Assessing the use of natural colonization to create new forests within temperate agriculturally dominated landscapes

There is a global drive to increase forest cover to protect biodiversity and help combat climate change. Tree planting is widely used to increase forest cover, although there is growing interest in using natural processes. However, predicting the outcome of natural colonization is challenging as it is a highly variable process and evidence is sparse, especially in temperate landscapes. Our study aims to evaluate the spatial and temporal scale over which natural colonization may be an effective approach to create new forests within temperate agriculturally dominated landscapes. We examine the spatial patterns in tree density and height across 90 sites in the United Kingdom that attempted to use natural colonization to create new forests between 1994 and 2004. Tree density and tree height were measured using light detecting and ranging point cloud data, an increasingly common technique for surveying forest characteristics. This research shows that natural colonization is a highly variable, relatively slow process in temperate agricultural landscapes, spatially restricted to a narrow fringe around existing forests and trees (105 m; 95% CI: 70–174 m, within approximately 19 years), although exact distances vary with former land use. This suggests that in some circumstances natural colonization will need to be assisted and supplemented with tree planting to increase forest cover, especially in areas away from a seed source. In reality, a blend of approaches will be needed to meet policy and land manager objectives, increase forest cover and tackle the biodiversity and climate crises.


Introduction
The world is currently experiencing a dual biodiversity and climate crisis (IPBES 2019), with biodiversity declining at unprecedented levels (Butchart et al. 2010) and the declaration of a Climate Emergency (Ripple et al. 2020).The drivers of these joint crises are multifaceted but a primary cause is the humandriven loss and degradation of habitats globally (Brooks et al. 2002;Díaz et al. 2019).This is especially true for forests, which have suffered global losses of 178 million ha between 1990 and 2020 (FAO & UNEP 2020) often as a result of human activities including logging, agriculture, and urbanization (Hansen et al. 2013;Curtis et al. 2018).Forest loss and degradation on this scale will facilitate biodiversity declines and alter forest communities such that specialist species are replaced by habitat generalists (Naaf & Wulf 2010;Betts et al. 2017).
There is now a global drive to increase forest cover, seen by many as a cost-effective and readily available nature-based solution to protect and restore biodiversity (Newmark et al. 2017;Leclère et al. 2020;Wang et al. 2022) and to help combat climate change (Canadell & Raupach 2008;Zomer et al. 2016;Leclère et al. 2020).For instance, the Bonn Challenge launched in 2012 is a global effort to restore 350 million ha of deforested and degraded land by 2030 (Dave et al. 2019;Seddon et al. 2019;Leclère et al. 2020)."Active" tree planting is used and promoted by many as a primary method to increase forest cover (Bastin et al. 2019;Griscom et al. 2020); others suggest that tree planting should not be seen as a panacea (Seddon et al. 2019;Holl & Brancalion 2020) following large-scale failures, an overly narrow focus on fast-growing plantations, overestimates of benefits and poor targeting (Temperton et al. 2019;Veldman et al. 2019;G omez-Gonz alez et al. 2020).As a consequence, there is growing interest in using natural processes to create and restore forests (Lamb et al. 2005;Crouzeilles et al. 2017Crouzeilles et al. , 2019)), often referred to as "passive restoration" or "rewilding" (Broughton et al. 2021;Chazdon et al. 2021).
The process of natural colonization, which allows trees to establish on areas of previously non-forest land by their own means, is receiving increased attention as a more "passive," ecologically sound approach to create new forests and tackle both the biodiversity and climate crises (Chazdon et al. 2021;Di Sacco et al. 2021).This argument assumes that forests created by natural colonization will be cheaper to establish (with lower reliance on planting stock), more effective (in terms of success, suitability, species, and structural diversity), and carry fewer risks (such as tree disease and pests) (Di Sacco et al. 2021). However, Reid et al. (2018) identified a degree of selection bias in the comparison of natural processes (colonization and regeneration) with tree planting.They suggested that a level of success was almost guaranteed for studies of natural processes, as these were generally conducted on former forest sites with high levels of natural regeneration from previously established trees.By contrast, studies of tree planting were often focused on more challenging and degraded non-forest sites whose colonization and establishment would likely be highly limited in the absence of planting (Reid et al. 2018).
