Amazonian secondary forests are greatly reducing fragmentation and edge exposure in old-growth forests

The position of secondary forests within the wider landscape plays a critical role in determining their growth rates and biodiversity (figure 1). Proximity to old-growth forest is advantageous throughout succession, supporting faster forest growth and greater species diversity in the early stages (Jakovac et al 2015, Toledo et al 2020), as well as providing the diverse seed sources required later (Hawes et al 2020). High surrounding forest cover has positive effects on biomass recovery (Martínez‐Ramos et al 2016, Toledo et al 2020), meaning that secondary forests growing in relatively intact forest landscapes are likely to have higher carbon accumulation potential than those in highly deforested landscapes (Chazdon 2003, Bihn et al 2010). A key determinant of species richness in secondary forests is whether they are connected to an extensive area of old-growth forest (Mayhew et al 2019)—high landscape connectivity accelerates the recovery of seed dispersal mechanisms, particularly for large mammals (Estrada-Villegas et al 2023). However, secondary forests are influenced not just by the quantity, but also the quality of surrounding old-growth forest cover. Degraded old-growth forests that have had their composition altered by structural disturbance events (e.g. forest fires, selective logging) are unsuitable as habitat for many forest species (Mestre et al 2013, Barlow et al 2016, Moura et al 2016), while small fragments will not hold many of the large vertebrates that could support the movement of large-seeded plant species into secondary forests (Laurance et al 2002, Lees and Peres 2006). Together, these relationships indicate that secondary forests may benefit considerably from proximity to large areas of structurally intact old-growth, whereas development of secondary forests in landscapes with highly fragmented old-growth forest cover or extensive forest disturbance is likely to be limited by lack of seed sources (Matos et al 2020).

Zoom In Zoom Out Reset image size

Secondary forests could provide additional environmental benefits if they help buffer old-growth forests against edge effects (figure 1). Edge effects are a well-documented phenomenon that drastically impact the structure and functioning of a forest (see review by (Laurance et al 2002). Field studies have demonstrated that the ecological and physical impacts of edge-effects extend hundreds of metres into a forest (Laurance et al 2002), while remote sensing studies have revealed their substantial impact on carbon emissions across large spatial scales (Silva Junior et al 2020a). However, these impacts may be mitigated by secondary forest growth adjacent to old-growth forest edges. Secondary forest buffers may significantly reduce edge-related tree mortality (Mesquita et al 1999), faunal edge avoidance (Stouffer and Bierregaard 1995) and the intensity of changes in microclimate (Didham and Lawton 1999). However, the benefits of secondary forest buffers are likely to be influenced by the temporal relationship between edge exposure and subsequent regeneration. As much as 90% of edge-induced carbon loss occurs within 5 years of exposure (Silva Junior et al 2020a), but it takes time for secondary forests to reach a level of structural development at which they may begin to offer protection from abiotic changes or offer habitat to core-forest species (Laurance et al 2002). Nonetheless, landscape-wide increases in the occupancy and abundance of specialist species have been found following regeneration by secondary forests (Rocha et al 2018).

Secondary forests may also benefit old-growth forests by increasing fragment connectivity (figure 1, Newmark et al 2017). Even small breaks in forest cover can present impassable barriers to some species (Lees and Peres 2009) and fragment isolation has been found to explain more variation in species composition than either forest type or forest age (Mayhew et al 2019). By connecting old-growth forest fragments, secondary forests could mitigate isolation effects by supporting the movement of animals between old-growth forest fragments, even if the secondary forest is not necessarily suitable habitat for those species (Newmark et al 2017).

