The flickering connectivity system of the north Andean páramos

2.1 Geographical features

The Northern Andes (ca. 448,000 km²) covers parts of Venezuela, Colombia and Ecuador (Figure 2a), and can be partitioned into six principal mountain ranges or ‘cordilleras’ (Figure 2c), namely the Sierra Nevada de Santa Marta (SNSM), Cordillera de Mérida, Eastern, Central and Western Cordillera and the Ecuadorian Cordilleras. Most of the Northern Andes is considered a highly to extremely high rugged landscape (Figure 2b; See mountain illustrations by Von Humboldt, 1845) where the high peaks and deep inter-Andean valleys cause strong contrasts in climate throughout the region (Flantua et al., 2016). Surface area in mountains does not decrease monotonically with elevation as has been shown previously in southern Colombia by Flantua et al. (2014) and on a global scale by Elsen and Tingley (2015). The Northern Andes shows a decrease of surface area going upslope where there is a slight peak around 900–1,200 m above sea level (a.s.l.) but then continues to decrease up to 6,260 m a.s.l. (Figure 2d), following a typical ‘pyramid shape’. However, the different cordilleras show different patterns of elevational surface area (Figure 2d) where the Eastern Cordillera shows a sharp peak around 2,600 m a.s.l. and the Ecuadorian Cordillera shows high values of surface area at much higher elevations than the other cordilleras (for more details see Table S1.1, Appendix S1 in Supporting Information). The páramos today are spread out over the Northern Andes as a ‘mountain archipelago’ of small and highly fragmented páramo complexes (See Figs S2.1, S2.2 for more details and photos of different páramo complexes) but their full range also cover isolated páramo islands in Costa Rica and northern Peru (Luteyn, 1999). Of all tropical alpine floras, such as in East Africa and New Guinea, the páramos are home to the highest species richness and endemism (Luteyn, 1999; Sklenář et al., 2014), with low between-mountain similarity in species (Sklenář et al., 2014). They also provide numerous ecosystem services on a local and regional scale (Herzog, Martínez, Jørgensen, Tiesse, 2011) and references therein), and especially in terms of hydrological services, they are vital for the provision of fresh water to several large cities in South America, such as Bogotá, Medellín, Quito, Cuenca, Piura and Cajamarca.

2.2 Quantifying temperature and upper forest line based on fossil pollen data

To quantify temperature fluctuations during the Pleistocene (and consequently estimate páramo connectivity), we used fossil pollen data from the Northern Andes. The composite pollen record Funza09 (4.83°N, 75.2°W; 2,550 m a.s.l., Fig. S2.1. Red star) reveals vegetation and climate dynamics over the past 2.25 Myr (Torres et al., 2013). We reconstructed the páramos’ elevational fluctuations, and consequently páramo connectivity, by estimating the upper forest line (UFL; the transition from the upper montane forest to the páramos) from the Funza09 record. Though this record covers the last 2.25 Myr, we only used the last 1 Myr as this interval reflects continuous lake conditions in comparison with variable hydrological conditions between 2.2 and 1 million years ago (Ma) which makes a quantification of changes to the UFL less precise. We follow the methodology described and implemented by Hooghiemstra (1984), Groot et al. (2011) and Hooghiemstra et al. (2012) to derive the Andean UFL and palaeotemperature curve (for detailed methodology on the UFL reconstruction see Appendix S3).

