The study area comprised the Colombian Andean Páramo biogeographical province (2,732–5,212 m, Figure 1), including the high-elevation northern Andes and the Sierra Nevada de Santa Marta, but excluding areas with similar altitudes in the Cordillera de Merida and Central American mountains (Figure 2). It was delimited in the north by the Sierra Nevada of Santa Marta (11°N), and in the south by the Colombia-Ecuador border (1 N). It spanned 36 Páramo complexes, as in Morales et al. (2007) and Londoño et al. (2014), denoting 49% of the world’s Páramos. The surveyed Páramo zones were covered in a 33% by protected areas¹, and surrounded in a 30% by forest patches². Meanwhile, agriculture and mining, respectively, disturbed 6 and 7%³ of these Páramos and their buffer belt (2 km around the tree line). The handling of these data sources is detailed below as part of the computation of the “Adaptive Capacity at the Study Area.”

Climate sensitivity in the Espeletia complex was modeled by comparing current and future (2050 RCP 8.5) distributions under a niche conservatism scenario (i.e., range shift via migration). Occurrences of Espeletia samples for the 36 Colombian Páramo complexes were downloaded as georeferenced specimen data from providers served by GBIF⁴. From an initial set of 15,466 specimen records downloaded, a total of 1,432 (Supplementary Table S1), comprising 28 different Espeletia taxa according to Cuatrecasas (2013) – Supplementary Table S2, were retained after filtering for georeferencing consistency (i.e., non-redundant records within the study area) and number of records. In order to avoid bias due to distributions with low number of input points (Wisz et al., 2008), specially for presumably rare taxa (Fois et al., 2018), only putative species with at least 10 original records were retained (three taxa with more than 100 records, Supplementary Figure S1A). This threshold follows Van Proosdij et al. (2016) recommendation for narrow-ranged species, that is those with a moderate prevalence class (i.e., 27 taxa were present in less than seven different Páramo complexes, and only one taxon was present in 20 Páramos, Supplementary Figure S1B). This latter trend is expected in Espeletia species, for which distance (Diazgranados and Barber, 2017; Padilla-González et al., 2017) and ecology (Cortés et al., 2018a) are the major speciation drivers.

Historical climate data was extracted from the WorldClim⁵ database, at a 30 s resolution, using the georeferencing of each record. All 19 bioclimatic variables were downloaded in the baseline period 1970–2000. The same variables projected to 2050, under the IPCC RCP 8.5 scenario, were gathered using CCAFS’s downscaled delta method (Navarro-Racines et al., 2020). This projection averages 33 different global climatic models at a 1 km² resolution. A Digital Elevation Model (DEM) variable was further included. Prioritization of environmental variables (Zeng et al., 2016) is crucial to enhance model complexity (Warren and Seifert, 2011) for climate sensitivity assessments. By improving variable selection (Cobos et al., 2019), collinearity (Feng et al., 2019) and sampling bias (Syfert et al., 2013) can be minimized. Collinearity in particular refers to a strong correlation between two or more predictor variables, and may cause instability in parameter estimation (Dormann et al., 2013). A measure to diagnose collinearity is the Variance Inflation Factor (VIF), the square of the multiple correlation coefficient resulting from regressing a predictor variable against all other predictor variables (Chatterjee and Hadi, 2006). Hence, we performed variable selection by computing the VIF score using the usdm (Uncertainty Analysis for Species Distribution Models) package (Naimi et al., 2014) in R v.3.3.1 (R Core Team), and drew Pairwise Pearson correlations (r) among all variables by means of the same software (Supplementary Figure S2). Five bioclimatic variables and DEM did not exhibit collinearity (Naimi et al., 2014) at VIF < 10 (Supplementary Table S3), with absolute r-values below 0.7 from 0.05 to 0.65, and therefore were retained hereinafter.

