Forest fragmentation and its associated edge-effects reduce tree species diversity, size, and structural diversity in Madagascar’s transitional forests

Tropical forests are being cleared at an accelerating rate, despite being one of the most important habitats for global biodiversity. Many remaining tropical forest tracts are now highly degraded and fragmented, which presents a major problem for sensitive and threatened forest-dwelling species that depend on this habitat for survival. In this study, we assessed the impacts of forest fragmentation, and its associated edge-effects, on tree species diversity, tree size, and structural diversity within the transitional forests of north west Madagascar. Using data collected from 9,619 trees within 200 vegetation plots, we calculated species diversity indices, a range of dendrometry measurements, and Shannon-Weaver diversity indices of structure, which we compared among core and edge areas of a continuous forest and a fragmented forest. We found that species diversity, tree size, and structural diversity was significantly reduced in fragmented forest, and within forest edge areas in comparison to core, continuous forest. We also observed species diversity and structural diversity to be strongly influenced by the total size, core area size, and shape of forest fragments. Whilst we found fragmentation and edge-effects to individually impact tree species diversity, size and structural diversity, fragmentation and edge-effects are strongly correlated and affect natural forest synergistically. Our results provide evidence that forest fragmentation seriously degrades habitat quality and integrity of transitional forests, which is of great concern for the threatened species that inhabit them. Urgent conservation efforts are needed to halt ongoing forest fragmentation throughout the tropics, and reforestation and restoration efforts are required to reconnect isolated forest patches and to reduce forest edge area.


Introduction
Tropical forests are arguably the most ecologically important habitat type in the world as they harbor more than 50% of the world's terrestrial floral and faunal species (Wilson 1988;Duivenvoorden et al. 2002;Bonan 2008). Despite their importance for global biodiversity, tropical forests are being destroyed at a faster rate than any other ecosystem (Myers 1988) and it is now estimated that over half of the world's tropical forests (~ nine million km²) have been cleared in the last century (Poorter et al. 2016;McFarland 2018). The primary driver of tropical deforestation has historically been human population growth (Godfray et al. 2010;Busch and Ferretti-Gallon 2017), as the increasing population has resulted in an increased global demand for food and other resources, resulting in large-scale agricultural expansion and resource extraction within tropical forest habitats (Myers 1993;Foley et al. 2005). Due to the ongoing deforestation, much of the remaining forest habitat in tropical regions is now extensively fragmented (Taubert et al. 2018), and areas of isolated forest fragments that are geographically separated by expanses of grassland now characterize landscapes that were once continuous forest tracts (Wade et al. 2003;Brinck et al. 2017). Further to this, many of the remaining forest fragments have high forest edge areas in comparison to continuous forests (Brinck et al. 2017), resulting in a high level of edge-effects which have an influence on the functioning of the fragment's ecological processes (Fahrig 2003;Laurance et al. 2007). Many forest fragments are also degraded, of poor habitat quality and low ecological integrity (Coe et al. 2013;de Paula et al. 2015) and forest fragments vary greatly in their size and shape (Hill and Curran 2003;Taubert et al. 2018). Tropical forest fragmentation is therefore regarded as one of the biggest threats to animal and plant populations worldwide (Turner, 1996).
Many forest-dwelling species now survive only in fragmented tropical forest (Shelton 1985), and it is paramount to understand how the forest fragmentation process impacts on forest population composition and diversity (Zuidema et al. 1996). This is also important for the protection and restoration of forest habitat and the species-specific conservation of forest species, many of which are classified as threatened on the IUCN Red List (Pimm et al. 1995). The effects of fragmentation and forest area on species diversity are already wellstudied (e.g., Laurance and Bierregaard 1997;Pimm 1998;Brosi et al. 2008;Dayananda et al. 2016), and the relationship between animal and plant species diversity and forest fragment size (e.g., Bierregaard et al. 1992;Benítez-Malvido and Martínez-Ramos 2003;Pardini et al. 2005;Harvey et al. 2006;Tripathi et al. 2010;Santana et al. 2021) and fragmentation associated edge-effects (e.g., Laurance 1991;Laurance et al. 2007;Sundarapandian and Karoor 2013;Vetter et al. 2013;Razafindratsima et al. 2018;Santana et al. 2021) is particularly well documented. However, conservation of threatened species, and conservation of the forest ecosystem as a whole, also requires a specific understanding of how forest fragmentation effects the forest habitat itself (Arroyo-Rodríguez and Mandujano 2006; Echeverría et al. 2007). Some studies have already found that forest fragmentation impacts the floral species diversity (e.g., Scariot 1999;Paciencia and Prado 2005;Viña and Estévez-Varón 2019), structural diversity (e.g., Benitez-Malvido 1998;Laurance et al. 2001;Wassie et al. 2010) and the availability of potential food sources for animals (e.g., Hart et al. 2013;Hadley et al. 2014) within a forest. Most studies document forest fragmentation as having a negative effect on forest floral and structural diversity, and some investigations also report positive relationships between forest fragment size or shape and tree species diversity (e.g., Ranta et al. 1998;Hill and Curran 2003;Page et al. 2010) and differences between forest interior and edge (e.g., Oliveira-Filho et al. 1997;Oliveira et al. 2004). As forest fragmentation becomes ever more profound within tropical regions, efforts to understand the relationships between habitat quality and forest fragmentation, fragment size, shape, and edge-area must be intensified in order to maintain and manage what forest habitat remains, and to mitigate fragmentation drivers (Hill and Curran 2003).