2.3 Calculations of connectivity per páramo ‘island’

To calculate the degree of connectivity between páramos, we used a graph-based habitat availability index called probability of connectivity (PC) metric. This metric takes into account the area of the páramo ‘island’ itself and the distances to other islands where a user-defined distance threshold defines the ‘reachability’ of other islands (Saura, Estreguil, Mouton, & Rodríguez-Freire, 2011; Saura & Pascual-Hortal, 2007), even if they are not physically connected (i.e. ‘structural connectivity’, Tischendorf & Fahrig, 2000). The metric assigns a value to each páramo island representing its contribution in maintaining the overall connectivity of the páramo biome (Saura & Pascual-Hortal, 2007; Saura et al., 2011). The total PC is built up in three ‘fractions’, namely the ‘intrapatch’, the ‘flux’ and the ‘connector’ fractions (Saura & Rubio, 2010). The first fraction focuses on the available surface area and habitat quality (if applicable) within the individual island. The second fraction assesses how well the individual island is connected to other islands given additional importance to the other islands’ attributes (surface and quality) and its strategic position to other páramo islands. The third fraction quantifies the contribution of the island to maintain connectivity between the rest of the islands, in other words its role as an intermediate stepping stone between non-adjacent islands. Additionally, we calculated the equivalent connected area (ECA), which is derived directly from the PC, as a measure of the overall connectivity of a region (Saura et al., 2011). Here, Conefor Sensinode (V2.2; Saura & Pascual-Hortal, 2007; Saura & Torné, 2009) and ESRI ArcGIS 10.3 (ESRI, 2014) were used to calculate the straight-line distances between islands, the PC and ECA. We calculated connectivity for the entire Northern Andes and for each mountain range separately.

2.4 Calculations of corridors between páramo islands

We identified corridors between páramo islands within and between cordilleras under different climatic conditions. We used the Gnarly Landscape Utilities (V0.1.3; McRae, Shirk, & Platt, 2013) with ESRI ArcGIS 10.3 to create a raster grid of ‘landscape resistance’ based on ruggedness (Figure 2b) and habitat suitability. We assumed an increased landscape resistance with increased ruggedness, assigning values between 0 (no resistance) to 100 (maximum resistance) using an equal interval classification. For the habitat suitability map, we started by assigning a ‘perfectly suitable’ score of 100 to each páramo island, while outside the island the score of 0 reflects maximum unsuitability. To soften this boundary, an exponential decay function was then used by increasing resistance in five elevational steps of 100 m where we assigned a suitability score of 40 to the boundary of the páramo. As a result of the decay function the highest suitability of páramo – its core area – was restrained 200 m above the UFL and 200 m below the snowline.

We used Linkage mapper to calculate the least-cost pathways, or corridors, based on the produced raster grid of landscape resistance (McRae & Kavanagh, 2011). These corridors are expressed as ‘conductance maps’ that represent gradients of cumulative corridors. Where the densities of corridors is highest, it is assumed that there is a high probability of dispersal and migration possible between islands (McRae, Dickson, Keitt, & Shah, 2008). The full landscape of the Northern Andes is considered an area where corridors could exist, with exception of the region between SNSM and the Sierra de Perijá (Fig. S2.1).

We resampled the 30 m Digital Elevation Model (DEM, Figure 2) to a 1 km resolution to reduce computing time for each Linkage mapper down to on average 2 hr. We allowed Linkage mapper to create corridors through (instead of only between) core areas to represent the full arsenal of connectivity through the landscape. Only corridors between páramo islands larger than 1 km² were considered at any given moment in time. From the final output maps, only values lower than 200k conductance (default threshold) are selected to highlight the strongest corridors. The outputs were weighted according to the percentage of time they occurred during the last 1 Myr.

3.1 A million years of temperature fluctuations

Temperatures at Funza (2,550 m a.s.l.) are estimated to have fluctuated between ca. 15 and 6°C causing an estimated maximum 1,600 m elevational shift of the UFL between ca. 3,500 and ca. 1,900 m a.s.l. (Figure 3). The Pleistocene glacial-interglacial dynamics were not replicated cycles of temperature change showing repeated patterns of high and lows, but display a high temporal variability between each glacial-interglacial cycle. Conditions similar to the current warm, interglacial conditions occurred several times during the last 1 Myr and accounted for around a quarter of the time. Extreme cool glacial conditions, ~6–8°C cooler than today, were relatively rare, occurring less than 10 percent of the time. On the whole, intermediate cool stadials and mild interstadials dominated the last 1 Myr, occurring over two-thirds of the time.