As the world's largest remaining expanse of tropical forest, increasing our understanding of the dynamics of regeneration in the Amazon is of critical importance to mitigating climate change. While trends in the location of secondary forests relative to old-growth forests have been demonstrated at regional-scale (Schwartz et al 2017a, Smith et al 2020, 2021), to date no analyses have directly addressed this issue at landscape-scale for the entire biome. Here, we conduct the first Amazon-wide assessment of the location of secondary forests in relation to old-growth forests. We explore the positioning of secondary forests within the landscape, asking (1) where are secondary forests located in relation to old-growth forests of different quality? (2) What extent of old-growth forest edges are buffered by secondary forests? And (3) how does secondary forest affect connectivity between old-growth forest fragments? We also investigate the exposure time of old-growth edge forest, as well as the age of secondary forests acting as buffers or connecting fragments, to provide insights into the potential benefits of their proximity. Enhancing our understanding of the spatial and temporal relationships between old-growth edges and secondary forests would help to refine our estimates of edge-related carbon emissions and habitat loss and could aid in the design and prioritisation restoration programmes.

The article is based on chapter 4 of the lead author's doctoral thesis (see Smith 2022).

The Amazon biome spans 6.7 million km² and is the largest remaining expanse of tropical forest on earth. This study focuses on upland terra firme forests (92% of total forest cover), as wetland forests are very different structurally and in their species assemblages. Any pixels classified as wetland by (Hess et al 2015) or (Gumbricht et al 2017) are excluded from our forest classes.

We assessed old-growth and secondary forest cover in the Amazon biome in 2020 using the MapBiomas Amazonia Collection 3.0 dataset (MapBiomas 2021), which provides annual 30 m resolution land cover maps from 1985 to 2020. We reduced the MapBiomas schema to two classes (forest and non-forest) and then applied a change detection algorithm to identify which forest pixels were secondary forests. Following the method of (Smith et al 2021), any pixel (900 m²) in the 'forest' class in the first year of the time series (1985) was considered old-growth forest until it transitioned to 'non-forest'. Pixels that transitioned from 'non-forest' to 'forest' were labelled as secondary forest. Secondary forest age was measured as the number of consecutive years a pixel was labelled as secondary forest. Monoculture tree planting is classified as agricultural land use in the original MapBiomas schema and has been excluded from the forest mask used in this study.

As the MapBiomas time series begins in 1985, any secondary forest that began growing before this date is classified as old-growth forest and the maximum age of secondary forests detectable with this method is 34 years. As such, our method may be overestimating old-growth forest extent and underestimating secondary forest extent. However, we believe the impact of this on our results to be small as secondary forests typically have low residence times and high turnover rates (Schwartz et al 2020, Smith et al 2020), so it is unlikely that much of the pre-1985 secondary forest remains in 2020. Furthermore, as the earliest reliable satellite imagery covering the whole Amazon is from 1985, MapBiomas provides the best available map of historic forest cover.

We created three classes of old-growth forest quality based on structural disturbance (see definition from Bullock et al 2020) and fragment size: (1) any old-growth forest, (2) structurally-intact old-growth forest, and (3) extensive (>50 km²), structurally-intact old-growth forest. These classes are intended primarily as a proxy for the availability of faunal seed dispersers and seed rain diversity, based on evidence showing that disturbance or fragmentation in old-growth forests alters seed-dispersal processes (e.g. Cordeiro and Howe 2003, Liu et al 2019, Hawes et al 2020).

Bullock et al (2020) provide a spatially explicit dataset of forest degradation from 1995 to 2017. They defined degradation (distinct from deforestation) as a natural or anthropogenic disturbance that does not change a pixel's land cover category—including fire, windthrow, selective logging, and damage to standing forests during expansion of roads or other development. We resampled the condensed Bullock et al (2020) dataset to align with MapBiomas, then identified structurally-intact old-growth forest as those pixels that had not experienced a disturbance event in the last 10 years (relative to 2020). We selected a 10 year threshold as forests recover over time (Rutishauser et al 2015, Silva et al 2018) and some species may eventually be able to return if further disturbance events are avoided (Mestre et al 2013, Mollinari et al 2019). While the (Bullock et al 2020) dataset is the most recent map of forest degradation currently available for the Amazon, there are two limitations to its use in this study that may mean we are underestimating the extent of structural disturbance. First, it does not extend to our study year (2020) so we cannot account for the three most recent years of structural disturbance. Second, (Bullock et al 2020) conducted their analysis for the Amazon EcoRegion as defined by (Olson et al 2001), which means a small portion of our study area (the RAISG-defined Amazon biome) is not included.