The island of Madagascar has arguably experienced the highest deforestation and forest fragmentation rates of any country in the world (Goodman et al. 2018). It has been estimated that up to 80% of Madagascar's original forest cover has been cleared in the last century alone (Harper et al. 2007;Vieilledent et al. 2018) and much of the little forest habitat that remains is now fragmented, degraded, and sporadically distributed within a matrix of grassland and savannah (Vieilledent et al. 2018). Madagascar is considered a biodiversity hotspot and the ongoing forest fragmentation has put enormous pressures on its biodiversity, of which 85% of its animals and 90% of its plants are endemic (Ganzhorn et al. 2001;Isambert et al. 2011), and the island is now one of the world's primary conservation concerns . Some studies of the effects of forest fragmentation on Madagascar's forest floral diversity and structure have been conducted at various localities throughout the island, and forest fragmentation has been documented to have mixed effects on forest habitat between different studies undertaken at different locations (Cadotte et al. 2002;Irwin et al. 2010;Malcolm et al. 2017;Razafindratsima et al. 2018). In addition, floral species diversity and structural diversity have been observed to be variable between forests at different localities (Ratovonamana et al. 2011;Andriaharimalala et al. 2012), over degradation gradients (Ganzhorn 1987;Hending et al. 2020) and between edge and core areas (Lehman 2016;Malcolm et al. 2017), suggesting that the effects of fragmentation on forest habitat quality are variable and possibly site-specific. These past studies, coupled with Madagascar's high bioclimatic zonation (Brown et al. 2016), advocate that an understanding of forest fragmentation induced changes in forest habitat is required on a localized scale in order to successfully manage and restore forest habitat within areas of high conservation priority. Further, this localized knowledge is also required to protect Madagascar's threatened animal taxa, which are often confined to small areas and are thus specialized to the forests within their respective ranges (Wilmé et al. 2006).
We conducted this study within the Sahamalaza-Iles Radama National Park (hereafter referred to as SIRNP) of the Sambirano domain, north west Madagascar. Forest habitat structure has already been reported to vary throughout SIRNP and strong habitat and microhabitat preferences have already been documented for resident animal species, particularly lemurs, that live within it (e.g., Seiler et al. 2013Seiler et al. , 2014Hending et al. 2017aHending et al. , 2021a. However, there are little data on how fragmentation affects tree species diversity, dendrometry (physical structure and size of a tree), and structural diversity. Further, nothing is known of the effects of fragment spatial geometry on tree diversity, size, and structure within the National Park, and the wider Sambirano domain more generally. In this study, we investigated how both forest fragmentation and edge-effects impact tree species diversity, size, and structural diversity, and whether the effects of fragmentation and edge-effects are independent or synergistic. Tree species diversity, tree size, and forest structural diversity have been observed to vary between continuous and fragmented forest, between forest edge and core areas, and among forest fragments of varying area and shape in Madagascar and other areas of the tropics more-generally (e.g., Didham and Lawton 1999;Brown and Gurevitch 2004;Harper et al. 2005;Krishnadas et al. 2019;Albiero-Júnior et al. 2021). However, fragmentation and edge-effects are often correlated and not independent of each other Malcolm et al. 2017). We therefore tested the hypothesis that forest fragmentation and edge-effects impact tree species diversity, tree size, and structural diversity synergistically. Specifically, we predicted higher tree species diversity, higher mean tree size, and greater forest structural diversity in continuous forests, core areas, and larger forest fragments than in fragmented forests, edge areas, and smaller forest fragments respectively, and that fragmentation effects and edge-effects would be strongly correlated and non-independent (as in Hill and Curran 2003;Ewers et al. 2007;Krishnadas et al. 2019).