The six non-collinear variables fed Maxent model v.3.4.1. (Phillips et al., 2017), a “machine learning” outline that uses maximum entropy and Bayesian inference to estimate occurrence probability distributions. Ten split-sample models were run, after checking for stabilization in the ROC curves (Spiers et al., 2018), with an average of 10 k pseudo-absences for each of the 28 putative taxa. Each replicated run type was set to “cross-validate” with 500 iterations (Buffum et al., 2015). Presence ≥ 95% was binary reclassified (Liu et al., 2005) and the average across repetitions was recorded in each case. This procedure was performed for the historical dataset (1970–2000). To predict distribution changes of Espeletia, we projected trained models on current presence localities and present background climate to 2050 under the RCP 8.5 scenario through Maxent’s “projection layer” tool. Current model validation considered AUC, an aggregate measure of performance across all possible classification thresholds. Minimum average testing AUC was above 0.98 (Supplementary Table S4). As AUC alone (Supplementary Table S5) may be deceiving (Lobo et al., 2008; West et al., 2016), the AIC and BIC criteria were also computed and checked for consistency (Supplementary Table S6).

All 56 distribution maps were converted to binary data, using a threshold of the 20th percentile, as suggested by the Maxent v.3.4.1 software (Phillips et al., 2017). These maps were then averaged to generate spatial information of current and future presence probabilities across all 28 taxa. Finally, the current distribution map was subtracted to the one obtained under a climate change scenario in order to estimate climate sensitivity in the Espeletia complex. Values close to zero in the sensitivity index meant few changes in the forecasted distribution of Espeletia due to climatic variation. On the other hand, negative and positive values corresponded to areas where the presence probabilities were, respectively, diminished or boosted by climate change.

We computed the adaptive capacity as the scaled additive effects of six main variables that expressed the capacity of the overall Espeletia biodiversity to respond without migration to the effects of climate change. Enhancing factors of the biodiversity’s adaptive potential were diversity, protected areas, and forest area around the Páramo. On the contrary, limiting factors of the plastic response were agriculture, mining, and population density.

Diversity may buffer the effects of climate change by enriching complex interactions, such as adaptive inter-specific introgression (Marques et al., 2019) and hybrid speciation (Coyne and Orr, 2004; Abbott et al., 2013; Payseur and Rieseberg, 2016). Hybridization may allow for local adaptation in transitional environments and bi-directional transfer of adaptive variation through introgression (Isabel et al., 2020). Espeletia in particular exhibits weak species boundaries, rampant gene-flow and reticulate evolution (Cortés et al., 2018a; Pouchon et al., 2018), leading to a high incidence of natural hybrids (Rauscher, 2002; Cuatrecasas, 2013). In order to account for the enhancing role of diversity on the adaptive potential, the number of Espeletia taxa was totalized in a 1 km grid using the occurrence data downloaded from GBIF (see footnote), as described in the previous section, and scaled from 0 to 1.

Another enhancing factor of the biodiversity’s adaptive potential is protected areas, where human intervention does not represent a direct threat, allowing biodiversity to be conserved while naturally coping with climate change. Protected areas selected for this analysis spanned 1,050 natural areas, according to the System of National Natural Parks (see footnote), and included 43 National Natural Parks, 663 Civil Society Reserves, 54 Regional Natural Parks, and 57 National Protected Forest Reserves. These categories were, respectively, rated as 1, 0.3, 0.1, and 0.1, following CIAT (2017). Higher ratings were assigned to areas that presumably provide greater protection for biodiversity due to their low anthropogenic pressure and extent.

Forest patches around the Páramo complexes (Henao-Díaz, 2019) may also contribute buffering climate change effects by facilitating (Bueno and Llambí, 2015) migration and gene-flow, which could increase the frequency of existing genetic variants adapted to particular conditions (Bridle and Vines, 2007). Additionally, forests provide conditions that may favor thermal and water regulation, decreasing the magnitude of climate change itself. Therefore, forest area in the buffer zone of the Páramo (2 km around the tree line) was gathered from the IDEAM (see footnote), and converted to binary points (1 = presence) in a 1 km grid.