We measured the size of forest fragments in our 2020 land cover map as the number of contiguous pixels classified as each forest type. Fragment size was calculated using the Accounting tool from the Guidos Toolbox software. We applied 8-way connectivity and identified 'extensive' forest as those fragments over 50 km².

2.4.1. Distance to nearest old-growth forest

2.4.2. Landscape context

To map old-growth edge forest, we calculated the Euclidean distance from every old-growth pixel to the nearest old-growth forest edge using the Guidos Toolbox Distance tool. Any pixel within 120 m of an edge was marked as edge forest. The most intense edge effects occur within 100 m of an edge (Laurance et al 2002); 120 m is the closest distance beyond this that is measurable using 30 m resolution pixels. In cases where a forest fragment is less than 120 m in width, the entire fragment is considered edge forest, as all forest within the fragment would be subject to edge effects. We repeated this analysis for all forest cover (old-growth and secondary forest combined). Any old-growth forest pixels that were no longer in the edge zone after the inclusion of secondary forests were identified as buffered edge forest, while those that remained within the edge zone were considered exposed edge forest. We conducted this analysis for every year in the time series to produce annual maps of edge forest exposure (1986–2020). Old-growth edge forest present in 1985 was excluded from the analysis as we could not determine its age or duration of exposure. Excluding this edge forest also served to remove natural edges from our analysis. This method is similar to that of Silva Junior et al (2020a); differences in results can be attributed to the longer timeseries used in our study. We repeated our analysis for three secondary forest age classes (any, >5 years old, >15 years old).

From our annual maps of forest edges, we calculated the age of old-growth edge forests in 2020 as the total number of years since a pixel was within the edge zone. We randomly sampled the exposed and buffered edge ages and used the Wilcoxon rank sum test to assess whether the samples were drawn from different distributions. We repeated this process 10 000 times and recorded the mean p-value. We undertook these analyses with a variety of sample sizes. However, results were insensitive to sample size (table S2) and we report results for n = 1000. We also used our annual edge maps to calculate the total number of years each pixel of edge forest was classified as exposed or buffered.

To explore the spatial relationship between old-growth forest cover and edge buffering, we divided the Amazon biome into a regular grid of ∼58.9 km² cells (65 536 pixels; pixel size: 0.0009 km²; size determined by computational efficiency). For each grid cell, we calculated the percentage of land cover that was occupied by old-growth forest and the percentage of old-growth forest edges that were buffered by secondary forest. Cells with >99.9% of pixels with a land cover classification of 'other' (i.e. where less than 0.1% of the cell area is capable of being forest) and cells where no forest edges are present were excluded from the grid level analysis. We then applied Spearman's rank-order correlation to a sample of grid cells to test the association between old-growth forest cover and buffering by secondary forests. To enhance the dispersal of selected grid cells across the study area, we used stratified sampling with replacement such that 25% of pixels were situated in each quadrant of the Amazon biome, while within-quadrant selection was random. We repeated this process 10 000 times, recording the mean correlation coefficient. We undertook these analyses with a variety of sample sizes. However, results were insensitive to sample size (table S1) and we report results for n = 1,000.

We measure the influence of secondary forests on old-growth forest connectivity by calculating the change in the number of isolated fragments and fragment size. First, we used the Guidos Toolbox Accounting tool to identify individual old-growth forest fragments and determined their area. To measure the effect of secondary forests on fragmentation, we repeated this analysis for three secondary forest age thresholds (all secondary forest, secondary forests >5 years old, and secondary forests >15 years old) and four minimum secondary forest bridge widths (0 m, 60 m, 120 m, 240 m). To identify secondary forest bridges meeting each minimum width threshold, we calculated the Euclidean distance from every secondary forest pixel to the nearest forest edge using the Guidos Toolbox Distance tool. To test for differences in median fragment size resulting from secondary forest bridges of different ages and width, we randomly sampled fragments and used the Kruskall–Wallis H-test. We repeated this process 10 000 times and recorded the mean p-value. We undertook these analyses with a variety of sample sizes. However, results were insensitive to sample size (tables S3 and S4) and we report results for n = 1,000.

