Keywords

Introduction

Conservation in the twenty-first century is predicted to be dominated by mitigation of the effects of man-made climate changes and the ramifications to natural systems and the species that depend on them (Bonan 2013; Laurance and Williamson 2001; Lewis 2006; Thomas et al. 2004; van Aalst 2006). Some changes are already being observed in temperatures, precipitation levels, cloud formation, and secondary impacts such as changes in plant phenologies that are predicted to have drastic consequences for ecosystems and species (Bertin 2008; Dore 2005; Lenoir et al. 2008; McCarty 2001; van Aalst 2006; Walther et al. 2002). Of the world’s major biomes, tropical montane forests will be one of the most severely affected (Bubb et al. 2004; Foster 2001; Herzog 2011; Still et al. 1999), with many more localized climate changes seen in air temperatures, cloud formation, and cloud capture (Pielke et al. 2002). This will affect many primates and other species that have restricted distributions or specialized habitat requirements (Newbold et al. 2014). Currently, all but one primate species listed as Critically Endangered on the IUCN Redlist of Threatened Species have restricted distributions (IUCN 2013), 26 of which have distributions in Montane habitats, five of which are entirely restricted to montane and pre-montane areas (Shanee 2013).

Another major concern for conservationists is the continued expansion of agricultural frontiers to support a growing human population and its demand for food and other resources (Fearnside 1983; Garland 1995; Newbold et al. 2014; Perz et al. 2005; Sanchez-Cuervo and Aide 2013; Wyman and Stein 2009). As much of the worlds suitable lands have already been converted to agricultural production, new frontiers are opening up in areas less suited to clearance, including montane ecosystems (Cayuela et al. 2006; Hall et al. 2009). The clearance of montane areas for agriculture works in tandem with climate changes to intensify local scale effects, increasing air temperatures, which in turn slows cloud formation and lowers precipitation levels, slowing forest regeneration. Heavier downpours increase erosion on slopes, further limiting forest regeneration, lowering soil fertility, necessitating the clearance of more areas for cultivation, resulting in a dangerous cycle (Laurance and Williamson 2001; Pielke et al. 2002).

The montane and pre-montane forests of northern Peru lie at the heart of the Tropical Andes Biodiversity Hotspot and are among the most threatened forested areas in the world (Myers et al. 2000; Robles Gil et al. 2004). Peru’s northern regions of Amazonas and San Martin suffer from the highest immigration and deforestation rates in the country (INEI 2008; PROCLIM/CONAM 2005; Reategui and Martinez 2007) accounting for approximately 18 % of Amazonian forest loss in Peru in the year 2000 (INRENA 2005). The tropical Andes are home to incredible levels of biodiversity with ~30,000 vascular plant species, 50 % of which are endemic, and the highest number of vertebrate species of any “Biodiversity Hotspot” (Myers et al. 2000). This includes 584 species and 69 genera of endemic birds. Diversity and endemism of mammals is similarly high with at least 75 species and five monotypic genera endemic to the area (Myers 2003; Myers et al. 2000).

Peru’s cloud forests account for only 5 % of the country’s 700,000 km2 tropical forests (Bubb et al. 2004) but species diversity is comparable to that of the much more extensive eastern Amazonian lowlands (Pacheco et al. 2009). In particular, the area between the Marañón and Huallaga rivers, ~8000 km2, has very high levels of endemism but is also severely threatened by logging, slash and burn and industrialized agriculture (Schjellerup 2000; Shanee 2012a), subsistence and commercial hunting (Shanee 2012b), and the cultivation of illicit crops such as coca ( Erythroxylum coca ) and opium poppies ( Papaver somniferum ). The production of these illicit crops is a double threat to the environment. The production causes habitat loss and contamination whilst the measures employed to control them, such as defoliant sprays and forced crop clearances, not only remove the crops but also affect neighboring forests and increase deforestation when producers relocate to new areas (Dourojeanni 1989; Fjeldså et al. 1999, 2005; Young 1996).