On the other hand, major limiting factors of the plastic response are agriculture (Avellaneda-Torres et al., 2020), mining (Pérez-Escobar et al., 2018) and population density within the study area and in its buffer zone (Vásquez et al., 2015). In order to account for these threats, agricultural and mining areas were downloaded from the Geographic Institute Agustín Codazzi (see footnote) for the period 2010–2012, and transformed to binary data (0 = presence) in a 1 km grid within the study area and in its buffer zone (2 km around the tree line). The population density indicator was computed as the inverse value of the normalized rural population density, scaled from 0 to 1, so that 1 meant absence of human settlements and 0 was the population density at the most populated rural zones in the study area. The population information was gathered from data projected to 2015 by the National Administrative Department of Statistics⁶.

In order to estimate climate vulnerability at the overall Espeletia complex, we simultaneously coupled its climate sensitivity with its adaptive capacity. Negative values of the sensitivity index meant areas where the presence probabilities of Espeletia were diminished due to climate change. Meanwhile, lower adaptive capacity scores spoke for a prominent role of limiting factors of the plastic response (i.e., agriculture, mining, and population density) as compared to enhancing drivers of the biodiversity’s adaptive potential (i.e., diversity, protected areas, and forest area around the Páramo). Therefore, the climate vulnerability score was computed as the complement value of the additive contributions of the sensitivity and the adaptive capacity indices. The climate vulnerability score was scaled from 0 to 1 for comparative purposes. Pairwise Pearson correlations (r) among all estimated scores were computed, and boxplots and ridgeline plots were drawn for the three indices in all 36 Páramo complexes. Computing and plotting were done in R v.3.3.1 (R Core Team).

Climate sensitivity was estimated across all 36 Colombian Páramo complexes by comparing current (baseline period 1970–2000, Figure 2A) and future (2050 RCP 8.5, Figure 2B) average distribution probabilities of 28 different Espeletia taxa (nine had average sensitivity scores below zero). Values surrounding zero in the sensitivity index (−0.029 to 0.030, 42% of the study area, Figure 2C) stood for few changes in the forecasted distribution of Espeletia due to climate change, and were predominant in the north of the Eastern Cordillera, the Cruz Verde-Sumapaz Páramo complex south of the Bogotá montane savanna (the southwestern wetland complex of the larger Eastern Cordillera’s Andean highland plateau), and the small sky islands of the Western Cordillera.

Given a niche conservatism scenario, negative estimates of the climate sensitivity (≤−0.030, 32% of the study area, Figure 2C) predicted range losses based on the averaged presence probabilities. Páramo areas more likely to experience range losses were those in the Sierra Nevada de Santa Marta, and the Central Cordillera South of Las Hermosas (i.e., Nevado del Huila-Moras, and Guanacas-Puracé-Coconuco), as well as those in the northern corner of the Eastern Cordillera (i.e., Sierra Nevada del Cocuy, Pisba, and Tota-Bijagual-Mamapacha), where diverse summits surrounded the topographically intricate Chichamocha inter-Andean canyon. On the other hand, range shifts via migration enriched presence probabilities in other localities (positive estimates of the climate sensitivity ≥ 0.031, 27% of the study area, Figure 2C). Range gains were only predicted for Páramo areas in the Central Cordillera (Los Nevados, Chilí-Barragán, Las Hermosas, and La Cocha-Patascoy, Supplementary Figure S3) where the dominant taxon was E. hartwegiana, as well as in the summits of the northern tip of the Eastern Cordillera (Perijá).