Study site
SIRNP is a 26,000 hectare protected area located primarily on the remote Sahamalaza Peninsula in Madagascar's Sofia region, between the latitudes of 14°04'S − 14°37'S and longitudes of 47°52'E − 48°04'E ( Fig. 1). In addition to its National Park status, SIRNP has also been designated a UNESCO Biosphere Reserve since 2007 (Rode et al. 2013). Despite this current protection, it has historically undergone severe deforestation and habitat fragmentation (Seiler et al. , 2014. Much of SIRNP is now characterized by grassland and savannah, although there are some remaining isolated forest tracts and matrices of gallery and scrub forest (Volampeno et al. 2011). SIRNP is located in Madagascar's Sambirano domain, an area of the north west of the island that is characterized by sub-humid transitional forests, and the vegetation of the National Park is comprised of a mixture of both deciduous and evergreen plant species, many of which are locally endemic (Koechlin 1972;Du Puy and Moat, 1996). SIRNP has a hot, sub-humid and seasonal climate, with a distinct hot, wet season (November -April) and a cooler dry season (May -October). The mean temperature range of SIRNP is 20.6-32.0 °C, with an extreme temperature range of 13.2-39.1 °C (Volampeno et al. 2011;Hending et al. 2017b) and an annual mean precipitation of ~ 1,600 mm (Schwitzer et al. 2007).
SIRNP is an ideal site at which to investigate the effect of forest fragmentation and its associated edge-effects on tree species diversity and forest structure. This is firstly because the National Park contains two remaining forests: the 1,169 ha Anabohazo forest, a continuous, undisturbed forest block that has undergone little disturbance (Randriatahina et al. 2014), and the 1,020 ha Ankarafa forest, a fragmented forest composed of ten fragments of varying size, shape, and with varying levels of forest edge-core ratios . Secondly, the edge-core gradient of SIRNP's forest has already been established from transects measuring the variation of microclimatic and abiotic variables (temperature, humidity, wind prevalence, and light intensity) from the forest edge into the forest core. This edge buffer has been reported as 165 m (Mandl 2018, unpublished thesis). The Anabohazo and Ankarafa forests are separated by only ~ 25 km of savannah, and this provides a rare opportunity to study the effects of forest fragmentation on a localized scale, where environmental covariates are more easily controlled, in comparison to other studies that have occurred over a larger area (e.g., Scariot 1999;Laurance et al. 2001).

Sampling of vegetation diversity and structure
Between 06/03/2019 and 15/05/2022, we conducted 200 vegetation plots, each measuring 20 × 20 m, to characterize tree species diversity, tree size, and forest structure within both the Anabohazo (N = 100) and Ankarafa (N = 100) forests. The 100 Ankarafa plots were located within ten separate forest fragments (10 plots per fragment) that varied in size and shape (Supplementary File 1). In Anabohazo forest, 50 of these plots were located within the forest core whilst 50 were within the forest edge area, defined as being within 165 m of the forest edge. In Ankarafa, 45 plots were located within the forest core whilst 55 were within the forest edge area (five plots in both edge and core for nine of the 10 Ankarafa fragments); this is because the Ankarafa II fragment had no core area (Supplementary File 1), and so all 10 plots for this fragment were located within a forest edge area. The vegetation plot locations were situated at ~ 50 m intervals along a system of established transects within each forest (Anabohazo N = 7, Ankarafa N = 10 (one per fragment). The plots themselves were established a minimum of 10 m from the transect line. Within each plot, we identified all trees with a circumference at breast height (CBH) of ≥ 16 cm (Armstrong et al. 2011) using the knowledge of local guides and we noted their colloquial (local) names. We then consulted the existing literature of the flora of Madagascar (Schatz, 2001) to identify the trees to species or genus level where possible and to determine whether they were an endemic native, a non-endemic native, or an introduced species to Madagascar. For each identified tree in the plot, we also measured the CBH with a measuring tape and we used a laser rangefinder (Excelvan, El Monte, California, USA) to measure the tree's total height, bole height and crown diameter. We measured crown diameter by positioning two members of the field team under opposite sides of the widest point of the tree's crown, and we measured the distance between the two researchers with the laser rangefinder. We converted our CBH values to diameter at breast height (DBH) values by dividing CBH by π. To measure tree density, we used the point-centre quarter method (Ganzhorn 2003), where we measured the distance from each identified tree to the nearest adjacent tree over two metres tall in four geographic directions using a measuring tape. We then calculated the mean of these four point-to-plant distances, squared it, and then calculated its inverse to obtain density values of the number of trees per square metre (N/m 2 ) as measure of the microhabitat density around each tree.