Three of Peru’s eight endemic primates (Boubli et al. 2012; Alfaro et al. 2012a, b; Marsh et al. 2013; Matauschek et al. 2011; Wilson et al. 2013) have distributions restricted to the Marañón-Huallaga landscape (Shanee et al. 2013b). The yellow-tailed woolly monkey (Lagothrix flavicauda), the San Martin titi monkey (Plecturocebus oenanthe), and the Peruvian night monkey (Aotus miconax) are all considered threatened by the IUCN (2013), with L. flavicauda and P. oenanthe considered Critically Endangered and A. miconax considered Endangered. Until recently, very little was known about these species, but recent conservation-based research programs have provided basic data on the distribution and ecology of these species (For example, Bóveda-Penalba et al. 2009; Buckingham and Shanee 2009; deLuycker 2007a, b; Shanee 2011; Shanee et al. 2011, 2013a, 2015). This information has been vital in providing a better understanding of threats, conservation need, and population trends for these species. The very restricted distributions of the three species are probably a result of the high levels of habitat heterogeneity in the area which is almost completely surrounded by the Maranon and Huallaga river valley’s (Fig. 1) divided by isolated areas of dry forests, high mountain ridges, and deep river valleys, impeding dispersion into new areas and creating localized bioclimatic zones (Shanee et al. 2013b, c, 2015).

Fig. 1
figure 1

Map of the study area showing major landmarks and the Maranon and Huallaga river systems and clipped area used in modeling, with inset showing location of study area in Peru

Full size image

Computer-based modeling of species distributions and ecological niches has become popular in recent years with better access to more powerful processors and dedicated software (Bocedi et al. 2014; Brown 2014; de Souza Muñoz et al. 2011; Goodchild et al. 1996; Guo and Liu 2010; Phillips et al. 2006; Skidmore 2004; Thuiller et al. 2009). Computer-based modeling is particularly useful when field surveys are made difficult by the physical impediments of the terrain or sociopolitical factors limiting researchers’ access to some areas within a species distribution. Many different modeling techniques exist, each with its own advantages and disadvantages (Elith and Graham 2009; Elith et al. 2006; Guisan et al. 2007; Guo and Liu 2010; Thuiller et al. 2009), in recent years ecological niche modeling with Maxent program (Phillips et al. 2006) has proven to be a robust presence-only modeling technique that balances accuracy with limitations on data availability, time, and model complexity. In addition, many tools and recommendation on how best to use Maxent for modeling have been developed to further robustness of models (Brown 2014; Warren et al. 2010). Similarly, computer modeling provides the best option for predicting the effects of future climate changes. These predictions are constantly being refined with many models now freely available to researchers (Hijmans et al. 2005; Kriticos et al. 2012).

For this chapter, I aim to model the possible effects of future climate changes on the distributions of three of Peru’s endemic and most endangered primate species. Building on this, I will model the effect of various simple thresholds as proxies to simulate expansion of the agricultural frontier and hunting pressure . This is done to highlight the challenges and opportunities climate changes may present for conservation for these species. Also, to examine the utility of GIS-based predicative modeling, balancing complexity with robustness.

Methods

Species and Distributions

Lagothrix flavicauda, Plecturocebus oenanthe, and Aotus miconax are endemic to a small area of northern Peru in the departments of Amazonas and San Martin (Shanee et al. 2015; Aquino and Encarnación 1994; Bóveda-Penalba et al. 2009; Shanee 2011). Lagothrix flavicauda and A. miconax are sympatric throughout the majority of their distributions (Fig. 2a, c) on the eastern slopes of the Andes in a thin band of montane cloud forest between approx. S78° 12′30″ and S75°24′55″ at altitudes ranging from 1500 to 2800 m above sea level, (MSL) although in some areas they are found at slightly higher or lower altitudes (Allgas et al. 2014; Campbell 2011; Shanee et al. 2013b, c). Plecturocebus oenanthe is restricted to the pre-montane area of the Rio Mayo valley south to the west of the Rio Huallaga as far as the Rio Huyallabamba (Fig. 2b) in lowland terra firme and tropical dry forests at elevations between 200 and 1200 MSL (Bóveda-Penalba et al. 2009). This species has been also been reported in small areas outside of these boundaries (Bóveda-Penalba et al. 2009; Shanee et al. 2013b; Vermeer et al. 2011).

