We defined the composition of the coastalscape as the proportion of land cover classes (i.e., artificial surfaces, agricultural areas, forest and seminatural areas, wetlands, and water bodies) inside the circular buffers. We conducted a Permutational Analysis of Variance (PERMANOVA hereafter) to compare coastalscape composition of Pelliciera spp. between periods and basins, and between periods and Caribbean regions (Cartagena, Morrosquillo, and Urabá). We used two different land cover layers, comprising the first decade of the 21st century: 2000–2002, and 2010–2012. These layers represent the official benchmark for recent land cover dynamics, also matching the timeframe of most of the Pelliciera spp. records used in this analysis.

The extent of urban areas was remarkable in some coastalscapes within each of the three Caribbean regions (see Figure 1 and Supplementary Figure 2). A Local Urban Index (LUI) was computed to indicate an absolute difference among Caribbean regions (Table 2). As expected, Cartagena scored the highest LUI (most urban), due to the greatest road length, transformed area, and artificial surface per square kilometer. Contrary to the expected, Morrosquillo was the least urbanized region, at least within the coastalscapes of Pelliciera spp. In Morrosquillo, occurrences were only reported within or near marine protected areas (Cispatá Bay Regional District of Integrated Management and Boca de Guacamaya Natural Regional Park) but mangroves also occur in the proximity of two small populated centers (Tolú and Coveñas). Low urbanization was expected within the coastalscapes in Urabá, however, the presence of Pelliciera spp. in the vicinity of Turbo, a high-density populated center with a highly compact urbanization, offsets the wild coastalscape observed in the Atrato River Delta where additional presences were recorded (see Supplementary Figure 2). Therefore, the LUI was greater in Urabá than in Morrosquillo.

Prior to the analysis, we visually defined areas of maximum and minimum urban extent within each region of the Caribbean. The LUI was consistent with the preliminary classification of minimum and maximum urban extent, except in the Barú Peninsula, where low urbanization rather than high was observed. It was probably because only two small towns are present, and secondary roads are almost absent, while there is only a two-lane primary road connecting the area to Cartagena (boat transportation has been historically prevalent).

We also calculated a General Urban Index that describes the coastalscapes according to theoretical maximums and minimums of anthropogenic land covers in the coastalscape. It classified Cartagena, and Turbo as Highly Urban. Highly compact areas in the seaside within these cities may be comparable to other heavily urbanized areas in the Caribbean region. Coastalscapes in Morrosquillo were classified as Moderately Urban or Not Urban.

This study reports the gradient in human-domination along the coasts of northern South America. Our results support previous studies indicating that the Pacific coast of Colombia remains dominated by forested biomes with low extents of agricultural and pastoral lands (Etter et al., 2008; Sánchez-Cuervo et al., 2012). On the contrary, the Caribbean coastalscapes were dominated by extensive rangelands (despite shrimp aquaculture was also observed in the proximity to Cartagena, a pattern consistent with previous national-scale studies; Larsson et al., 1994; Etter et al., 2008; Sánchez-Cuervo et al., 2012). These findings are also consistent with the reports using the anthrome framework (Ellis and Ramankutty, 2008) and the human footprint index for Colombia (Correa-Ayram et al., 2020).

The lack of significant temporal variability in coastalscape composition in the study area suggests a reduction in the deforestation rates and, in cases, an increase in afforestation rates, as observed in Latin America during the first decade of the 21st century (Sánchez-Cuervo et al., 2012; Graesser et al., 2015). Specifically, mangrove deforestation rates worldwide have been slower since the 2000s (Friess et al., 2019; Goldberg et al., 2020). The lack of temporal variation in coastalscape composition may also suggest the reach of an advanced stage in the deforestation cycle where prime lands for agriculture and other types of exploitation have been already transformed and monopolized (Meyfroidt and Lambin, 2011). Besides, further stages in the economic development of different countries in Latin American and the Caribbean, have implied conversion of former agricultural areas into urban and suburban areas, but not a reduction in forest cover (e.g., Grau et al., 2003; López-Marrero et al., 2012).