Predicting the outcome of natural colonization is challenging because it is a highly variable process (Brancalion et al. 2016;Meli et al. 2017;Crouzeilles et al. 2020).This variation produces some of the benefits of utilizing natural processes, such as species and structural diversity (Di Sacco et al. 2021), but also leads to uncertainty about the benefits that may accrue on a given site and time frame (Brancalion et al. 2016).Outcomes at a site level may be influenced by abiotic factors, such as prior land use (Reid et al. 2018), ongoing management, and the harvesting of forest products (Chazdon 2013).In addition, biotic factors such as the proximity, diversity, and density of seed sources (Molin et al. 2018;Crouzeilles et al. 2020), grazing pressure (Murphy et al. 2022), and the dispersal mechanisms of local tree species (Grashof-Bokdam & Geertsema 1998;Martínez & García 2017) may also influence site-level variation.Natural colonization also varies greatly across biomes.In the Brazilian Amazon, Crouzeilles et al. (2020) found that 90% of passive tree colonization occurred within 192 m of forested areas in a 20-year period.In contrast, tree colonization has been found to be a slow and spatially restricted process in temperate agriculturally dominated landscapes (Harmer et al. 2001).
Specifically, in Britain, Murphy et al. (2022) found that after 12 years oak colonization was largely limited to within 20 m from the nearest seed source.
Woodland creation and restoration in temperate landscapes occur primarily on former agricultural land or after land abandonment (Westaway et al. 2023).However, the evidence base for the spatial and temporal scale of natural colonization is sparse and variable, dominated by studies from the tropics (Reid et al. 2018).As a result, policymakers, practitioners, and land managers view natural colonization as a challenging and less favorable approach to create new forests, particularly when they have specific forest objectives (FitzGerald et al. 2022).Recent natural colonization studies from temperate landscapes are generally small-scale and often based on single former land use types (Broughton et al. 2021;Murphy et al. 2022), although it is well known that former land use and soil types influence ecosystem restoration (Brudvig et al. 2013).This makes it hard to generalize principles to guide and effectively target restoration actions, especially in less studied temperate agricultural landscapes.
Our study aims to evaluate the spatial and temporal scale over which natural colonization may be an effective approach to create new forests within temperate agriculturally dominated landscapes.We examine the spatial pattern in tree density and height using light detecting and ranging (LiDAR) data (van Leeuwen & Nieuwenhuis 2010; Zhen et al. 2016), across 90 sites in the United Kingdom, which attempted to use natural colonization to create new forests.These sites were identified through an objective site selection process from a broader suite of sites that received funding for woodland creation under the English Woodland Grant Scheme (Forestry Commission 2004).This site selection process attempts to avoid survivorship bias to learn from both success and failure.The process of natural colonization was initiated on these sites between 1994 and 2004, which represents both an ecologically realistic and policy-relevant timescale to evaluate tree colonization (Forestry Commission 2022).

Site Selection
The aim of the site selection process was to identify a network of sites that had nationwide coverage, were a similar age, had similar starting conditions, and had limited survivorship bias.To achieve this, we selected suitable sites from those that received funding for forest creation under the English Woodland Grant Scheme 3 (WGS3) (Forestry Commission 2004).The WGS3 ran from 1994 to 2004 and included an option for "New Natural Regen" (n = 247), which supported both natural regeneration within existing forests and natural colonization of non-forested open habitats.
Historical aerial imagery on Google Earth Pro (2022) was used to identify which of the "New Natural Regen" sites had areas of natural colonization on open, previously unforested land.Historical images were taken within AE8 years of the start of natural colonization.Sites were excluded if they had been returned to agriculture, had evidence of planting, were previously forested, such as clear fell, or were managed as parkland.The former land use was then identified from the Land Cover Map 1990 (UKCEH 1990) and sites were only retained if they occurred within one of four land use types: acid grassland, arable, heathland (heather and heather grassland), or improved grassland.Finally, sites that did not have full LiDAR point cloud data in the Environment Agencies National LiDAR Programme (Environment Agency 2022) were excluded.
Our final site selection comprised 90 sites: 21 acid grassland sites, 29 arable sites, 14 heathland sites, and 26 improved grassland sites (Fig. 1).The mean age of selected sites was 19 years (range 14-26 years) as detailed in Table 1.Site ages were ascertained by calculating the time difference in years between the start of the grant scheme and the date on which LiDAR surveys were conducted on a given site (Table S1).