Fig. 2
figure 2figure 2figure 2

Map showing the estimated distributions of the focal species in northern Peru, (a) A. miconax, (b) P. oenanthe, and (c) L. flavicauda, with insets showing location of study area. Distribution maps adapted from Rowe and Myers (2012)

Full size image

Data Collection and Preparation

I used point data for each species from confirmed presence localities in previously published studies and my own fieldwork (Shanee et al. 2015; Bóveda-Penalba et al. 2009; Shanee 2011). This gave a total of 48 points for Lagothrix flavicauda, 110 points for Plecturocebus oenanthe, and 73 points for Aotus miconax. To remove spatially non-independent localities and clusters of points, I spatially rarefied occurrence data using three natural breaks between 5 and 25 km2. For bioclimatic variables layers, I used freely available data sets from Worldclim (Hijmans et al. 2005). This provided me with 19 bioclimatic layers representing different environmental variables (Table 1). All layers were clipped to the bounds of a polygon layer of Peru’s national borders. I then carried out a principle component analysis (PCA) of climate heterogeneity, variable layers that showed high levels of homogeneity were then removed from subsequent analyses. I then made bias files for selection of locations for background and pseudo-absence points for use in predictions to limit errors of commission (e.g., overprediction of the model) (Anderson and Raza 2010; Phillips et al. 2009). Bias files were created using a minimum convex polygon (MCP) buffered to 200 km outside of sample presence points. I then carried out a spatial jackknife to evaluate which model performed best and used this for final modeling (Boria et al. 2014; Radosavljevic and Anderson 2014; Shcheglovitova and Anderson 2013).

Table 1 Data sets used in analyses
Full size table

For future predictions, I downloaded four sets of layers representing the same bioclimatic variables produced by the International Panel on Climate Change fifth assessment (Raper 2012; Rogelj et al. 2012). I chose the models from the NASA Goddard space institute, these layers are freely available from Worldclim (Hijmans et al. 2005). The future bioclimatic layers represented predictions of conditions under different greenhouse gas representative concentration pathways (RCP) in different years (Moss et al. 2010): RCP = 26 and 85 for 2050 and 2070. These layers had the same spatial resolution as the layers used in the initial analysis and were also clipped within the bounds of the Peru polygon. The final model, based on the results of spatial jackknifing, was then projected onto these climate scenarios. As standard, ROC (Receiver Operating Characteristic) curve and AUC (Area Under Curve) were used as measures of the predictive power and fit of the models (Peterson et al. 2007; Merckx et al. 2011).