Analysis of vegetation species diversity
To analyze how tree species diversity varied between the forests, and between edge and core areas, we first used the EstimateS software (Colwell 2013) to obtain values for the expected number of species (S) in the edge and core areas of each forest using rarefaction (Colwell and Coddington 1994). We formed subsets of data by pooling the plots for each forest and the separate edge and core areas of each forest, and we specified the number of knots for the rarefaction as the number (N) of trees included within each subset of data (Colwell et al. 2004). We specified for EstimateS to extrapolate the data up to a maximum of 5,000 knots for all data subsets, as the sampled number of trees was uneven between the two forests, and edge and core areas. We then constructed species-individual curves (with 95% confidence intervals) using the rarified and extrapolated S values so that tree species diversity could be compared between the Anabohazo and Ankarafa forests, and the edge and core areas. We did this by assessing the overlap between the 95% confidence intervals of the curves as a conservative means of establishing statistical significance (Colwell 2013).
We also used the EstimateS software to calculate Chao 1 species diversity indices (with 95% confidence intervals) for each forest's core and edge areas to statistically compare species diversity among sampled areas. We chose this approach, rather than conducting analysis on standardized N species/N trees values, as these indices compensate for bias caused by undetected species and therefore represent lower bounds for the true species diversity of each sampled area (Chao 1984(Chao , 1987. For the species-individual curves, we compared Chao 1 species diversity between the two forests and their edge and core areas by assessing the overlap between 95% confidence intervals of the curves. Chao 1 indices cannot be extrapolated, and so we compared the indices at the point of maximum runs for the area with the fewest sampled individuals (Hending et al. 2023). We constructed all species-individual curves and Chao 1 index-individual curves in R Studio (2020) using the 'ggplot2' package (Wickham 2016).

Analysis of vegetation structure
Prior to analysis, we assigned our DBH, tree height, crown diameter, and tree density values into frequency bins. The frequency bins were as follows; DBH: 5.0-9.9 cm, 10.0-14.9 cm through to 45.0-49.9 and ≥ 50.0 cm; height: 1.50-2.99 m, 3.00-4.49 m through to 13.50-14.99 m and ≥ 15.00 m; crown diameter: < 1.99 m, 2.00-2.99 m, 3.00-3.99 m through to 9.00-9.99 m and ≥ 10.00 m; tree density: 0.00-0.49 n/m 2 , 0.50-0.99 n/m 2 through to 4.00-4.49 n/m 2 and ≥ 4.50 n/m 2 . We then used these frequency datasets to calculate Shan-non-Weaver structural diversity indices (H) (Shannon and Weaver 1949) for each variable for each plot using the formula: where P i = the fraction of the entire population made up of species i, and S = total number of classification bins. This was so that we could then conduct analysis of structural diversity (following Ganzhorn 1987;Andersohn 2004).

Statistical analysis
We conducted all statistical analyses in R Studio using a 95% confidence interval. To check that all data fields were of normal distribution and homogenous variance, we performed Levene tests using the 'car' package (Fox and Weisberg 2019) on each data field that we wanted to compare. Whilst all Levene test results were not significant at the plot level, all Levene test results were significant at the individual tree level (Supplementary File 2). We therefore used Multivariate Generalized Linear Mixed Models (GLMMs) in the 'lme4' package (Bates et al. 2015) to compare tree species diversity, dendrometry, and structural diversity between (1) the two forests, and (2) between edge and core areas, and to test the relationship of these two fixed effects. Within all GLMMs, we assigned the tree diversity, dendrometry, or structural diversity metric that we were comparing as the response variable (one response variable per model), and we used two grouping variables (forest and edge/core). Although our study was localized within SIRNP to minimize random effects from environmental and climatic variables between the 200 plots, we included mean annual temperature and annual precipitation values as random factors to control for fine-scale environmental differences between plots. We obtained the values of mean temperature and precipitation for each vegetation plot from WorldClim (Fick and Hijmans 2017: 1 km resolution) using the 'raster' (Hijmans 2021) and 'sp' (Pebesma and Bivand 2005) R packages. We did not include elevation as an additional random factor, as elevation correlates with temperature and precipitation (Hending 2021a). To compare mean tree species per plot and structural diversity Shannon-Weaver indices between forest areas, we conducted our GLMMs at the plot level (N = 200), where all data points were vegetation plot means. To compare tree dendrometry measurements between forest areas, we conducted our GLMMs at the individual tree level (N = 9,619), where all data points corresponded to individual trees. Although individual trees can be regarded as independent data points, we ensured that any plot-based effects were removed from the subsequent GLMM results by including the vegetation plot ID, mean temperature and precipitation as random factors within the models. The full results of all GLMMs can be viewed in Supplementary File 3. In addition, we also performed GLMMs with one grouping variable with four categories (Anabohazo core, Anabohazo edge, Ankarafa core, and Ankarafa edge) to further explore how tree species diversity, tree size and tree structural diversity varied between the four surveyed forest areas. The results of these single fixed-effect GLMMs can be viewed in Supplementary File 4. To investigate the relationship of tree diversity, size, and structural diversity with fragment size, core area size, and fragment shape (Patton 1975), we performed linear models (LMs, full results in Supplementary File 5). As with the GLMMs, we included mean temperature and annual precipitation values for each forest fragment as random factors within our LMs.