3.2 Calculations of páramo connectivity

Our estimations on the spatial and elevational extent of ancient páramos and their connectedness at different times in the past reveals that páramos underwent frequent spatial alterations between fragmented and connected spatial configurations, but the exact patterns were highly dependent on mountain chain topography (Figure 4a,b. See Appendices S4 and S5). The páramos in the Ecuadorian Cordillera generally maintained a high degree of connectivity over the last 1 Myr, rarely enduring severe fragmentation. Fragmentation did however occur when the snowline plunged significantly during colder and wetter glacial periods, causing a break up of páramo areas on lateral flanks of the mountains. Likewise, the level of connectivity between páramos on the Central Cordillera fragmented substantially through a descending snowline, breaking the upper elevation limit of páramo connectivity. In contrast, the Eastern Cordillera shifted substantially between periods of connectivity and fragmentation, always, however, maintaining two large páramo islands surrounded by smaller ‘satellite islands’. Páramos in the Cordillera de Mérida seem to have been restricted during interglacials to one core area only, while during colder periods a relatively high fragmentation is observed possibly due to glaciers pushing páramos to lateral distributions. Here, connectivity increased mainly towards the southwest and during colder periods (UFL ≤ 2,300 m a.s.l.). The páramos of the SNSM and the Western Cordillera endured the highest degree of rates of change in fragmentation of all ranges. In the latter, páramo habitats are estimated to have often completely disappeared. In contrast, páramos of the Central Cordillera maintained a long latitudinal distribution, forming a chain of isolated populations in small patches that on the whole remained connected. Even in very cold conditions, no continuous connectivity of core areas seems to have been possible between the Eastern Cordillera and Cordillera de Mérida, or the region of Sierra de Perijá. Towards the south of the Eastern Cordillera a low-elevation barrier was possibly crossed at 1,900 m a.s.l. forming a brief bridge suitable for páramo habitat into the Macizo Colombiano of the Central Cordillera.

The reconstruction of putative corridors shows a complex spatial pattern through the mountainous landscapes of the Northern Andes (Figure 5a,b). The long ridge of the Central Cordillera forms the starting point of numerous corridors to the páramos in the Western Cordillera. The Eastern Cordillera shows a complex internal pattern of corridors, where there are neither strong corridors towards Sierra de Perijá in the North, nor towards the Cordillera de Mérida, while a high concentration of corridors is found between the large páramos complexes in the Eastern Cordillera (Páramos of Boyacá and Cundinamarca, Fig. S2.1). In the Ecuadorian Cordillera a more lateral pattern of high/low potential corridors is observed following the intra-Andean valleys and peaks within this mountain range. Corridors to the southernmost páramos of Ecuador as also the northernmost páramos of the Western Cordillera are weak and occurred infrequent during the last million years, shown by the thin lines.

3.3 Flickering connectivity systems

Páramo connectivity through time shows a highly variable pattern (Figure 6a) introduced by Flantua and Hooghiemstra (2018) as a flickering connectivity system (see visualization video in Appendix S6). We find support for the hypothesis that this system with fluctuating, highly variable connectivity in spatial and temporal dimension is unique for each mountain range of the Northern Andes (Figure 1). For example, changes in connectivity within the Ecuadorian Cordillera are substantial but the system ‘flickers’ around a high average when compared to other mountain ranges. The flickering connectivity systems within the Eastern and Central Cordillera are surprisingly similar, though the peaks of connectivity during glacial periods and the dips of connectivity during interglacials are more extreme in the former (Figure 6a). The Western Cordillera is a larger mountain range than the Cordillera of Mérida and the SNSM (Table S1.1), and its variation of connectivity has been correspondingly larger (Figure 6b) but with the lowest occurrence of connectivity compared to the other mountain ranges (Figure 6a). Considering only the frequency in the distribution of data (Figure 6b), the Ecuadorian Cordillera and the SNSM stand out for their relatively small within-mountain range variation in connectivity, compared to the Eastern and Central Cordillera (similar patterns) and the Western Cordillera.