A significant landmark of the Anthropocene in northern South America is the urban development in various spots along the Caribbean coast. While high scores of local and general urban indices were found in Cartagena and Urabá, additional urban settlements are present within the distribution range of Pelliciera spp., according to the present study. Moreover, if an urban index would be computed for the entire mangrove habitat regardless of the presence of vulnerable species, a south-north gradient would be also evident. On the contrary, there was no urbanization gradient along the Pacific and only two major port cities are found (Buenaventura and Tumaco) but immersed in coastalscapes dominated by forested lands (agriculture to a lower extent) consistent with previous reports (López-Angarita et al., 2018; Fagua et al., 2019). While the low degree of urbanization in the Pacific is an exception of the generality in the Tropics, the compact urbanization in Turbo (Urabá) and Cartagena is more alike to the wider Caribbean. Turbo is the typical case of a small and compact commercial port city that experienced rapid population growth between 1980 and 2000s (like others in the Antilles; Parés-Ramos et al., 2008; López-Marrero et al., 2012). Cartagena is a typical example of a coastal metropolis experiencing rapid urban sprawl into neighbor rural areas, similar to the observed elsewhere in the Caribbean, Brazil, West Africa, and Southeast Asia (Martinuzzi et al., 2009; Nfotabong-Atheull et al., 2013; Friess et al., 2019). Morrosquillo is an example of low density and dispersed urban development (Tolú and Coveñas) found in tourism districts in the Antilles and the Caribbean basin of Central America and Mexico (e.g., Hirales-Cota et al., 2010; Díaz-Gallegos et al., 2011).

Therefore, the spatial template offered by urbanization in the Caribbean basin of South America provides an opportunity for studying the ecological and evolutionary consequences of urbanization on mangrove ecosystems. Despite the biogeographic differences between the Pacific and the Caribbean basins, it would be interesting to study the ecological consequences of specific forms of human domination on the coastalscapes. We recommend focal studies on urban and peri-urban mangroves (e.g., Buenaventura and Tumaco) and coastalscape analyses in the South Pacific where rapid landscape transformation occurred in coastal watersheds since the 2000s as the consequence of armed conflicts and proliferation of illicit cash crops (Quintero-Angel et al., 2021). Finally, we also recommend the use of innovative remote sensing techniques such as city lights and LIDAR for high-resolution analyses of urban sprawl into coastal wetland areas, particularly in Cartagena. The suburbanization and compaction processes in Cartagena seem to be similar to those of large cities in the northern Andes and the Caribbean, where the application of such techniques has been instrumental for improving scientific understanding, nature conservation, and urban planning (Parés-Ramos et al., 2013; Álvarez-Berríos et al., 2013; Martinuzzi et al., 2018; Branoff and Martinuzzi, 2020).

Mangrove configuration, at 5-km radius coastalscapes, was different between basins. The Caribbean exhibited higher fragmentation with a greater fractal dimension of patches and lower median patch size, mangrove area, and core area, as previously reported for terrestrial forests and mangroves worldwide (Taubert et al., 2018; Bryan-Brown et al., 2020). The differences in mangrove configuration related to the degree of human domination in the coastalscapes and could represent differences in the rates and drivers of deforestation in peripheral areas (as in Etter et al., 2006). A greater mangrove area reflects the wild condition of the Pacific coastalscapes and their low deforestation rates (Hamilton and Friess, 2018; López-Angarita et al., 2018; Mejía-Rentería et al., 2018; Simard et al., 2019). Therefore, this study suggests that urbanization translates into a patchy configuration of the remaining mangrove areas.