Tree species diversity
In total, we identified and measured a total of 9,619 trees from 225 different species (Anabohazo core N = 2,446, Anabohazo edge N = 2,095, Ankarafa core N = 2,250, Ankarafa edge N = 2,828). We confirmed the presence of 203 tree species within Anabohazo forest (core N = 189, edge N = 134), whilst we were able to confirm the presence of 118 tree species in Ankarafa forest (core N = 95, edge N = 107) (Supplementary File 6). We were able to identify 94 tree species to the species level, 120 species to the genus level, whilst we could not identify 11 tree species. The mean number of tree species per vegetation plot was significantly different among forest areas, with Anabohazo core plots having a higher mean number of tree species than Anabohazo edge and both areas of Ankarafa (Supplementary File 4). Anabohazo forest had a significantly higher expected number of tree species than Ankarafa forest when comparing the interpolated data (at the point where N = 2,095 individuals) and the extrapolated data (N = 5,000 individuals) ( Fig. 2A). Anabohazo core also had a significantly higher expected number of tree species than Anabohazo edge. Both Anabohazo forest areas had a significantly higher Chao 1 species diversity index than both Ankarafa forest sites, although there was no difference between core and edge areas in both forests (Fig. 2B). Both forest fragmentation (Estimate = -7.568, df = 25.144, t = -4.488, P < 0.001) and edge-effects (Estimate = -5.939, df = 193.255, t = -5.066, P < 0.001) influenced tree species diversity, and there was a significant interaction between the two variables (Estimate = 3.921, df = 193.065, t = 2.364, P = 0.019, Correlation = 0.310) (Supplementary File 3).
Our linear models highlight that tree species diversity correlates positively with forest fragment size (R = 0.554), forest core area percentage (R = 0.452), and fragment shape index (R = 0.566) (Fig. 3). However, the results of these LMs were not statistically significant (Supplementary File 5).

Dendrometry
Tree DBH, height, crown diameter, and tree density were significantly different between the four compared forest areas (Table 1; Fig. 4, Supplementary File 4). Both forest fragmentation and edge-effects had significant impacts on mean DBH, height, crown diameter and tree density, ( Our LMs highlight that mean tree DBH and mean crown diameter had a strong positive correlation with forest fragment size (R = 0.919 and R = 0.773 respectively), forest core area percentage (R = 0.868 and R = 0.719 respectively), and fragment shape index (R = 0.779 and R = 0.304 respectively), whilst mean tree height and tree density had a strong negative correlation with fragment size (R = 0.887 and R = 0.754 respectively), core area percent-age (R = 0.877 and R = 0.420 respectively), and shape (R = 0.841 and R = 0.538 respectively) (Fig. 5). Although almost all of these results indicated strong correlations (R > 0.700), the results of our LMs indicate that most of these correlations were not statistically significant (Supplementary File 5).

Tree structural diversity
DBH, height, crown diameter, and tree density Shannon-Weaver diversity were significantly different between the four compared forest areas ( Table 2; Fig. 6, Supplementary File 4). Forest fragmentation had significant effects on DBH, crown diameter and tree density Shannon-Weaver diversity, whilst edge-effects significantly impacted DBH, tree height and crown diameter Shannon-Weaver diversity (Table 2, Supplementary File 3). Interactions between fragmentation and edge-effects were significant for DBH and tree height Shannon-Weaver diversity, but not for crown diameter or tree density ( Table 2).
Our linear models highlight that tree DBH, height, and crown diameter structural diversity correlated positively with forest fragment size (all R > 0.700) and forest core area percentage (all R > 0.630). Whilst DBH structural diversity also correlated positively with fragment shape index (R = 0.717), height and crown diversity correlated negatively (R = 0.469 and R = 0.337 respectively). Tree density diversity indices correlated negatively with all three fragment-related variables (Fig. 7). Similar to the dendrometry variable results, structural diversity variables also had very strong correlations with fragment size, core area, and shape (R > 0.700), yet results of the LMs indicate that most results were not statistically significant (Supplementary File 5).