Final Modeling

Even using bias files to limit the possible extent of niche predictions the Maxent models can over predict possible niche area when compared to known geographical barriers limiting the species’ distributions, often including areas outside of a species’ actual or historical distribution. To rectify this all model outputs for A. miconax and L. flavicauda were clipped to areas within the bounds of a polygon, representing the area between the Maranon and Huallaga rivers (Fig. 1). Similarly, outputs for P. oenanthe were clipped within a polygon representing the area between the Maranon and Huallaga rivers north of the Huyllabamba River and south of the eastern cordillera that forms the margin of the Mayo River Valley (Fig. 1). I then divided model predictions into ten equally sized classes representing different probability levels of species presence, the two lowest classes (0–9.9 and 10–19.9 %) were then removed to reduce errors of commision. The remaining eight classes were then divided into two subclasses representing two levels of probability (Good and Very Good). I then calculated the area of each subclass and overall as a measure of original habitat extension for each species (Table 2). To calculate the current area of occupancy of each species, I overlaid a forest cover layer (Hansen et al. 2013) to the outputs, removing areas with less than 50 % forest cover from predictions as areas without suitable habitat (Shanee et al. 2015; Wyman et al. 2011) and calculated the area in km2 of each subclass and overall to estimate the current distribution of the species. I repeated this for the four future climate scenario predictions. To predict the effect of future expansion of the agricultural frontier and the effect of hunting on the species, I created two thresholds of moderate and high hunting pressure/possibility of deforestation within the 35- and 55-year time frames used in the climate change analysis. Thresholds were set at areas <1 km (high pressure) and <5 km (moderate pressure) away from human settlement and infrastructure for high and moderate pressure, respectively. Additionally, I remodeled these thresholds including only habitat outside of protected areas to see if the current protected area system will be sufficient to support viable populations of the three species taking into account possible changes in niche occurrence with future climate changes.

Table 2 Results of ecological niche modeling and future threat analyses, all results are in km2
Full size table

Results

After multi-distance spatial rarefying of the original species occurrence points (48 for Lagothrix flavicauda, 110 for Plecturocebus oenanthe , and 73 points for Aotus miconax) to remove spatially non-independent localities and clusters, the number of data points for L. flavicauda, P. oenanthe, and A. miconax used in subsequent analyses were 34, 45, and 39, respectively. The PCA for climatic heterogeneity showed high autocorrelation of variables in 11 of the 19 bioclimatic layers; therefore, I used only eight in the final model, Annual Mean Temperature, Mean Diurnal Range, Isothermality, Temperature Seasonality, Annual Precipitation, Precipitation Seasonality, Precipitation of Warmest Quarter, and Precipitation of Coldest Quarter.

Model Results

The results from models projected onto the future bioclimatic layers showed no significant differences between predictions for years or RCP levels (All p > 0.001); therefore, results presented here are averages across the four different year/RCP combinations for each species.

Aotus miconax

The final ecological niche model for Aotus miconax gave an ROC curve AUC of 0.913 for training data. Minimum training presence was 0.473. Results of the jackknife test showed the environmental variable with highest gain when used alone was annual precipitation. The environmental variable that decreased gain the most when omitted was annual mean temperature.

When clipped to within known geographical boundaries and excluding cells in the lowest two probability levels, the total original possible extent of occurrence of A. miconax was ~37,220 km2. After reclassification into two subclasses representing Good and Very Good probabilities of species presence (20–59.9 and 60–100 %), estimated original niche sizes were 22,640 and 14,580 km2 for each subclass. After deforested areas were removed from these predications (areas <50 % forest cover) the current maximum possible extent of occurrence is ~29,990 km2 of which 17,930 km2 was classed as Good and 12,060 km2 was classed as Very Good. Future climate changes are predicted to reduce niche availability for A. miconax by a further 9 %, the most affected probability subclass is predicted Very Good category with a further 16 % loss. Including the 1 and 5 km future deforestation/hunting pressure buffers there is a 16 and 53 % loss of niche availability. This is reduced to 15 and 44 % for the two respective buffer thresholds when assuming no future habitat loss or hunting within protected areas.

Plecturocebus oenanthe

The final ecological niche model for Callicebus oenanthe gave an ROC curve AUC of 0.951 for training data. Minimum training presence was 0.370. Results of the jackknife test showed the environmental variable with highest gain when used alone was precipitation of the coldest quarter. The environmental variable that decreased gain the most when omitted was annual mean temperature.