Data Processing and Outputs
LiDAR uses airborne sensors to emit laser pulses and measure the return time of reflected signals, to quantify the distance of a reflecting surface.Discrete returns from many laser pulses generate a "point cloud" that illustrates the three-dimensional structure of all reflecting surfaces (Beland et al. 2019).From these point clouds, it is then possible to obtain measures such as tree location, maximum height, crown diameter, and vegetation structure (Holmgren & Persson 2004;Guo et al. 2021), as illustrated in Figure 2. LiDAR point cloud data for each site and a 500 m buffer surrounding it was obtained from the Environment Agency National LiDAR Programme (Environment Agency 2022).The collection always occurs in the winter and there are a maximum of four discrete returns per pulse.Canopy height models for each site were then created using a pit-free algorithm with the "lidR" v4.0.1 package (Roussel et al. 2020).
Canopy height models were cross-referenced with historical aerial images to identify trees that were present within site boundaries and the 500 m buffer surrounding each site, before the start of natural colonization.All trees, hedgerows, and forests that existed before the natural colonization within the site boundaries and within 500 m of the site were identified and polygonized.The site-level polygons were then refined to exclude any existing trees.Clipping the point cloud data to these adjusted polygons allowed existing trees to be removed from the dataset, preventing their detection as colonizing trees, which has the potential to exaggerate the extent of natural colonization.Furthermore, mapping mature trees that existed within and surrounding sites as source trees results in a more accurate estimation of the effect of distance to seed source on the extent of natural colonization (Guy, Forster, and Watts, unpublished data).
Point clouds were then normalized via interpolation using the knnidw() function, to account for changes in terrain elevation when estimating the height of trees from point cloud data (Roussel et al. 2020).The location and size of individual trees across the sites were identified using Li et al. (2012), tree segmentation algorithm within the "lidR" package (Roussel et al. 2020) using the default parameters.The algorithm used here only identifies trees over 2 m and with a crown diameter greater than of 1.5 m, to reduce the possibility of low-lying vegetation being mislabelled as trees.Previous validation of the algorithm indicated that 86% of trees were successfully detected and that 94% of segmented trees were correct (Li et al. 2012).These data were then summarized in a tree density and tree height raster with 10 m Â 10 m pixel resolution (each pixel contained data on tree density and mean tree height).
The distance from each pixel to the nearest potential seed source was then calculated (using polygons created in the process described above).The process of site selection and data processing to produce the three required data outputs (distance to seed source, tree density, and tree height) for further analysis is illustrated in Figure 3.
To analyze the impact of site age on the density and height of colonizing trees, we calculated the number of years between the start of the grant scheme and the year in which LiDAR surveys were conducted.It is also important to note that site age data were only available for 82 sites and these ages are approximate, as natural colonization may have begun on some sites before their inclusion in the WGS3 grant scheme.