Tree species diversity
The results of our study demonstrate that forest fragmentation and forest edge-effects both individually have detrimental effects on tree species diversity within the transitional forests of north west Madagascar (Fig. 2). Further, there were significant positive interactions between the effects of forest fragmentation and edge-effects, and so both processes appear to impact tree species diversity synergistically. We observed fewer tree species in smaller fragments and those with lower core forest areas, suggesting that tree species diversity correlates to the magnitude and severity of the forest fragmentation (Fig. 3). These findings conform to our original hypothesis, and mirror the results from other studies that have assessed how vegetation diversity is impacted by forest fragmentation (Benítez-Malvido and Martínez-Ramos 2003; Hill and Curran 2003; Aguirre-Jaimes et al. 2021; Santana et al. 2021). Whilst our study highlights a higher species diversity in continuous forest in comparison to fragmented forest, tree species diversity is likely to be even greater in con- Fig. 3 The relationship between forest fragment total area (red), forest fragment core area percentage (blue), and forest fragment shape index (green) with mean tree species per vegetation plot for 10 forest fragments within Ankarafa forest of the Sahamalaza-Iles Radama National Park, north west Madagascar. Relationships investigated with linear models (full results in Supplementary File 5). Shaded areas represent 95% confidence intervals Table 1 Summary (mean ± standard deviation (range)) of four dendrometry measurements (mean tree DBH, mean tree height, mean tree crown diameter, mean tree density) of each tree within vegetation plots located within core and edge forest areas of two forests in the Sahamalaza-Iles Radama National Park, north west Madagascar, with results of Multivariate Generalized Linear Mixed Models assessing the impacts of forest fragmentation, edge-effects, and their interaction with each other, on each tree structural diversity variable. Values with the same superscript letter within a row were not found to be significantly different from each other in a Generalized Linear Mixed Model comparing forest areas (corrected for multiple comparisons, following sequential Holm-Bonferroni correction). Superscript letters are ordered in descending size order of the forest area category tinuous forest (Anabohazo) than we report, as our species-individual and Chao 1-individual curves had not yet plateaued at our point of statistical comparison (Fig. 2). This postulates that a high number of rare tree species are present within Anabohazo. Both curves had however plateaued for fragmented forest (Ankarafa), suggesting that forest fragmentation, and indeed fragment size (Fig. 3, Hill and Curran 2003), likely limits tree species diversity within a forest area (Almoussawi et al. 2020). The greater species diversity in continuous forest can be attributed to the presence of rare species (Fig. 2B), species sensitive to habitat degradation ) and shade tolerant species (Metzger 2000), which would otherwise be absent in fragmented forests. Fragment age is also likely to influence tree species diversity, as species diversity can be expected to be lower in older fragments (Ross et al. 2002;Goosem et al. 2016); as SIRNP forest fragment age data is not something that we were able to obtain either by historical GIS maps or discussions with local guides, this was something we could not investigate or control for in this study. Interestingly, within fragmented forest, we observed a higher tree species diversity in forest edge area than in core area. This observation contrasts to core-edge comparisons from the continuous forest (Figs. 2 and 3). Tree species diversity is likely to be higher in core areas of continuous forest due to the presence of species sensitive to edge-effects and degraded habitat . However, these species may be less frequent in fragmented forest, making them more difficult to detect, or absent entirely. The presence of tall, emergent trees within core areas of fragmented forests would also restrict the growth of edge-tolerant species (Budke et al. 2010), which would be more abundant at the forest edge; we observed shade-intolerant species such as Harungana madagascariensis and Strychnos spinosa almost exclusively within forest edge areas (Supplementary File 6). This, coupled with the presence of successional and exotic species (Waddell et al. 2020), may result in higher species diversity within edge areas of fragmented forests. Sensitive species will continue to decline as fragment size and core area decreases (Hill and Curran 2003;Laurance et al. 2018), and we observed diversity decline in correlation with decreasing fragment size here (Fig. 3). Whilst our results were not statistically significant despite high R values, this can be attributed to our small sample size (N = 10 fragments).