When clipped to within known geographical boundaries and excluding cells in the lowest two probability levels, the total original possible extent of occurrence of C. oenanthe was ~6,992 km2. After reclassification into two subclasses representing Good and Very Good probabilities of species presence (20–59.9 and 60–100 %), estimated original niche sizes were 4,335 and 2,657 km2 for each subclass. After deforested areas were removed from these predications (areas <50 % forest cover), the current maximum possible extent of occurrence is ~5,547 km2 of which 3,628 km2 was classed as Good and only 1,919 km2 was classed as Very Good. Future climate changes are predicted to increase niche availability for C. oenanthe by almost 24 %, the largest increase is predicted to be in the Very Good category with an increase of over 100 % in niche availability. Including the 1 and 5 km future deforestation/hunting pressure buffers there is a loss of 6 and 72 % of niche availability. This was reduced to 26 and 50 % for each respective buffer threshold when assuming no future habitat loss or hunting within protected areas.

Lagothrix flavicauda

The final ecological niche model for Lagothrix flavicauda gave an ROC curve AUC of 0.910 for training data. Minimum training presence was 0.387. Results of the jackknife test showed the environmental variable with highest gain when used alone was precipitation of the warmest quarter. The environmental variable that decreased gain the most when omitted was precipitation seasonality.

When clipped to within known geographical boundaries, and excluding cells in the lowest two probability levels, the total original possible extent of occurrence of L. flavicauda was ~57,910 km2. After reclassification into two subclasses representing Good and Very Good probabilities of species presence (20–59.9 and 60–100 %), estimated original niche sizes were 37,150 and 20,760 km2 for each subclass. After deforested areas were removed from these predications (areas <50 % forest cover), the current maximum possible extent of occurrence is ~39,060 km2 of which 22,460 km2 was classed as Good and 16,600 km2 was classed as Very Good. Future climate changes are predicted to reduce niche availability for L. flavicauda by a further 7 %, the most affected of the probability subclasses is predicted Good category with a further 18 % loss, the Very Good category is predicted to increase by 7 %. Including the 1 and 5 km future deforestation/hunting pressure buffers there is an additional predicted 16 and 54 % loss of niche availability. This is reduced to 12 and 46 % for each respective buffer threshold when assuming no future habitat loss or hunting within protected areas.

Discussion

The areas of the original ecological niches modeled here for A. miconax and L. flavicauda are similar to those from previous GIS-based studies (Shanee et al. 2015; Buckingham and Shanee 2009). The largest difference found was in the estimated original niche size for P. oenanthe, which is much smaller than previous studies have estimated (Ayres and Clutton-Brock 1992; Shanee et al. 2011). Similarly, future climate changes are predicted to reduce the available niche area for A. miconax and L. flavicauda, whereas niche area for C. oenanthe is actually predicted to increase with future climate changes, even when taking into account expansion of the agricultural frontier. Actual levels of habitat loss for all three species are estimated here to be much lower than previous predictions (Buckingham and Shanee 2009; Shanee et al. 2011). The use of a 50 % forest cover threshold for species habitat does not include the effect of hunting pressure, which is high for all three species, particularly L. flavicauda (Shanee 2012b), nor does it take into account the effect of habitat fragmentation on the species’ dispersal ability. Using the <1 and <5 km thresholds may give a truer picture of actual presence of species, as many available areas which have the correct bioclimatic conditions may not currently hold populations of these species.

As with all modeling, the predictions presented here are only as good as the data available. I am confident that I have used the most complete data sets for species presence points, including results from several recently published exhaustive field studies (Shanee et al. 2013b, 2015; Bóveda-Penalba et al. 2009; Shanee 2011). By using only published data and localities from my own recent field surveys and those of researchers whose methods are known, I have avoided problems of unreliability of data downloaded from internet databases, museum collections, and other sources where accuracy of species data points is uncertain (Chan et al. 2011; Graham et al. 2008).