When frequencies of connectivity are weighted by the amount of time that connectivity lasted two main patterns emerge (Figure 6c). The first is shared by the Western, Central and Eastern Cordilleras, which all display an elongated pattern where the highest values are around a centroid, resembling a ‘humming top’ or, as Elsen and Tingley (2015) recognized in mountain hypsographies, a ‘diamond’ shape. Ecuadorian Cordilleras, Cordillera de Mérida and SNSM instead reveal a different pattern with a narrower centroid that widens towards the upper and lower section, resembling an ‘hourglass’ shape. Here, the Ecuadorian Cordillera and SNSM show a surprising similarity though at different connectivity ranges. The Central and Eastern Cordilleras are strikingly similar overall.

4.1 Variable degrees of past connectivity of different mountain ranges

Although currently isolated, evolutionary radiations and the assembly of the páramo ecosystem formed during times when the páramos were flickering in and out of different degrees of connectivity (Figure 6). The concept of ‘mountain fingerprints’ (Flantua & Hooghiemstra, 2018) proposes that the region's complex topography would have meant that páramos in different mountain regions would have fragmented and connected at different periods of time, at different elevations, and with different rates and frequencies (as summarized in Figure 1). This means that in some mountain ranges the páramos are a mix of somewhat even occurrence of connectivity and fragmentation events through time (Figure 1b, exemplified by the Eastern Cordillera), or could have been dominantly fragmented (Figure 1a, e.g. Western Cordillera), or more connected (Figure 1c, e.g. Ecuadorian Cordilleras). These regional differences in the temporal and spatial variation in past páramo connectivity (Figures 4-6) are likely to have resulted not only in regional differences in biogeographical patterns through time, but also varying ecological and evolutionary processes. We therefore propose that our data and models can be used to test hypotheses of the drivers of species richness, endemism and degrees of Pleistocene diversification in the Northern Andes, and the approach applicable to other mountain regions around the world.

4.2 Evolutionary implications of the flickering connectivity system

Several insightful schematic representations of Pleistocene diversification models in the Neotropics have been developed in recent years (Flantua & Hooghiemstra, 2018; Hazzi, Moreno, Ortiz-Movliav, & Palacio, 2018; Ramírez-Barahona & Eguiarte, 2013; Rull, 2005). Phylogeographical and phylogenetic synthesis work within and among páramo taxa is currently still largely lacking (see for instance Yu et al., 2019 for the Qinghai-Tibet Plateau), inhibiting the direct testing of these models. However, here we highlight several recent studies that are considered valuable in the light of the flickering connectivity system reconstruction (see Table 1), emphasizing the expectation that the rapidly growing body of phylogeographical/phylogenetic literature in the region will support future comparative analyses.