Tree size
Our results for tree size mirror those for species diversity; tree size and density was higher in continuous forest than in fragmented forest, and tree size was greater in core forest areas in comparison to edge areas (Table 1; Fig. 4). Similar to tree species diversity, fragmentation and edge-effects effect tree size individually, but there were strong positive relationships between both variables indicating that fragmentation effects drive edge-effects and vice-versa. These findings align with our hypothesis, and the findings of previous studies (Laurance et al. 2000(Laurance et al. , 2006Magnago et al. 2015;Almeida et al. 2019). Whilst we found fragmented forest and edge areas to contain some large, fast-growing species (e.g., Canarium madagascariensis, Ceiba pentandra, Mangifera indica), these species were also present within continuous and core forest, along with specialised, old-growth, emergent canopy species, contributing to the significantly higher mean tree height in continuous and core areas. Large, specialised species are often absent in fragmented, degraded, and edge areas of forests, where they are replaced by successional, photophilic and generalist species (Laurance et al. 2006), often with lower biomass (Razafindratsima et al. 2018). This could be an explanatory factor for lower mean tree size in fragmented forests and edge areas. Tree size in forest edge areas is often limited by factors that cause extensive damage to vegetation, such as microclimate, wind exposure, soil desecration, fire, and browsing animals that are unable to access core forest areas (Laurance and Curran 2008;Magnago et al. 2015;Almeida et al. 2019). Such edge-effects would be much more prominent within fragments with lower core areas, resulting in altered tree size in fragmented forests. Further, additional disturbance and anthropogenic factors that affect tree size, such as cutting by humans, can often be observed many hundreds of metres into a forest patch (Laurance 1991). Tree size would therefore remain unaffected only in areas of continuous forest far from edge boundaries, which explains lower tree diversity in fragments of all sizes.
We found tree size and density to vary among our sampled fragments, an observation mirroring previous studies in other areas of the tropics (Tripathi et al. 2010;Almeida et al. 2019). Whilst tree DBH and crown diameter correlated strongly with fragment size, core area, and shape (Fig. 5), tree height and density had a weak negative correlation with these variables. Our results suggest that fragment size and shape affect tree dendrometry, yet our  Table 1). Boxplot notches represent 95% confidence intervals. Dots represent individual trees Fig. 5 The relationship between forest fragment total area (red), forest fragment core area percentage (blue), and forest fragment shape index (green) with mean tree DBH (first row), mean tree height (second row), mean tree crown diameter (third row), and mean tree density (fourth row) per vegetation plot for 10 forest fragments within Ankarafa forest of the Sahamalaza-Iles Radama National Park, north west Madagascar. Relationships investigated with linear models (full results in Supplementary File 5). Shaded areas represent 95% confidence intervals Table 2 Summary (mean ± standard deviation (range)) of calculated Shannon-Weaver (SW) structural diversity indices for four dendrometry variables (mean tree DBH, mean tree height, mean tree crown diameter, mean tree density) within vegetation plots of core and edge forest areas of two forests of the Sahamalaza-Iles Radama National Park, north west Madagascar, with results of Multivariate Generalized Linear Mixed Models assessing the impacts of forest fragmentation, edge-effects, and their interaction with each other, on each tree structural diversity variable. Values with the same superscript letter within a row were not found to be significantly different from each other in a Generalized Linear Mixed Model comparing forest areas (corrected for multiple comparisons, following sequential Holm-Bonferroni correction

Tree structural diversity
Similar to tree species diversity and size, our results demonstrate that forest structural diversity is higher in continuous forest and forest core areas in comparison to fragmented forest and edge areas ( Table 2; Fig. 6). Whilst fragmentation and edge-effects had individual impacts on three of the four structural diversity variables each ( Table 2), fragmentation did not have an individual significant effect on tree height structural diversity, and edgeeffects did not significantly affect tree density structural diversity on an individual level. However, structural diversity of DBH and height appears to be synergistically affected by both fragmentation and edge-effects, whilst crown diameter and tree density appear are only impacted by each variable individually (i.e., fragmentation does not drive edge-core differences in crown diameter and tree density structural diversity, and vice-versa). The findings for DBH and height conform to our original hypothesis, and our findings for all variables align with trends observed in other geographic areas of Madagascar and the tropics (Benitez-Malvido 1998; Laurance et al. 2001;Wassie et al. 2010;Muluneh et al. 2021). Forest structural diversity also appears to have a general positive correlation with fragment size and core area size (with the exception of tree density SW diversity), and negatively with shape index (with the exception of DBH SW diversity) (Fig. 7, Supplementary File 5). These results provide evidence to suggest that high structural diversity can only be retained within intact, undisturbed forest core areas, most likely due to higher species diversity, the retention of both smaller understorey species and large, old-growth species, and a greater range of tree forms (Memiaghe et al. 2016;Ali et al. 2020). Structural diversity loss also appears relative to the scale and degree of forest fragmentation and habitat degradation (e.g., Didham and Lawton 1999; Brown and Gurevitch 2004;Echeverría et al. 2007;Hending et al. 2020). However, we must restate that our fragment characteristic-related LMs were mostly not statistically significant, despite high R values.