The adaptive capacity across 36 Colombian Páramo complexes was computed using a scaled additive model of enhancing (Figures 3A–C) and limiting (Figures 3D–F) factors of the biodiversity’s adaptive potential and the plastic response. The adaptive capacity index (Figures 4B, 5A, Supplementary Figure S3A, and Supplementary Table S7) was higher (≥0.72, 34% of the study area) in species-rich Páramo areas (Pearson’s r = 0.19 ± 0.01, Figure 3A and Supplementary Figure S4), and matched extensive National Natural Parks (r = 0.59 ± 0.01, Figure 3B and Supplementary Figure S4), like the Eastern Cordillera’s Sierra Nevada del Cocuy, Chingaza and Sumapaz. The Páramo adaptive capacity index was also moderately superior (0.62–0.71, 29% of the study area, Figure 4B) in the most representative National Natural Parks in the Central Cordillera (Los Nevados and Nevado del Huila) as well as in the Sierra Nevada de Santa Marta. In turn, the western slopes of the Páramo complexes at the Eastern and Central Cordilleras exhibited lower adaptive capacity values (≤0.61, 37% of the study area), and were enclosed by less forest patches (30% on average, Figure 3C, r = 0.39 ± 0.01, Supplementary Figure S4).

However, positive effects due to protected areas and forest coverage on the adaptive potential were offset by limiting factors of the plastic response, primarily agricultural (6% of the study area and its buffer zone, Figure 3D, r = −0.47 ± 0.01, Supplementary Figure S4) and mining (7% of the study area and its buffer zone, Figure 3E, r = −0.41 ± 0.01, Supplementary Figure S4) pressures, mainly in the western slopes of the Páramo areas at the Eastern and Central Cordilleras. This let to the lowest adaptive potential estimates (≤0.52, 20% of the study area, Figure 4B) at the western slopes of the Páramo complexes in the Cauca province, in the Central Cordillera, and the Boyacá and Santander provinces (i.e., Almorzadero and Altiplano Cundiboyacense Páramo complexes), in the Eastern Cordillera. Rural population (Figure 3F) hampered the adaptive potential in all three Cordilleras and Páramo areas (r = −0.28 ± 0.01, Supplementary Figure S4).

The scaled complement of the additive contributions of the sensitivity (Figures 2C, 5B and Supplementary Figure S3B) and the adaptive capacity (Figure 4B) indices were used to compute the climate vulnerability score (Figures 4C, 5C and Supplementary Figure S3C). The most vulnerable (≥0.57, 11% of the study area, Figure 4C) Páramo areas coincided with diverse Páramos in the Eastern Cordillera (Figure 3A) because of more distribution losses (r = 0.39 ± 0.01, Supplementary Figure S4). Agriculture (r = 0.48 ± 0.01, Supplementary Figure S4), mining (r = 0.36 ± 0.01, Supplementary Figure S4) and rural population density (r = 0.23 ± 0.01, Supplementary Figure S4) also boosted their vulnerability, mainly on the western slopes in the middle of the Central Cordillera (i.e., Guanacas-Puracé-Coconuco), and the northeast corner of the Eastern Cordillera (i.e., Almorzadero, Pisba, and Altiplano Cundiboyacense). On the other hand, the less vulnerable (≤0.38, 57% of the study area, Figure 4C) Páramo areas were those with the highest adaptive capacity (≥ 0.72, Figure 4B, r = −0.93 ± 0.01, Supplementary Figure S4) due to large National Natural Parks (Sumapaz, Chingaza, Sierra Nevada del Cocuy, Figure 3B), as well as those for which predicted range shifts were neutral (−0.029 to 0.030, Figure 4A, i.e., the Cruz Verde-Sumapaz Páramo complexes in the Eastern Cordillera) or positive (≥0.031, Figure 5B, Los Nevados, Chilí-Barragán, and Las Hermosas in Central Cordillera).