In this analysis we used age thresholds rather than discrete age classes, as species that are sensitive to secondary forest age will typically have a lower age tolerance limit but not an upper age limit—meaning if a species can move through a 5 year-old secondary forests, it can also make use of any secondary forest that is older than that threshold.

In 2020, there was 189 451 km² of secondary forest in the Amazon biome (figure 2); found in more than 5.3 million spatially discrete patches that ranged in size from 4500 m² (the smallest area detectable by our analysis) to 125 km². 41.2% of all secondary forest (78 059 km²) was directly adjacent to an old-growth forest, while 94.1% of secondary forest was within a fragment connected to an old-growth forest fragment (the latter includes the former as directly adjacent secondary forest is also within a connect fragment). These figures were only slightly lower when we restricted our assessment to structurally-intact old-growth forests, with 33.2% of secondary forest directly adjacent and 92.6% within a connected fragment. However, the proximity measures were much lower when we restricted our assessment to extensive, structurally-intact old-growth forest, with only 20.1% of secondary directly adjacent and 57.4% within a connected fragment.

Zoom In Zoom Out Reset image size

The median (inter-quartile range; IQR) distance of secondary forest pixels to old-growth forest was 30 m (30–120), 60 m (30–210), and 1110 m (30–8010) when we compared to any old-growth forest, structurally-intact old-growth forest, or extensive structurally intact old-growth forest, respectively (figure 2). Across the biome, less than half of all secondary forests were within 1 km of extensive, structurally-intact old-growth forest (48.9%; 92 674 km²), but the majority were within 1 km for any old-growth forest or structurally-intact old-growth forest (98.6% and 96.8% respectively; figure 2(C)).

Less than half of secondary forest pixels (45.5%; 85 759 km²) had >50% old-growth forest in the surrounding landscape (0.99 km radius) and 21.7% (40 907 km²) had <10% surrounding old-growth forest cover. Restricting the analysis to structurally intact old-growth forest, 33.0% (62 604 km²) of secondary forests had >50% surrounding old-growth forest cover, while 30.8% (58 271 km²) had <10% surrounding old-growth forest cover. With further restriction to extensive structurally intact old-growth forest, only 23.0% (43 545 km²) of secondary forests had >50% surrounding old-growth forest cover, and 63.2% (119 666 km²) had <10% surrounding old-growth forest cover.

In 2020, 6.4% (348 903 km²) of old-growth forest in the Amazon biome was within 120 m of an anthropogenic edge created after 1985. However, 41.1% (143 392 km²) of this old-growth edge forest was buffered by secondary forests (figure 4(A)). The extent of buffering reduced to 32.2% (112 448 km²) when restricted to secondary forests more than 5 years old, and 22.9% (79 902 km²) for secondary forests more than 15 years old.

The proportion of buffered edges varied spatially with old-growth forest cover, with less buffering in areas with low old-growth forest cover (p ⩽ 0.01; table). This trend was less pronounced when restricting buffering to secondary forests of ⩾5 years or ⩾15 years old (figure 3). In highly deforested landscapes (<20% forest cover remaining), the mean ± SD percentage of buffered edge forest was 24.6 ± 17.5%, while in more intact regions (>80% forest cover remaining) it was 50.4 ± 27.7%.

Zoom In Zoom Out Reset image size

The median age (years since a pixel became an edge forest) of edge forest present in 2020 was 19 years (IQR, 8–27; also see (Silva Junior et al 2020a), but buffered edge forest was typically older (23 years, 15–29) than exposed edge forest (14 years, 4–24; Wilcoxon rank sum: W = 9.81, p ⩽ 0.01). One-year old edge forest (i.e. edge created in 2020) was almost double the extent of any other annual age class (1 year-old: 25 769 km²; older edge classes: 6472–15 114 km^2; figure 4(A)). Edges had a median (IQR) exposure time of 6 years (2–14) and over a third (36.9%) of edges had never been buffered by secondary forest (figure 4(C)). Where edges were buffered, on average, this occurred within 3 years (2–5) of creation.