Statistical Analysis
Associations between tree density and distance from seed source were analyzed with negative binomial generalized linear mixed  models (GLMMs), fitted with natural splines.Before constructing models, the heather and heather grassland levels for former land use were combined into a single heathland level, due to small sample sizes in the original groups.GLMMs were constructed with tree density as the response variable, distance from seed source, former land use, and site age as fixed effects and site as a random effect.Natural splines with three degrees of freedom were then added to the distance from the seed source variable to account for the non-linear association between tree density and distance from the seed source.Candidate models were constructed using additive or interacting fixed effects, alongside either site-level random intercepts or site-level random slopes for the effect of distance from the seed source.Akaike information criterion (AIC) scores were used to guide model selection (Burnham et al. 2011).Tree density estimates from the model are interpreted in the context of the English Woodland Creation Offer, which defines a success threshold for natural colonization as 100 trees/ha over 60% of a site after 10 years and compared to an initial minimum stocking density of 1,100 trees/ha for planted woodland (Forestry Commission 2022).
To analyze associations between tree height and distance from the seed source, the same model structure was used, this time assuming a Gaussian distribution with a log link function and without natural splines as data visualization suggested splines would not improve model fit.

Tree Density
The best candidate model predicted a non-linear relationship between tree density and distance from the seed source, tree ).In the Site selection phase, (5) natural colonization was identified at WGS3 sites using historical aerial imagery and (6) the former land use was confirmed using LCM 1990.During Data processing (7) all seed sources present before natural colonization were mapped within 500 m of the site using LiDAR data and historical aerial imagery.In addition, (8) all trees within the site before natural colonization started were removed and tree segmentation identified the location of individual trees.These two data sets were then used to produce three Data outputs for analysis, 10 m Â 10 m pixels covering the entire extent of the site, each with associated (9) distance to the nearest seed source, (10) tree density, and (11) mean tree height.
density increased to a peak around 20 m from the seed source before declining (Figs. 4 & S1).The maximum tree density varied between land uses (Table 2; Fig. 4).Presented alphabetically: in former acid grassland sites tree density peaked at 167 trees/ha (95% CI: 124-225); in former arable sites density peaked at 384 trees/ha (95% CI: 296-497); in former heathland sites the density peaked at 176 trees/ha (95% CI: 122-254); and, in former improved grassland sites the density peaked at 407 trees/ha (95% CI: 310-535).The model indicates that after approximately 19 years (range 14-26 years) a tree density of 100 trees/ha would occur at a spatial scale of around 70 m (95% CI: 45-115 m) from the nearest seed source in former acid grassland sites, 135 m (94-upper CI exceeds the maximum data range of 305 m) from a seed source in former arable sites, 73 m (45-127 m) in former heathland sites and 140 m (97-upper CI exceeds the maximum data range of 149 m) on former improved grassland sites (Table 2).There was substantial between-site variation in overall tree densities, although the spatial pattern was consistent.
The best model explaining spatial variation in tree density did not include an interaction term between former land use and distance from seed source or site age (model F, Table S2).These models had lower AIC scores when compared to the additive models (Table S2), but were rejected as they predicted unrealistically high CIs at further distances away from source trees, likely due to a lack of samples at these distances.In addition, as site age was not included the best model is fitted using all 90 sites.

Tree Height
The best model to explain variation in tree height with the distance from the seed source predicted a negative linear relationship in all former land use types, with the tallest trees adjacent to the nearest seed sources (i.e.0 m) (Figs. 5 & S2).Maximum tree height was predicted to be 5.4 m (95% CI: 4.57-6.34m) on acid grassland sites, 6.49 m (5.44-7.73m) on arable sites, 5.31 m (4.41-6.40m) on heathland sites, and 7.54 m (6.12-9.28m) on improved grassland sites (Table 2).Site age had a modest but statistically significant positive linear effect on tree height (1.03, 95% CI: 1.00-1.06,z = 2.46, p = 0.014) (Fig. S3).
The best model (lowest AIC score; model 5, Table S4) explaining the relationship between tree height and distance from seed source did not include an interaction term between former land use and distance from seed source and included an effect of site age (Table S5).As age is included, the model is fitted using 82 sites (those that included age data; Table S1).The inclusion of fewer sites did not change the qualitative conclusions that were drawn from the model.