Widespread range losses in the distribution of Espeletia are expected by 2050, a pessimistic prediction in line with previous modeling efforts (Buytaert et al., 2011; Crandall et al., 2013; Mavárez et al., 2018; Helmer et al., 2019; Peyre et al., 2020; Young and Duchicela, 2020). This tendency is broadly defined as thermophilization, a progressive decline of cold mountain habitats and their biota (Gottfried et al., 2012). Large reductions in modeled area and important upward shifts in the distribution of 28 Espeletiinae are already predicted by 2070 at the Venezuelan Páramo (Mavárez et al., 2018). Despite our projection is carried out for a shorter time frame at another, wider, study area (Colombian Andes spanning the Eastern, Central and Western Cordilleras besides Sierra Nevada de Santa Marta), overall results point toward the same trend and are complementary. Similar predictions of climate sensitivity at the Páramo complexes of the Ecuadorian Cordilleras (1 sp.) were not considered as part of this study because of the difficulties in compiling and comparing data from different national repositories, specifically for some of the variables that compose the adaptive capacity score (i.e., agriculture and mining pressures). Yet, considering analogous approaches in the Ecuadorian Páramo (Crandall et al., 2013; Helmer et al., 2019; Peyre et al., 2020) is essential to gather a more comprehensive understating on the consequences of climate change at high Andean ecosystems north of the Huancabamba depression, a major biogeographical barrier for high-elevation plant taxa (Weigend, 2002).

There are two important exceptions for the widespread range losses in the distribution of Espeletia in the Colombian Andes. First, small and fragmented Páramo sky islands in the Western Cordillera exhibit neutral climate sensitivity, a counterintuitive projection that Mavárez et al. (2018) also pointed out for Espeletiinae in isolated Páramo areas at Cordillera de Mérida. This reiterative finding suggests that scarce topographical complexity (terrain ruggedness) in recently colonized small islands (Flantua et al., 2019) may have favored habitat generalists, while intricate local-scale environmental heterogeneity at larger Páramo complexes could have triggered an explosive radiation (Cortés et al., 2018a; Naciri and Linder, 2020) of habitat specialists, each with a narrow ecological niche for migration to occur (Cuesta et al., 2019b). Modern colonization and limited topographical complexity, together with higher connectivity, may also account for the second exception, which is that range gains are only predicted for Páramo areas in the Central Cordillera (Los Nevados, Chilí-Barragán, Las Hermosas, and La Cocha-Patascoy), where the dominant taxon is E. hartwegiana. In line with this, Páramo areas more likely to exhibit range losses are those in the topographically complex Eastern Cordillera’s northeast corner, where diverse (Luteyn, 1999; Cuatrecasas, 2013; Peyre et al., 2015,2019) summits surround the deep (1,100 m) Chichamocha inter-Andean canyon.

Range shifts presuppose niche conservatism, but also depend on the intrinsic dispersal ability of the taxa (Pelayo et al., 2019; Tovar et al., 2020). In organisms with limited seed dispersal such as Espeletia, other ways for migration are key (Cochrane et al., 2019). For instance, migration via sporadic pollen flow would not be as constrained as seed dispersal (Cuatrecasas, 2013). Alternately, hybrids with intermediate niche requirements (Rieseberg, 2003) may serve as bridges for migration in a plant complex with rampant natural introgression (Rauscher, 2002; Cortés et al., 2018a; Pouchon et al., 2018). Finally, sporadic episodes of habitat connectivity (Flantua et al., 2019) in the lifespan of a long-lived organism such as Espeletia (Cuatrecasas, 2013) could provide a window for migration across modern barriers.

Key traits may relax the niche conservatism assumption by conferring the ability to colonize new habitats. For instance, underground plant-soil interactions (Goh et al., 2013; Sedlacek et al., 2014; Little et al., 2016) across different water levels (Geange et al., 2017) and cellular-anatomical physiological adaptations (Sandoval et al., 2019) may serve as pre-adaptations for a wider range expansion (Cuesta et al., 2017). Plastic traits may also broad the spectrum of novel climates (Arnold et al., 2019). Coupling this modeling with eco-physiological studies in Espeletia (Leon-Garcia and Lasso, 2019; Llambí and Rada, 2019) is hence a key research line to better track future range shifts in the Páramo.