Zoom In Zoom Out Reset image size

Of the 681 027 km² of old-growth forest that became edge forest during our time series (1986–2020), almost half (48.8%) was subsequently deforested by 2020. The median (IQR) time between becoming edge forest and being deforested was 4 years (2–9), with buffered edge forests typically older (median, IQR; 9 years, 6–15) at clearance than exposed edge forest (4 years, 1–8; figure 4(B)).

In 2020, there were 3.3 million old-growth forest fragments in the Amazon biome (figure 5), with a median (IQR) fragment size of 4500 m² (1800–13 500; (figure 6(A)). Allowing any secondary forest to act as a bridge between old-growth fragments reduced the total to 1.3 million fragments (excluding those comprised of only secondary forest; figure 5), almost doubled the median (IQR) fragment area (8100 m², 3600–26 100; the area of old-growth forest within a mixed-forest fragment), and reduced the area of old-growth forest in small fragments (<50 km²) by 31.2% (77 456 km²; figures 6(A) and (B)). When applying a minimum age (of either 5 or 15 years) for secondary forest bridges, the effect of secondary forest on fragmentation was reduced but still pronounced (figure 6; Kruskal–Wallis: H = 104.48, p= ⩽0.01). Applying a minimum width to secondary forest bridges created a similar pattern—increasing minimum bridge width resulted in more fragments and smaller median fragment size (figures 6(C) and (D)); Kruskal–Wallis: H = 25.86, p = ⩽0.01).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In 2020, almost half of Amazonian secondary forest was growing directly adjacent to old-growth forest. This is potentially beneficial news for forest recovery potential, as, all else being equal, restoration success is highly associated with remaining forest cover, (Crouzeilles et al 2019) and proximity to old growth supports the recovery of high biomass and high diversity ecosystems (Jakovac et al 2015, Mayhew et al 2019). These advantages span the successional life of a secondary forest, supporting both the early (Toledo et al 2020) and later stages of succession, when they provide the diverse seed sources required to establish resilient secondary forests (Martínez‐Ramos et al 2016, Hawes et al 2020).

Our results show that the proximity metrics are sensitive to changing the extent and intactness of old-growth forest fragments, with secondary forest adjacency declining by as much as 64%. These findings highlight an important research gap, as it is not clear how old-growth condition or extent affect the spill over of benefits to adjacent secondary forests. As old-growth forests become increasingly fragmented (Montibeller et al 2020, Fischer et al 2021) and structural disturbance continues to spread (Bullock et al 2020, Matricardi et al 2020), furthering our understanding of these relationships will be integral to designing effective restoration programmes, especially in landscapes where there is little old-growth forest remaining (Crouzeilles et al 2019).

The proximity of secondary forests to old-growth forests indicates they could have a substantial role in mitigating edge effects: 43.1% of old-growth edge forests were buffered by secondary forests. However, there is scope for improvement, as over half (56.9%) of all old-growth edge forests are not buffered. We found opportunities for increasing edge buffering across the Amazon, with less than half of edge forests buffered in over two-thirds of landscapes (68.8%; figure 3). These opportunities are currently greatest in highly deforested landscapes, suggesting that policy measures aiming to encouraged restoration on private lands would benefit from incentivising buffering. There are opportunities too in high forest cover landscapes, where existing buffers are dominated by young secondary forests (figure 3). Incentivising secondary forest permanence could prevent major loss of carbon stocks from old-growth forest along newly-created edges (Silva Junior et al 2020a) and aid in preventing old-growth deforestation (Wang et al 2020).