However, we still need to better understand the patterns derived from the fragmentation process through multi-temporal analyses and high-resolution mapping efforts, which are urgently needed. These efforts would benefit from employing open-access satellite imagery, cloud-computing (e.g., Bhargava et al., 2020), or the available open access multitemporal global layers (e.g., Thomas et al., 2017; Bunting et al., 2018; Bryan-Brown et al., 2020; Richards et al., 2020). Efforts must concentrate around protected, urban and peri-urban areas, as well as on to assess the effectiveness of conservation efforts and to identify threats and deforestation hotspots.

In the Caribbean, the urban intensity was related to higher fragmentation, a pattern previously reported by Blanco-Libreros and Estrada-Urrea (2015) in the Urabá Gulf. The loss rate related to proximity to an urban center and patch density, while patch density and mean patch area were related to human disturbance (Blanco-Libreros and Estrada-Urrea, 2015). Increased patch density, due to trampling and logging, also promoted species shifts at the community level (Blanco et al., 2012; Blanco-Libreros and Estrada-Urrea, 2015). Small-scale deforestation and logging have been reported in the proximity of human settlements in many rural and peri-urban locations world-wide (e.g., Blanco et al., 2012; Nfotabong-Atheull et al., 2013; Palacios and Cantera, 2017; Scales and Friess, 2019). Thus, we recommend field surveys of Pelliciera spp. in different areas along the distribution range, but particularly comparing areas of maximum and minimum urbanization, for understanding the effects of patch geometry and mangrove habitat configuration on demographic variables. We also recommend using the protected versus non-protected contrast to achieve this objective and to further understand the ecological service of habitat provision to these and other species of plants and animals relative to patch size. Finally, we need to better understand why differences in mangrove configuration were not observed at smaller spatial scales.

Mangroves are the most threatened ecosystems across tropical coastalscapes, with the highest urban population density per habitat area, and this trend will seemingly continue until 2050 (McDonald et al., 2013). Urbanization has the potential to alter the quantity, quality, and spatio-temporal arrangement of resources for mangrove plant species and populations, similarly to the observed in other systems (Parris, 2016; Dubois and Cheptou, 2017). A main consequence of the urbanization process is the simultaneous decrease in habitat amount and the increase in habitat fragmentation over time (Liu et al., 2016). Fragmentation usually provides a better descriptor of trait variation than urbanization per se (Dubois and Cheptou, 2017). The main direct effects of fragmentation are the creation of smaller patches and the increased isolation of patches (sensu Wilcove et al., 1986).

In small forest patches, there is less habitat and resources, while more area is susceptible to edge effects, thus supporting smaller populations than large patches (Gascon et al., 2000; Laurance et al., 2002; Hobbs and Yates, 2003). In addition, small mangrove populations are more vulnerable to local extinction caused by environmental and demographic stochastic events and are less likely to be rescued if they are very isolated in areas of high cyclonic activity, such as the Caribbean region (Krauss and Osland, 2020). Besides, increased fragmentation can reduce the survival rate of seedlings, the production, and quality of fruits, and pollinator abundance, even in naturally-patchy habitats such as mangroves and in species such as Pelliciera spp. interacting with winged pollinators (birds, moths, and/or bats; Hermansen et al., 2014, 2017; Duke, 2020). This can result in reduced gene diversity and higher genetic structure in mangrove plant species (Arnaud-Haond et al., 2006). In addition, small Pelliciera populations in the Caribbean (presumably P. benthamii) have low intrapopulation genetic diversity and they are probably affected by genetic drift, inbreeding, and bottlenecks (Castillo-Cárdenas et al., 2012). Genetic drift can reduce the ability of populations to cope with novel environments and small populations can also experience fitness reductions as population size declines (Allee effects; Gilpin and Soule, 1986; Keller and Waller, 2002). Fitness reductions from drift, inbreeding depression, and Allee effects can reduce population size, creating a dangerous feedback loop (Young et al., 1996; Cheptou and Avendaño, 2006; Cheptou et al., 2017).