The fact that diverse areas in the Eastern Cordillera are vulnerable to climate change is not only counterintuitive, but it also means that the long-term rapid diversification rate, striking in the Espeletia complex, could be jeopardized. Diversity is regarded as a buffer of the effects of climate change by enriching interactions (Cáceres et al., 2014), such as facilitation (Bueno and Llambí, 2015; Wheeler et al., 2015; Llambí et al., 2018; Mora et al., 2018; Venn et al., 2019), adaptive inter-specific introgression (Schilthuizen et al., 2004; Seehausen, 2004) and hybrid speciation (Payseur and Rieseberg, 2016), of which there is abundant evidence in Espeletia (Rauscher, 2002; Cuatrecasas, 2013; Cortés et al., 2018a; Pouchon et al., 2018). Yet, these positive effects are offset by distribution losses in sensitive Páramo localities, besides rising mining, agriculture and population density. These are Páramo’s current major anthropogenic threats (Vásquez et al., 2015; Pérez-Escobar et al., 2018) primarily in the western slopes of the Eastern and Central Cordilleras.

Permeable species barriers due to pervasive hybridization make the species concept highly debatable within the Espeletia complex. Therefore we refrained from emphasizing single-species responses that are likely to be overwhelmed by meta-population dynamics. Pouchon et al. (2018) suggested that the entire subtribe Espeletiinae requires an in deep taxonomic revision (Mavárez, 2019b) because most of Cuatrecasas’ (2013) species are thought to be paraphyletic. This trend is not unexpected since the available taxonomy greatly relies on potentially plastic morphological and reproductive traits (Mavárez, 2020) that are susceptible to be driven by environmental effects, besides the fact that it disregards rampant hybridization (Rauscher, 2002). Then, a more appropriate scenario for the young rapidly-evolving Espeletia complex with weak species boundaries would be that ecotypes in the same locality, despite depicting microhabitat divergence, are capable of sharing through gene flow and introgression adaptive genetic variants (Cortés et al., 2018a). Gene swapping because of porosities in the species bounds would imply that (1) the effective size of “rare” taxa with few records is higher in terms of standing adaptive variation, and (2) concerted climate responses is feasible across lineages within the same Páramo sky island. Inter-mountain exchange is also plausible, as reported in other tropical sky islands (Chala et al., 2020; Tusiime et al., 2020).

A possible way forward for populations to persist is the very same driver that is thought to contribute with the unbeatable diversification rate in the Páramo flora (Madriñán et al., 2013) – local-scale adaptive variation (Huang et al., 2020; Todesco et al., 2020). As part of this study we considered major enhancers and constrainers of the adaptive potential at a regional scale. However, populations may still harbor intrinsic genetic variation naturally selected to cope with environmental differences at the microhabitat level (Ramírez et al., 2014). In highly heterogeneous mountainous environments, local-scale environmental effects are known for shaping genetic diversity and generating morphological diversity (Hughes and Atchison, 2015), which may prove useful in the reaction to climate change within the same locality. There is already compiling evidence of ecological-driven divergence in the radiation of Andean Espeletia (Cortés et al., 2018a), besides allopatric differentiation and isolation by distance (Diazgranados and Barber, 2017; Padilla-González et al., 2017, 2018). Ecological parapatric speciation is a likely consequence of local patterns of environmental variation (Naciri and Linder, 2020) typically found at the Páramo ecosystem (Monasterio and Sarmiento, 1991). For instance, frost is more extreme in the wind-exposed high valley slopes than in the wind-sheltered depressions of those valleys or in the upper boundary of the cloud forest. Soil moisture content, higher in the well-irrigated high valley depressions and in the upper boundary of the cloud forest than in the drier slopes of the valleys, could also contribute to this divergence. It now remains to be explored whether this mosaic of environmental heterogeneity has enforced enough heritable polygenic (Barghi et al., 2020) phenotypic variation as a pre-adaptation (Nürk et al., 2018) to the new selective forces imposed by climate change.