Discussion
This work has given us useful insight into the spatial and temporal scales of natural colonization across temperate agricultural landscapes.Our findings concur with previous studies that natural colonization is a slow process, gradually spreading from a seed source over time and largely spatially restricted to a narrow fringe around existing source forests and trees.We found that after approximately 19 years the maximum distance that successful natural colonization (based on a threshold of 100 trees/ha) occurred ranged between an average of 70 and 140 m and the maximum tree density occurs at around 20 m from a seed source, and then declines with increasing distance and tree height declines with increasing distance.The mechanisms for the patterns observed are likely to arise from a complex array of biotic and abiotic interactions influencing seed production, dispersal, establishment, and subsequent development.We found that natural colonization is a slow process, which is in line with other studies (Harmer et al. 2001;Broughton et al. 2021;Murphy et al. 2022).Colonization of trees requires a seed source and a seed to germinate and develop into a tree.Good seed production years in some temperate tree species can be irregular and infrequent (e.g.masting species seed every 3-15 years), which will have an effect on the rate of seed fall (Harmer 1999;Zwolak et al. 2016).Germination and development in trees is slow but can be further hindered through interactions with other species.Seed predation by rodents, beetles, and slugs can be high in ex-agricultural land (Bruun et al. 2010); competitive pressure from ruderal plant species and rank grasses, often quick successful early colonizers of ex-agricultural land, can be strong, preventing germination, and/or development (Alard et al. 2005;Hudjetz et al. 2014); and herbivory pressure from deer and livestock can also inhibit development (Kuiters & Slim 2003;Hudjetz et al. 2014;Murphy et al. 2022), especially in open habitats areas adjacent to forests (Halls & Alcaniz 1968;Thirgood & Staines 1989).
Trees density increases with distance from the seed source to a maximum of approximately 20 m, before declining to below 100 trees/ha at between 70 and 140 m from the seed source (Forestry Commission 2022).In all land use types this maximum was lower than the initial stocking density of 1,100 trees/ha for planted woodlands (Forestry Commission 2022), although tree density in planted sites of equivalent age would be expected to be lower than this due to management and natural processes.In addition, trees adjacent to the seed source were taller and height declined with increasing distance.Tree colonization of open habitats predominantly occurs through gravity or wind-dispersed seeds within 50-100 m of a seed source (Harmer 1999; Murphy et al. 2022).Bird and mammalian vectors can facilitate the wider dispersal of tree seeds (Worrell 2014).However, long-distance dispersal is likely to lead to sporadic, low-density seed distribution and some bird and mammal species movement can be restricted by habitat suitability and resource availability (Zwolak et al. 2016;Martínez & García 2017).The higher volume of seed adjacent to the seed source means a greater number of individuals to potentially escape seed predation and herbivory pressures.However, our results indicate that within 20 m of the seed source competition between these trees as well as the adjacent mature woodland for resources is resulting in a degree of natural self-thinning (Das et al. 2011).At distances greater than 20 m from the seed source, the tree density may be more linked to seed dispersal, which declines with distance  Natural colonization in temperate landscapes (Nakagoshi & Wada 1990;Letcher & Chazdon 2009;Broughton et al. 2021).Colonized trees are taller next to the seed source than further into the open habitat as they potentially arrived earlier and pre-existing mature trees (seed source) are likely having a commensal effect, protecting developing trees from wind, reducing water stress, and reducing competition from ruderal species (Meng et al. 2006;Craine & Dybzinski 2013;Hughes et al. 2023).We did not identify an effect on the pattern of tree distribution between former land uses, but each had markedly different baselines of tree density and height.Former arable and improved grassland sites supported a higher maximum density of trees and taller trees than acid grassland and heathland sites.In this study, acid grassland and heathland sites tended to be at higher elevations in the northern extent of England, around the Pennines, Lake District, North York Moors, and Northumberland.These upland environments are often more exposed and on steeper terrain, potentially offering less favorable conditions for tree establishment and growth due to increased wind buffeting (Miller et al. 1987;Messaoud & Chen 2011;Hughes et al. 2023).Former land use has long-term impacts on forest biodiversity (Dupouey et al. 2002), and early successional structural differences in forests can last for decades (Jakovac et al. 2021).Thus, former land use alongside topographic factors is important to consider when implementing natural colonization as it is with any forest creation strategy.