The dynamic history of the páramos elucidated by the flickering connectivity system can provide three important insights in terms of evolutionary processes. First of all, the regional differences in past páramo connectivity – the mountain fingerprint – support temporally and spatially discordant phylogeographical patterns (Massatti & Knowles, 2014; Papadopoulou & Knowles, 2015, 2016; Pennington et al., 2010). This means that the timing of diversification in the different mountain regions would not be expected to have occurred synchronously, even if all phylogenetic studies on páramo species could overcome current issues in model inference, taxonomy and distribution, spatial resolution and time-calibration points (Rull, 2011). Secondly, diversification rates might differ along the elevational gradient and this might be the rule rather than the exception. Elevational differences in surface availability and connectivity (Bertuzzo et al., 2016; Flantua & Hooghiemstra, 2017; Flantua et al., 2014) are likely to influence at what elevation the strongest geographical processes will occur, and these processes are thus expected to differ between mountain systems resulting in elevational differences of diversification (see e.g. Hughes & Eastwood, 2006; Kropf, Kadereit, & Comes, 2003; Lagomarsino, Condamine, Antonelli, Mulch, & Davis, 2016; Quintero & Jetz, 2018). Furthermore, the climate fluctuations of the Pleistocene caused connectivity to occur at different moments through time (Figure 1), a process facilitating the step-wise but irregular migration of páramo plant species throughout the landscape, such as Puya, Loricaria and Espeletiinae (Table 1). Thirdly, the flickering connectivity system, which is expected to cause phases of increased isolation followed by increased connectivity of populations, is expected to result in pulses of diversification (Knowles, 2000), possibly resulting in series of sub-radiations in the páramos. Where isolation resulted in allopatric, in situ speciation, connectivity triggered diversification through dispersal and settlement in new areas (‘dispersification’, Moore & Donoghue, 2007), and hybridization of previously isolated populations (Grant, 2014; Petit et al., 2003). Much evidence suggests that hybridization is not the processes of species becoming ‘reabsorbed’ into their parental forms but contributes by bringing evolutionary novelty and gene flow operating at different introgression rates (Dušková et al., 2017; Nevado, Contreras-Ortiz, Hughes, & Filatov, 2018; Pouchon et al., 2018), and thus a likely trigger of speciation and morphological diversity. Interestingly, population-level processes such as gene flow, dispersification and hybridization, alongside periods of isolation, have been increasingly recognized to play out at the phylogenetic scale, leading to (rapid) lineage diversification, for example in mountains (e.g. Hazzi et al., 2018; Knowles & Massatti, 2017), tropical rain forests (e.g. Onstein et al., 2017) and islands (e.g. Ali & Aitchison, 2014). Interestingly, the Funza09 pollen record shows a clear increase in the amplitude of climate change around the mid-Pleistocene transition (ca. 0.9 Ma) coinciding with accelerated diversification of high elevation birds (Weir, 2006) and the Espeletiinae in the Cordillera de Mérida (Pouchon et al., 2018; Table 1). Indeed, these studies signal a potential link between the intensity of the flickering connectivity system and biological radiations (Flantua & Hooghiemstra, 2018). Thus, the flickering connectivity system is expected to have left an imprint on geographical patterning of genetic divergence (between populations) and within-populations genetic diversity with obvious inter-cordillera differences. Furthermore, extinction events may further complicate the observed patterns of divergence between cordilleras.

4.3 Future research

Our spatio-temporal estimates of past connectivity lay a foundation for further research on elucidating the causal mechanisms of mountain diversifications (see also Appendix S7). Models of past connectivity (Figures 4-6), when combined with phylogeographic data, could help reveal the role of interspecific gene flow and allopatric speciation in driving radiations in the high Andes and contribute to a better understanding of the relative importance of geography versus adaptive radiation that underpin Andean diversifications. In such a complex system it may also be useful to pay attention to commonalities. For example, when considering both frequency and duration, our data show that two connectivity patterns emerge (i.e. hourglass versus non-hourglass; Figure 6c). Research could explore if cordilleras with shared connectivity patterns also share phylogenetic histories and contemporary (endemic) species’ biogeographies to test for universal mechanisms that have shaped present day alpine biomes. This would be especially useful if used in conjunction with information on the reproductive life histories, growth and dispersal capacities of specific taxa.

Finally, past patterns of connectivity are critical to interpret biogeographical patterns of currently isolated or fragmented systems in a wide variety of terrestrial ecosystems including mountains (Flantua & Hooghiemstra, 2018), islands (e.g. Simpson, 1974; Weigelt, Steinbauer, Cabral, & Kreft, 2016; Norder et al., 2018), fresh water systems (e.g. Dias et al., 2014), rain forests (e.g. Graham, Moritz, & Williams, 2006), grasslands (e.g. Lindborg & Eriksson, 2004; Münzbergová et al., 2013) and marine coastal ecosystems (Hoeksema, 2007) that similarly experienced major spatial changes during rapid sea-level fluctuations over the Pleistocene. The approach developed here, to quantify historical connectivity, can therefore provide a platform for interpreting contemporary biogeographies and past drivers of diversification in a wide array of both marine and terrestrial ecosystems where available space has been altered by climatic fluctuations. We postulate that quantifying flickering connectivity systems will facilitate a much more detailed and much needed quantitative basis to compare phylogeographic/phylogenetic patterns, e.g. the Tibeto-Himalayan region (Muellner-Riehl, 2019), and species (endemic) richness (e.g., Sklenář et al., 2014), from different mountain regions of the world.