects (100 million) were used to create the spatial framework for LCM2007. As the spatial resolution of OSMM is greater than that used for LCM (which uses 20 m × 20 m pixel satellite data), the OSMM was spatially generalized, removing unnecessary detail while retaining relevant information on the location of boundaries (Morton et al., 2011). These data were then converted from a polygon format to a vector format, and the vectors split at intersections in order to yield a linear framework suitable for individual feature attribution from raster datasets (e.g., NEXTMap).
Surface relief information was obtained from the NEXTMap dataset, which was chosen as it has a comprehensive coverage of GB.
NEXTMap includes both a DTM and a DSM, which were originally produced by Intermap Technologies in 2007, the same year as the Countryside Survey 2007. Data were generated by airborne survey using synthetic aperture radar (Carey et al., 2008) (SAR), and singlepass interferometry (IfSAR; Chiverrell, Thomas, & Foster, 2008).
NEXTMap digital elevation data were collected at a flight height of approximately 6,500 m; the data were supplied at a 5-m resolution.
A spatial mask was necessary in order to filter or areas consid- Littoral and sublittoral zones and land above 300-m altitude were considered unlikely locations for woody linear features; hedgerows have not been recorded in CS locations with these characteristics. The resulting 5-m resolution dataset was used in the model.

| Model evaluation
The model was evaluated by comparing the model results at three scales, National (GB), GB land class and at the 1-km resolution-these data were used as a "truth" to test the effectiveness of the model in Bunce, Barr, Gillespie, et al., 1996) (Cohen, 1960) was computed against the validation data for the 5-m point interval classification which compares the accuracy of the system to the accuracy of a random system, and it is a general statistic that can be used for classification systems.

| RESULTS
For the large-scale estimates, Figure 1 shows the model predictions for the density of woody linear features per km square in GB. The results indicate high densities toward the south of GB and much lower densities in the north. Table 1 Table 2 shows that the majority of features were correctly predicted.
The spatial accuracy of the classification is indicated in Figure 4 which shows Cohen's Kappa statistics for the point-based comparisons between modelled and field-recorded data. Figure 4a shows the results for all squares containing woody linear features, while do not appear to be spatially biased. Figure 5 compares the estimates of woody linear feature density based on CS2007 (5a) with the estimates based on the modelled linear framework (5b).

| DISCUSSION
The output of our study is a unique map describing the locations of, and classifying, individual linear boundary features at a national scale  (Table 1) and spatially consistent with CS results (Figures 2 and 5). However, the estimates are generally on average 40% lower than those generated from the CS sample. Figure 3 indicates that the model errors are more commonly associated with the omission of hedges rather than identification of false hedges.
The method of matching boundaries is not perfect, as the datasets being compared are independently derived and boundaries are often complexes of different features located very close to one another, which may include, for example, two hedges bounding another linear feature (such as a green road), or coincident lines of trees and hedges. Additionally, woody linear features in GB are highly variable dependent on individual hedge management practices, regional cultural norms, engagement with agri-environment schemes etc. and may vary between a short (<1 m) and narrow (<1 m) feature resembling a wall and a wide unmanaged hedgerow between 5 and 10 m wide including standard trees with substantial crowns (Countryside Survey, 2007). The match appears to be better in the southwest of England, and this may be because the hedgerow areas are often earthen banks topped with gorse (Ulex europaeus).
A measure of confidence in the output, expressed as Kappa statistics, can be seen in Figure 4, showing results for both individual CS survey squares and land class means. There was no relationship between hedgerow length and confidence (i.e., the model is not better at predicting hedges where there are a lot of them), and in general, there is "fair" to "moderate" agreement.
To date, CS data have been used for hedgerow assessments to underpin national policy on their management ).
While Figure 2 shows the strong agreement between land class means When available, these data which also use earth observation (EO) data will be compared to model outputs.
The model has a number of potential practical, scientific, and policy uses which are explored further here. These include its potential use as a methodology for administering rural payments relating to linear woody features (above). Although, as many payments commonly relate and Wales. This led to a number of regional field hedgerow mapping exercises being performed in order to take account of the extent and condition of hedgerow habitats. These exercises are resource intensive (even with the use of volunteers) and inconsistent in their coverage at a national scale due to their dependency on buy-in at regional levels (although much effort was made to ensure consistency of recording). A new consistent national dataset adds greatly to existing regional data in providing a better understanding of their role in providing habitats for local biodiversity and connecting up semi-natural habitats.
For potential business use, such as the development of hedges as a wood fuel resource (Wolton, 2014) for bioenergy, a dataset describing the woody linear network will be relevant to the identification of suitable locations for relevant infrastructure such as biomass generators and anaerobic digesters.
This dataset also enables us to increase our scientific understanding of landscapes and how they provide essential ecosystem services.  Norton, 2015). Similarly, it is known that hedges influence a whole range of ecosystem services from disease spread to the provision of clean water or climate mitigation (as detailed in the introduction); this dataset provides the potential for accounting for that influence alongside that of land cover in parcels in models of ecosystem function.
During the construction and testing of the model, a number of different approaches were taken including the use of CS field data to train the model, but the simplest model consisting of a simple query of the height characteristics proved to be both the most effective and robust. Detailed CS data might potentially provide a valuable dataset for understanding the performance of the model and thereby improving it, but this would require a significant amount of time and resource and F I G U R E 5 (a) Woody linear feature density (km/km 2 ) from CS2007 field survey, mapped as land class means; b) modelled linear woody feature network density (km/km 2 ) mapped as ITE land class means may only serve to highlight issues around the resolution and spatial accuracy of the spatial framework and the CS dataset. Options for improving the model in the future using national-scale data include the potential use of land cover information about the land parcels on either side of the boundary, which may be correlated with the linear feature type (this could be verified using CS sample square data). LiDAR data are also a potential data source for improving the product, and has been used in the development of regional models of woody linear features (Bailly, Lagacherie, Millier, Puech, & Kosuth, 2008;Ferraccioli et al., 2014). It has not however yet (to the authors' knowledge) been collected or interpreted in a consistent way across GB.
Another possibility may be to use citizen science to further validate and improve the quality of information, including historical data collected as part of the regional hedgerow mapping exercises described above, although this would need to be carried out using a strategic and consistent framework to ensure the consistency and quality of the data. Potential approaches include placing the data on the internet for users to validate. The approach could be linked to temporal reviews and revisions to provide statistics of hedgerow change. This could, for example, help to target particular management approaches including restoration or recreation of woody linear features under agrienvironment schemes.
Given the lack of any such product currently, the information it provides is valuable and although incomplete, the hedge model is generally accurate. Clearly, for users at local levels, there is a great opportunity (potentially through the development of appropriate software) to supplement the model data with new or more accurate data collected at local scales either through volunteer approaches on the ground or the addition of regional government data (where available).
The value of a national model is consistency of approach and as stated above, any enhancements to the model outputs at local levels should endeavor to retain and build on this consistency.
In conclusion, the information presented here offers great potential to further the management and conservation of hedgerows in GB, improve delivery of ecosystem services, and to improve landscape resilience, and the approach described is an easily translatable model that can be applied in different parts of the world given the availability of appropriate data.