We identified that natural colonization was highly variable at a site level; the best models for both tree height and tree density included a term for site-level random effects.The sites used in this study span the whole of England and will experience differences in local climate, impacts of pests and disease in the mature woodland, site-level nutrient availability and management as well as local pollinator communities, herbivore populations, and seed predator abundance.All of these factors in turn will affect the seed production, dispersal, establishment, and subsequent development of trees at individual sites.The large sitelevel variation means that it is advisable to be cautious when interpreting results for policy.
There are several caveats that need to be considered when interpreting the findings of this research.Although all these natural colonization sites were started by individual land managers between 1994 and 2004 as part of the same grant scheme, we know little about their specific objectives, motivations, or rationale for entering the scheme.Some managers may have been focused on reducing costs to create a productive timber forest while others may have been trying to create a more variable and open forest ecotone over a longer timeframe.These manager objectives along with subsequent interventions, such as initial ground preparation, fencing, and supplementary tree planting, of which we know little about, have obviously affected the trajectory of these sites through time and may explain some of the observed variations.
Although LiDAR has been used widely to examine forest structures (van Leeuwen & Nieuwenhuis 2010; Beland et al. 2019) we do not know how well it may capture tree density and height within these young and complex natural colonization sites, especially those with small trees (less than the current 2 m threshold) at the leading colonization edge.The minimum 2 m threshold is set as the default within the tree segmentation algorithm we used, because overlap among small trees may lead to an overestimation of the density of smaller trees (Roussel et al. 2020).Furthermore, Forest Research defines a tree outside of woodland as a woody perennial species over 2 m, which is self-supporting and produces a crown (Forest Research 2022).As a result, using this current approach we may be underestimating the density and spatial scale of colonization across these studies sites.In addition, there is no information on the actual tree species that have colonized these sites from source forests and trees and the benefits they may be providing in terms of biomass, carbon storage, or biodiversity value.Some sites may be dominated by a few early successional species or species which are able to compete in a fertile agricultural environment or survive high herbivore pressure, while other sites may consist of a diverse mix of species in more favorable and less challenging sites.Fieldwork is currently underway by the authors and others to investigate some of these natural and social science questions further, specifically to address the potential underestimation of tree density and the process of species colonization (e.g.dispersal processes) and establishment (e.g.site fertility and herbivory).
However, we acknowledge that a blend of approaches will be needed to meet a range of policy and land manager objectives, increase forest cover and tackle the biodiversity and climate crises (Brancalion & Holl 2020;Di Sacco et al. 2021).Areas of natural colonization may develop slower, have lower tree densities, biomass and stored carbon, than a similarly aged plantation, but they may offer considerable benefits such as reduced costs, reduced soil disturbance with accompanying carbon losses, fewer risks from trees disease and increased biodiversity value through the creation of a complex mosaic of trees, scrub, and grassland, which may be better aligned with more biodiversity-focused objectives.There is also some evidence that given enough time these sites within close proximity to the source forest may develop through natural succession into diverse and complex forests (Broughton et al. 2021).In general, natural colonization is most suitable on lightly degraded sites near existing source forests and trees (Di Sacco et al. 2021).In contrast, tree planting may be more appropriate where you aim to create a more predictable forest in terms of species, density, and speed of development and may be more aligned with productive and carbon-focused objectives, or for sites that are heavily degraded and more isolated from existing source forests or trees (Di Sacco et al. 2021).It has recently been shown that degraded and isolated sites in temperate agricultural landscapes may be very slow to develop, with very few trees colonizing after 33 years (Broughton et al. 2022).
Findings from this research can be used to refine the spatial targeting of grant schemes for natural colonization in temperate agricultural landscapes.In combination with future research and accounting for the caveats and limitations, this can also help land managers understand the risks and benefits of natural colonization and increase uptake where appropriate.
Natural colonization and tree planting should not be considered as binary or mutually exclusive approaches to increase forest cover.There is increasing evidence to suggest that a degree of spatially targeted tree planting, such as establishing small patches of trees, that is "woodland islets" and "applied nucleation" (Benayas et al. 2008), can actually assist rather than hinder natural processes.For example by enhancing seed dispersal, providing habitats and future seed sources, and ameliorating the local environment (Holl et al. 2020;Kulikowski et al. 2023).