The resolution of data layers used in modeling effect the robustness of results, with finer resolutions generally producing better results (Vale et al. 2014). The bioclimatic data sets I used have a resolution of ~1 km which allow for the models to include all but micro-scale gradients in niche presence (Elith and Graham 2009). Comparing the Maxent outputs and the distribution maps for A. miconax and L. flavicauda given by Rowe and Myers (2012) (Fig. 2a, c), this limitation can be seen clearly in central Amazonas, where the species are not present (Shanee et al. 2015; Shanee 2011) but the correct bioclimatic conditions exist (Figs. 3, 4, and 5). Even so, when the deforestation layer was applied to models the corresponding area is largely removed from the resulting distribution predictions. Scale is another factor that can influence applicability of models (Guisan and Thuiller 2005; Suárez-Seoane et al. 2014). By using bias files to limit gain and results of commission models were improved (Guisan and Thuiller 2005). I corrected problems of over prediction by clipping model outputs to known geographic barriers. Other problems in accuracy of modeling can occur from spatial autocorrelation of point data, inflating measures of accuracy (Veloz 2009), and spatial heterogeneity of bioclimatic layers. By carrying out a PCA of climate variables and spatially rarefying locality data, I was able to limit the possible effect of these problems on model results (Boria et al. 2014; Veloz 2009).

Fig. 3
figure 3figure 3figure 3figure 3

(a) Prediction of the original ecological niche area of A. miconax, (b) Current habitat availability for A. miconax, original niche area minus current deforestation, (c) Predicted future habitat availability for A. miconax based on modeling results, with areas of current deforestation removed, and (d) Predicted future habitat availability for A. miconax, including 1 and 5 km thresholds of predicted deforestation and hunting

Full size image
Fig. 4
figure 4figure 4figure 4figure 4

(a) Prediction of the original ecological niche area of P. oenanthe, (b) Current habitat availability for P. oenanthe, original niche area minus current deforestation, (c) Predicted future habitat availability for P. oenanthe based on modeling results, with areas of current deforestation removed, and (d) Predicted future habitat availability for P. oenanthe, including 1 and 5 km thresholds of predicted deforestation and hunting

Full size image
Fig. 5
figure 5figure 5figure 5figure 5

(a) Prediction of the original ecological niche area of L. flavicauda, (b) Current habitat availability for L. flavicauda, original niche area minus current deforestation, (c) Predicted future habitat availability for L. flavicauda based on modeling results, with areas of current deforestation removed, and (d) Predicted future habitat availability for L. flavicauda, including 1 and 5 km thresholds of predicted deforestation and hunting

Full size image

As expected future climate changes are predicted to reduce niche availability for A. miconax and L. flavicauda. This is because of general and localized changes in temperatures, precipitation levels, and cloud formations, all of which will in turn have drastic consequences on plant phenologies affecting habitat availability and quality (Bubb et al. 2004; Foster 2001; Herzog 2011; Pielke et al. 2002; Still et al. 1999). Interestingly, my models predicted a large increase in niche availability for P. oenanthe with future climate changes. These very different results could stem from the different habitat requirements of the species. Aotus miconax and L. flavicauda are restricted to higher elevation montane forests, which are predicted to be very sensitive to climate changes (Bubb et al. 2004; Herzog 2011; Still et al. 1999), whereas C. oenanthe is restricted to lower elevation pre-montane and tropical dry forests. Increased temperatures and reduced precipitation may account for the predicted increase in niche availability, particularly in dry forest areas. These differences in the predicted effects of future climate changes on niche availability for these species demonstrates the complexities involved in modeling such changes (Newbold et al. 2014). Caution needs to be used when interpreting this result as the increase in area is mainly outside of the species current distribution. In this case the species, or habitat, may not be able to adapt quickly enough to the geographic shift in niche location (Feeley and Silman 2010), and this could therefore constitute a significant decrease in the actual niche availability. When I applied the two thresholds of predicted future land use change and anthropogenic hunting pressure, the models all predicted reductions in niche availability for all species.