Loss of carbon stocks in exposed edge forest is rapid, with declines of over 20% within 1 year of edge creation (Silva Junior et al 2020a). Based on the edge exposure times found in this study and the 15 year carbon stock decay curve reported by (Silva Junior et al 2020a), as much as 33.4% of the carbon stock in old-growth edge forests (⩽15 years old) may have already been lost. The role of secondary forests as buffers is also relevant for biodiversity conservation. Changes in microclimate and structure at forest edges (Laurance et al 2002) mean that for many species, the amount of forest interior habitat available in landscapes with highly fragmented forest is considerably smaller than the total forest cover. Our findings show that across the biome 6.4% of remaining old-growth forest cover is within 120 m of an edge, meaning core-forest species that avoid edges, (e.g. as reported by Pfeifer et al 2017), may have lost as much as 348 903 km² of habitat in addition to that already lost to deforestation. However, if the secondary forest that borders almost half of that old-growth forest edge can mitigate microclimatic changes, it could be playing an important role in increasing habitat availability for forest-specialist species.

The potential benefits of secondary forests as buffers for old-growth forest edges are numerous and include reduced edge-related tree mortality (Mesquita et al 1999). However, more research is needed to quantify the scale of these effects, and the underlying mechanisms that drive them. With 63% of old-growth edge forests in 2020 having been buffered at some point during the study period, an important question is whether secondary forest buffering simply prevents further degradation or enables old-growth edge forests to recover some of their ecological integrity. Equally, the increase in height, leaf area, and canopy complexity of secondary forests during succession (Pena-Claros 2003, Feldpausch et al 2005) suggests that the age of a secondary forest would strongly influence its effectiveness as a buffer. Given that the proportion of buffered edge forest drops from 41% to 22% when restricting buffers to secondary forests ⩾ 15 years-old, the impacts of an age-related relationship could be substantial. The effectiveness of a secondary forest buffer in mitigating different edge effects in old-growth forest could also vary with its width (Laurance et al 2002). However, we lack sufficient data to demonstrate the form of these relationships, or identify thresholds that could be used to guide management. A more comprehensive understanding of how secondary forests influence old-growth edge forest could help improve estimates of carbon emissions and guide conservation planning for some of the world's most vulnerable species.

The considerable reduction in old-growth fragment isolation resulting from secondary forest bridges further demonstrates that the value of secondary forests extends far beyond their ∼189 451 km² extent. The connectivity provided by secondary forests could enable movement of organisms between old-growth forest fragments, maintaining or restoring genetic diversity and alleviating the extinction risk where fragments that are too small to support long-term viable populations (Metzger et al 2009, Newmark et al 2017).

However, the value of secondary forests for connectivity remains very uncertain, and future research needs to quantify the parameters of viable secondary forest bridges. For example, species differ in their ability to move through secondary forests of different ages (Powell et al 2013, 2015). Our analysis suggests this is a key criterion at the landscape level, as there were an additional ∼400 000 isolated fragments of old-growth forest when we restricted bridges to secondary forests ⩾ 5 years old. The width and length of a secondary forest are also likely to be important in determining its functionality as a bridge. The width of a secondary forest bridge is important as edge effects alter abiotic conditions (Laurance et al 2002), especially in very narrow bridges that may suffer from additive edge effects (Porensky and Young 2013). Secondary forest bridge length, although not quantified in the present study, may also be important as some species may have maximum dispersal distances that they will not move beyond (Powell et al 2015). Our study evaluates the effect of secondary forests on direct connectivity, however, secondary forests may also provide indirect connectivity by acting as stepping-stones for species that are able to cross gaps (Urban and Keitt 2001, Anon 2007) . As species differ in their ability to cross gaps (Lees and Peres 2009), assessing how secondary forests affect the distance between old-growth forest fragments may also be an important metric to consider when quantifying the value of secondary forests.

In satellite imagery, pixels at the intersection of different land cover types may contain a mixture of land covers and are therefore prone to temporal variation in classification. Our method of identifying secondary forests through change detection means that if a mixed pixel was misclassified as undergoing deforestation, it would subsequently be classed as secondary forest by our algorithm if it appeared to 'return' to a forested state in future years. As such, it is possible that some of our adjacent secondary forest pixels may be a misclassification of exposed old-growth edge forest, which would lead to an overestimation of the extent of secondary forest and its role in buffering edge forest. The sharp increase in buffered edge forest in 1998 and 2016 also suggests that some degradation from fire may have been misclassified as deforestation during the 1997–1998 and 2015–2016 El Niño events—which would result in degraded forest being counted as secondary forest (e.g. Lapola et al 2023). However, the impact of such misclassifications on our results is likely to be limited as the MapBiomas dataset undergoes a rigorous validation process (MapBiomas 2021). Nonetheless, further work will benefit from improved assessments of land use change at forest edges.