One potential caveat of our results precisely concerns data availability at the micro-scale level. Bioclimatic data is reliably modeled throughout the Páramo range at a 1 km² spatial resolution, which unfortunately overlooks micro-environmental drivers of the local-scale genetic adaptation (De La Harpe et al., 2017; Leroy et al., 2020). This is of particular importance at mountain/alpine ecosystems (Ramírez et al., 2014; Cortés and Wheeler, 2018), where microhabitats may serve as refugia (Zellweger et al., 2020), like across the treeline due to active landform processes (Bueno and Llambí, 2015; Arzac et al., 2019; Gentili et al., 2020). Populations may rely on specific small-scale habitat attributes that are likely to be overlooked by SDMs (Sinclair et al., 2010), such as topographic, geomorphic, and edaphic features, as well as the distribution of other taxa – e.g., competitors and facilitators (Yackulc et al., 2015). Due to this, SDMs may exhibit substantial uncertainty (Buisson et al., 2010) and lack of validation (Botkin et al., 2007) at local scales. Therefore, niche models could be insensitive to micro-environmental processes, leading to an underestimation of the mechanisms governing populations’ responses. Even if bioclimatic models aim representing the realized niche, they may then disregard the fundamental niche (Loehle and Leblanc, 1996). Despite statistical imputation-like approaches such as environmental downscaling have been considered to model climate at a higher resolution for specific Páramo complexes (Mavárez et al., 2018), consistently applying such techniques across cordilleras seems unfeasible because accuracy would be constrained by the accessibility of fine-scale bioclimatic data from regional weather stations. These restraints are likely to be overcome soon by incorporating microclimate into SDMs (Lembrechts et al., 2019) through mechanistic algorithms (Lembrechts and Lenoir, 2020) physiological models (Cortés et al., 2013; López-Hernández and Cortés, 2019), hydrological equations (Rodríguez-Morales et al., 2019; Correa et al., 2020), in situ data logging, and remote sensing (Ramón-Reinozo et al., 2019; Zellweger et al., 2019).

Another potential caveat of our inferences relate with data availability across the entire Páramo area in the northern Andes. Some components of the adaptive capacity score (i.e., protected areas, agriculture, and mining pressures) are downloadable only through national repositories, making standardizations and comparisons across all Páramo complexes in different countries impractical. Yet, it is worth to clarify that climate sensitivity, adaptation capacity and vulnerability indices are not meant to be absolute scores but rather relative values to compare among Páramo complexes. In this sense, the power of our inferences to rank Páramo areas is unlikely limited by the pessimistic predictions of the future climate scenario RCP 8.5, making our vulnerability index robust. Furthermore, our whole integrative analytical pipeline is in line with the most up-to-date approaches to carry out climate change vulnerability assessment (Ramirez-Villegas et al., 2014; Valladares et al., 2014; Foden et al., 2018).

By assessing climate sensitivity and adaptive capacity across Páramo sky islands, we predicted climate change vulnerability in the rapidly evolving Espeletia complex, a syngameon (Cannon and Petit, 2020) likely shaped by incomplete ecological speciation (Nosil et al., 2009) and adaptive introgression (Fitzpatrick et al., 2015; Martin and Jiggins, 2017). However, in order to gather a more comprehensive view of climate change responses in the high mountain ecosystems of the northern Andes, other diverse and highly abundant plant genera in the Páramo (Hughes et al., 2013) should be considered under a similar analytical framework as the described here. Immediate candidates to follow up this approach are Bartsia (Uribe-Convers and Tank, 2015), Chusquea (Ely et al., 2019), Diplostephium (Vargas et al., 2017), Hypericum (Nürk et al., 2013), Loricaria (Kolar et al., 2016), Lupinus (Hughes and Eastwood, 2006; Vásquez et al., 2016; Contreras-Ortiz et al., 2018), Oreobolus (Chacón et al., 2006; Gómez-Gutiérrez et al., 2017), Puya (Jabaily and Sytsma, 2013), and Senecio (Duskova et al., 2017; Walter et al., 2020). These combined efforts would ultimately reveal whether the fastest evolving biodiversity hotspot on earth, and in general tropical high mountain ecosystems (Hedberg, 1964; Sklenáø et al., 2014; Chala et al., 2016), have a chance to persist under current environmental and anthropogenic threats.