Natural inaccessibility and socioeconomic instability played major roles in protecting A. miconax and L. flavicauda, and to a lesser extent P. oenanthe, from anthropogenic pressures for many years (deLuycker 2007b; Ellenbogen 1999; Kent 1993; Shanee 2011; Young 1996). Since the paving of the main highway from Peru’s Pacific coast to the Amazonian lowlands, increased immigration from the high mountain sierras of Peru’s interior has caused widespread deforestation and substantial increases in hunting rates (deLuycker 2007b; Dreyfus 1999; Morales 1986; Shanee 2012a). From remote, unsettled regions, this area now has the highest immigration and deforestation rates in Peru (INEI 2008; PROCLIM/CONAM 2005; Reategui and Martinez 2007). This has caused severe fragmentation of habitat for all three species (Shanee et al. 2011, 2015; Bóveda-Penalba et al. 2009; Leo Luna 1987; Shanee 2011). As in all areas, deforestation, fragmentation, and the presence of livestock and waste products have many negative impacts on populations of wildlife (Newbold et al. 2014) including, increased competition for resources (Andrén 1994; Estrada and Coates-Estrada 1996), increased hunting pressure (Jerozolimski and Peres 2003; Michalski and Peres 2005; Peres 2001), increased zoonotic infections (Chapman et al. 2006; Fahrig 2003; Gillespie et al. 2005; Goldberg et al. 2008; Sanchez-Larranega and Shanee 2012), and reduced connectivity between populations, reducing genetic fitness (Bergl et al. 2008; Brenneman et al. 2012; Marsh et al. 2013).

Protected area networks have been a mainstay of conservation for many years but have been criticized for shortfalls in effectiveness in protecting species (Cantú-Salazar et al. 2013; Geldmann et al. 2013; Rodrigues et al. 2003; Seiferling et al. 2012), increasing the need for landscape level solutions that include local communities in gap areas (Gálvez et al. 2013; Porter-Bolland et al. 2012). Community-managed forests provide a solution for conservation in highly populated areas and often perform better then protected areas (Porter-Bolland et al. 2012). The inclusion of conservation programs in gap areas is of particular importance as levels of land development around protected areas has a direct influence on their effectiveness as conservation units (Durán et al. 2013; Leroux and Kerr 2013). In northern Peru, the inclusion of communities is of particular importance as human populations are relatively high and increasing (PROCLIM/CONAM 2005; Shanee et al. 2014). The protected area network in northern Peru covers a fairly large area of forests including areas of current and future habitat for A. miconax and L. flavicauda but provides little protection for P. oenanthe. As with other areas in the Andes, protected areas in northern Peru may not be enough to safeguard these species from anthropogenic development activities (Swenson et al. 2012). Including the predicted increase in niche area for P. oenanthe, anthropogenic activities will still reduce total available area for the species, even assuming no more habitat loss within protected areas.

The results presented here show that multiple drivers of extinction risk combine to threaten species (Newbold et al. 2014) and that future man-made climate changes will have variable effects depending on a species’ habitat and ecological needs (Newbold et al. 2014). Although climate change is predicted to dominate conservation during this century (Bonan 2013; Laurance and Williamson 2001; Lewis 2006; Lewis et al. 2011; van Aalst 2006; Veech and Crist 2007), other anthropogenic activities are still and, in many cases, will continue to be the major drivers of extinctions (Feeley and Silman 2010; Hurtt et al. 2011; Krausmann et al. 2013; Newbold et al. 2014; Peres et al. 2010; Tilman et al. 2001). Future conservation actions should not only concentrate on mitigating the effects of climate change but should also concentrate on reducing other anthropogenic pressures which are driving species to extinction. This is particularly true for species with limited geographic ranges and habitat specializations (Newbold et al. 2014) that are intrinsically more at risk of extinction (Cardillo et al. 2005; Purvis et al. 2000a, b) but that also may not be able to adapt to changing climates and habitats in the near future.