Expanding Amazonian forest restoration is essential for achieving global targets on climate change mitigation and biodiversity conservation (Barlow et al 2021). Our assessment of the co-location of secondary and old-growth forests provides important insights for large-scale restoration success.

First, the substantial reductions in the edge buffering and connectivity achieved by secondary forests when restricting our analysis by secondary forest age highlight the importance of considering secondary forest permanence, and moving away from the high-turn-over, low-residency-time regime that currently dominates Amazonia's secondary forests (Schwartz et al 2017b, Nunes et al 2020, Smith et al 2020). Unfortunately, with Amazonian deforestation on the rise in recent years (PRODES 2021, Smith et al 2021), even just maintaining the existing extent of secondary forests will be a major challenge.

Second, for secondary forests to play a decisive role in avoiding the rapid loss of carbon stocks in old-growth edge forest (Silva Junior et al 2020a), they need to establish soon after deforestation, at the newest edges. Unfortunately, this is likely to be difficult to achieve in practice, as new edges are created in regions of active deforestation where forest loss rather than land abandonment dominates (Smith et al 2021). Furthermore, regions of active deforestation are characterised by low levels of governance and inadequate land tenure arrangements (Geist and Lambin 2002); the opposite of what is required for encouraging restoration (Barlow et al 2022). However, in regions where deforestation has been stopped, and governance is reinstated, incentivising natural regeneration along edges would be a sensible strategy (Chazdon and Guariguata 2016). Such incentives would offer win–win for conservation, particularly in highly deforested regions: providing additional habitat and improving the quality of adjacent forest while also preserving carbon stocks. Nonetheless, the development of any future incentive programmes must also carefully consider local socio-economic factors, as these are critical in determining the success or failure of incentivised restoration (Tedesco et al 2023).

Third, we show that even existing secondary forests have the potential to make an important difference to landscape connectivity, helping maintain species diversity and viable long-term populations in old-growth fragments (Michalski and Peres 2007) (Mayhew et al 2019). This could be particularly advantageous for critically endangered species, such as the Ka'apor capuchin monkey (Cebus kaapori), that are able to use secondary forest, (de Oliveira et al 2014). Future restoration—such as the state of Pará's Plan for the Amazon Now—could enhance the connectivity benefits of secondary forests in highly disturbed landscape seven further by incorporating large-scale systematic conservation into the planning process. However, it may be that simpler approaches are also effective: for example, studies simulating the ideal approach to restoration have shown that starting close to remaining old-growth fragments is an efficient method of improving functional connectivity (Tambosi et al 2014, Miranda et al 2021). This is not only a simple rule to apply to any landscape, it also maximises the chances of successful restoration and the potential benefits for buffering old growth forest edges.

Although our definition of secondary forest encompasses both restoration initiatives and natural regeneration, currently the vast majority of Amazonian secondary forest is natural regeneration. As an example, one of the most famous assisted regeneration schemes to date—Rede de Sementes do Xingu—has restored only 5000 ha of forest in a decade, a tiny proportion of all secondary vegetation (Barlow et al 2021). Nonetheless, the concepts surrounding proximity to old-growth forest can be applied equally to both types of secondary forest.

Our exploration of the spatial relationship between old growth and secondary forests is intended to highlight the potential importance of this rarely considered relationship, acting as an catalyst to encourage., more research that addresses the potential benefits. In particular, field and remote sensing studies are required to understand the importance of old-growth forest size, condition and proximity for secondary forest recovery; how that relationship changes over time, and if old-growth forest are most important at a particular stage. We also highlight the need for more research on the potential benefits of secondary forests as buffers of old growth edges, exploring how this is mediated by secondary forest widths and ages. Advancing these research areas will not only improve our understanding of restoration strategies, it will also help refine estimates of forest carbon stocks and habitat provision for tropical forest biodiversity.