We propose that mangrove patch shrinking and isolation, due to fragmentation, will interact with environmental changes thus contributing to lower effective population size, affecting birth and death rates, as well as immigration and emigration, for northern South America. The different pathways of interaction are illustrated in our theoretical framework shown in Figure 4. The conceptual framework is partly based on Hobbs and Yates (2003), Parris (2016), and Cheptou et al. (2017), and in a review of the literature on urban mangroves (see Supplementary Table 6 for details). In our study, we used Mean Patch Size (AREA_MN) and Patch Cohesion Index (COHESION) as spatial proxies of decreased patch size and increased patch isolation, respectively. We used Core Area Percentage of Landscape (CPLAND), Mangrove (Class) Area (CA), and Perimeter-Area Fractal Dimension (PAFRAC) as proxies of greater area susceptible to edge effects. We propose that Pelliciera spp. populations in the Caribbean (particularly the recently reported P. benthamii) are at greater risk of extinction due to their low genetic diversity (Castillo-Cárdenas et al., 2012), the small number of propagules that survive each year (Dangremond, 2015), and the high amount of area susceptible to edge effects (smaller total mangrove area, smaller core area, and more irregular patches when compared to the coastalscapes along the Pacific). For populations in Cartagena, we predict the highest susceptibility induced by the smallest core area (related to edge effects) observed across the study area, and their location in the limit of the distribution range (likely having lower gene flow; Eckert et al., 2008; Castillo-Cárdenas et al., 2012).

The study area covered the Pacific and Caribbean coasts of Colombia (1°23′29.4″N – 12°27′30.2″N, 71°07′18.6″W – 79°00′31.8″W), the area of distribution of Pelliciera spp. (Planchon and Triana Silva, 1863) in northern South America (Figure 1). Presence records of Pelliciera spp. were collected from the Colombian National Mangrove Assessment (HELIO_SP.CO v1; Blanco-Libreros and Álvarez-León, 2018), the Global Biodiversity Information Facility (GBIF.org, 2020), technical reports, regional experts, and scientific papers (For a detailed description of the occurrence data used and sources, see Supplementary Table 1). The presence records were georeferenced using QGIS Desktop 3.10.7. Only one record per square kilometer was chosen randomly to reduce spatial autocorrelation. A total of 107 records of Pelliciera spp. were compiled: 82 in the Pacific and 25 in the Caribbean.

Since the Caribbean basin exhibits a greater human-domination in the coastalscapes (Álvarez-León and Polanía-Vorenberg, 1996; Álvarez-León, 2003), the occurrences of Pelliciera spp. were pooled into three Caribbean regions (Urabá Gulf, Morrosquillo Gulf and two bays in the vicinity of Cartagena City). Within the three regions of the Caribbean, areas of maximum and minimum urban extension were defined to quantitatively evaluate urbanization as a driver of mangrove change, as proposed by Branoff (2020).

The composition of the coastalscapes was calculated using National Land Cover maps (Scale 1:100,000) for two different periods 2000–2002 and 2010–2012 (IDEAM, 2010b, 2014; available in http://www.siac.gov.co/catalogo-de-mapas). These vector layers are based on the Corine Land Cover Methodology adapted for Colombia and they included five land cover classes: (1) artificial surfaces (including ponds for marine and continental aquaculture), (2) agricultural areas, (3) forest and seminatural areas, (4) wetlands and (5) water bodies.

The coastalscapes were defined as circular buffers of 1 km radius, centered on the presence points of the species. The land cover layers were rasterized using a pixel size of 60 m × 60 m. R 4.0.2 (R Core Team, 2020) and the landscapemetrics package (Hesselbarth et al., 2019) were used to calculate the absolute and relative extent (CA and PLAND) of Forest and seminatural areas as proxies of potential habitat for Pelliciera spp., and Artificial surfaces and Agricultural areas as proxies of anthropogenic drivers of loss and degradation. The wetlands category was not used as a proxy of potential habitat for Pelliciera spp. because the floristic composition of wetlands comprises Pleustophyta, Rizophyta, and Haptophyta but not true mangrove plant species, according to the layer authors (IDEAM, 2010a).