Figure 1 .
Figure 1.Map displaying the spatial distribution of sites (n = 90) in the United Kingdom on which natural colonization was analyzed.Points are colored according to former land use, meaning the land use before the start of natural colonization.

Figure 2 .
Figure 2. On the left is an example LiDAR point cloud showing (from left to right) the transition from a mature source woodland into a naturally colonized site.On the right is the corresponding aerial view of the example site, with the blue box showing where the example LiDAR point cloud were sampled.

Figure 3 .
Figure 3. Data processing procedure.Data inputs were (1) open access WGS3 new natural regeneration grant data (Forestry Commission 2004), (2) Land Cover Map 1990 (UKCEH 1990), (3) Historical aerial imagery (2022), and (4) LiDAR point cloud data from the National LiDAR Programme (Environment Agency 2022).In the Site selection phase, (5) natural colonization was identified at WGS3 sites using historical aerial imagery and (6) the former land use was confirmed using LCM 1990.During Data processing (7) all seed sources present before natural colonization were mapped within 500 m of the site using LiDAR data and historical aerial imagery.In addition, (8) all trees within the site before natural colonization started were removed and tree segmentation identified the location of individual trees.These two data sets were then used to produce three Data outputs for analysis, 10 m Â 10 m pixels covering the entire extent of the site, each with associated (9) distance to the nearest seed source, (10) tree density, and (11) mean tree height.

Figure 4 .
Figure 4.The association of tree density with distance from seed source and former land use, across 90 sites in England.Solid colored lines illustrate mean model predictions for each land use type and ribbons show the 95% CI around the means.Dashed colored lines illustrate predicted site-level associations.Black dotted lines indicate the 100 trees/ha success threshold for natural colonization and the minimum density of 1,100 trees/ha for planted woodlands, as defined by the England Woodland Creation Offer (Forestry Commission 2022).Model predictions are displayed over 200 m, as most data fell below this cut-off and within 200 m of source trees is a policy-relevant spatial scale.A plot displaying predictions over the full distance range for each land use is available in Figure S1.The number of sites in each group is acid grassland (n = 21), arable (n = 29), heathland (n = 14), and improved grassland (n = 26).

Figure 5 .
Figure 5.The association of tree height with distance from seed source and former land use, across 90 sites in England.Solid colored lines illustrate mean model predictions for each land use type and ribbons show the 95% CI around the means.Dashed colored lines illustrate predicted site-level associations.Model predictions are displayed over 200 m, as most data fell below this cut-off and within 200 m of source trees is a policy-relevant spatial scale.A plot displaying predictions over the full distance range for each land use is available in Figure S2.The number of sites in each group is acid grassland (n = 21), arable (n = 29), heathland (n = 14), and improved grassland (n = 26).

Table 1 .
Number and mean age of natural colonization sites across the former land use types based on the date the scheme was started within WGS3 (Forestry Commission 2004) and the date of the LiDAR data (Environment Agency 2022) as detailed in TableS1.Pixel distance describes the distance between source trees and 10 Â 10 m pixels for which tree density and height were calculated, as an indicator of site size.
Mean age AE SE (range) (units = years) Mean pixel distance AE SE (range) (units = m)

Table 2 .
Summary statistics.Values are predictions based on model F, regarding tree density and model 5, regarding tree height.