In addition to lower tree species diversity, forest structural diversity within fragmented forest and edge areas is also likely to be caused by changes in microclimate, wind exposure, soil degradation, fire, and other factors that damage tree structure and limit growth (Laurance and Curran 2008; Magnago et al. 2015;Almeida et al. 2019). Whilst lower structural diversity is indicative of habitat degradation within forest ecosystems (Milodowski et al. 2021), reduced tree structural diversity is also correlated positively with species diversity of resident animals and non-woody plants (Tews et al. 2004;Lenière and Houle 2006;Mert and Yalçınkaya 2017;Hending et al. 2020). Lower structural diversity, coupled with reduced tree species diversity and density, diminishes the availability of travel strata, feeding trees, and sleeping sites for forest-dwelling animals (e.g., Seiler et al. 2013), many of which are threatened and dependent on forest habitat for survival. Forest fragmentationinduced structural diversity loss is therefore of great conservation concern, particularly for sensitive species that are specialised to continuous, pristine forest core habitat (Augusto et al. 2000). Fortunately, fragments and edge areas are valuable habitat for many generalist, gap-tolerant species, and many forest flora and faunal are capable of thriving within both pristine and degraded forest areas (Fahrig et al. 2019). This is definitely true for numerous species within our own study area, and Madagascar more-generally (e.g., Hending 2021b).

Conservation implications and conclusion
This study has demonstrated that forest fragmentation and edge-effects are highly detrimental to tree species diversity, tree size, and structural diversity, and the impacts of fragmentation and edge-effects are often correlated, and therefore synergistic. Even core areas within larger forest fragments exhibit significantly reduced tree species diversity and structural diversity, and such trends are linked with reduced overall floral and faunal biodiversity, and reduced ecosystem functionality (Tews et al. 2004;Hertzog et al. 2019). Our findings provide further evidence that continued forest fragmentation may drive forest ecosystem collapse, and only core areas of unfragmented, continuous forests are able to retain normal levels of tree species diversity and structural diversity. Tropical forests remain one of the most important habitat types for global biodiversity (Wilson 1988;Bonan 2008), and substantial habitat restoration efforts need to be made to halt fragmentation and limit the loss of intact core forest. In many areas of the tropics, including in our study area (Cotton 2021;Hending et al. 2021b), reforestation efforts are already underway to restore areas of primary forest lost due to unsustainable slash-and-burn agriculture, charcoal production, and mining Fig. 6 Summary of Shannon-Weaver structural diversity indices (SW) for DBH, height, crown diameter, and tree density for core and edge areas of Anabohazo and Ankarafa forest in the Sahamalaza-Iles Radama National Park of north west Madagascar. Boxplots display the median, interquartile ranges, and minimum and maximum values. Boxplot notches represent 95% confidence intervals. Dots represent individual data points (plot structural diversity SW values) Fig. 7 The relationship between forest fragment total area (red), forest fragment core area percentage (blue), and forest fragment shape index (green) with mean Shannon-Weaver structural diversity indices (SW) for tree DBH (first row), tree height (second row), tree crown diameter (third row), and tree density (fourth row) per vegetation plot for 10 forest fragments within Ankarafa forest of the Sahamalaza-Iles Radama National Park, north west Madagascar. Relationships investigated with linear models (full results in Supplementary File 5). Shaded areas represent 95% confidence intervals (Chazdon 2008;Locatelli et al. 2015;Taubert et al. 2018). These efforts will be crucial to reconnect isolated forest patches, increase and regenerate core forest areas, and safeguard the future of core forest ecosystems (Chazdon 2008;Di Sacco et al. 2021).
Forests throughout Madagascar and the wider tropics continue to be fragmented at an alarming rate (Harper et al. 2007;Taubert et al. 2018;Vieilledent et al. 2018). The situation in Madagascar is of particular concern, as over 80% of the island's original forest cover has already been cleared, and 46% of remaining forest areas are located within only 100 m of a forest edge (Vieilledent et al. 2018). Coupled with the ongoing threats posed by agricultural expansion, unsustainable natural resource extraction, and climate change (Myers 1993;Foley et al. 2005;Hending et al. 2022), continued forest fragmentation renders tropical forests as arguably the most threatened of all natural habitat types globally, despite being the most important for terrestrial biodiversity and carbon sequestration (Duivenvoorden et al. 2002;Bonan 2008;Keith et al. 2009). This study provides further evidence that urgent conservation actions, and new environmental policy and legislation mandates need to be implemented to protect, restore, and renew areas of forest throughout the tropics, and to ensure the future survival of the threatened taxa that inhabit these vitally important habitats.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.