A PERMANOVA (Anderson, 2001; McArdle and Anderson, 2001) was performed to test for differences in the composition of the Pelliciera coastalscapes between basins (the Caribbean versus Pacific) or among Caribbean regions (Urabá, Morrosquillo, and Cartagena). In addition, years (2000 versus 2010) were nested within basins or regions, respectively. Pairwise comparisons were employed when significant differences were detected. PERMDISP was used to detect differences in multivariate dispersions among groups. Bray-Curtis similarity index and Spearman rank correlation were used as proximity metrics, and the analysis was conducted in PERMANOVA+ for PRIMER v7 (Anderson et al., 2008; Clarke and Gorley, 2015).

Spatial datasets used for quantification of urban variables and the urban indexes are described in the Supplementary Table 3. All spatial analyses were performed in QGIS Desktop 3.10.7 using Group Stats plugin (QGIS Development Team, 2020; Szostok et al., 2020). The urban index proposed by Branoff (2020) was used to calculate a Local Urban Index (LUI) and a General Urban Index (GUI) for the three regions in the Caribbean (Cartagena, Morrosquillo and Urabá), and in the areas of maximum and minimum urban extent within each one. The urban indices are a representation of the relative intensity of urbanization, in which 100 is the most urbanized site and 1 is the least urbanized site. The index was calculated using the following equation:

where n is the number of variables used in the index and Yi represents the variables normalized to a range of 0 to 100 through the following equation in which Y and X represent the normalized and raw values, respectively:

For the LUI, we included road density, percentage of transformed land cover and artificial surfaces in the coastalscapes. The variables were normalized according to the distribution of the data. For the GUI, we excluded road density, and the variables were normalized according to an ideal distribution with a maximum of 100 and a minimum of 0. We expected to have a measurement of urbanization allowing comparison in the degree of urbanization of the Pelliciera spp. coastalscapes in our study area, with those in other urbanized coastalscapes within the range of occurrence. The regions were also classified into categories based on their quartiles of the urban index. For the LUI, the “Most Urban” sites were those with an urban index greater than the third quartile, “Urban” sites were those in the interquartile range and the “Least Urban” sites were those with an urban index lower than the first quartile. For the GUI, the “Highly Urban” sites were those with an urban index greater than the third quartile (more than 75), “Urban” sites were those in the interquartile range (between 25 and 75) and the “Hardly urban or Non-urban” sites were those with an urban index lower than the first quartile (less than 25).

The coastalscapes were defined as circular buffers of 500-m, 1-km, or 5-km radius to assess mangrove habitat configuration for Pelliciera spp. A pixel size of 35 × 35 m was used to calculate fragmentation metrics at the mangrove class, using the national mangrove layer. This layer was mapped from 2005 to 2009, with scales ranging from 1: 10,000 to 1: 50,000 (INVEMAR, 2009; available in https://gis.invemar.org.co/arcgis/rest/services/SIGMA/MANGLARES_COLOMBIA/MapServer).

All spatial and statistical analyses were conducted using R version 4.0.2 (R Core Team, 2020) using rgeos, rgdal, raster and landscapemetrics (Hesselbarth et al., 2019; Bivand and Rundel, 2020; Robert, 2020). PAFRAC, ENN_MN and AREA_MN were calculated, following Bryan-Brown et al. (2020), using Queen’s case contiguity when needed. CA, COHESION and CPLAND (with edges of 100 m and 500 m) were also calculated. Differences in configuration between Pacific and Caribbean basins were explored through Mann-Whitney test with wilcox.test function in R. In testing differences between Most Urban, Urban, and Least urban sites in the Caribbean, defined by the Local Urban Index, the Kruskal-Wallis rank sum test was used through the kruskal.test function of base R. The subsequent post-hoc differences were identified through Multiple Comparison Test After Kruskal-Wallis by the kruskalmc function of pgirmess package (Giraudoux, 2018).