Our results, at a conservation practitioner's scale, identify geographic areas in the eastern slopes of the Peruvian and Bolivian Andes with high concentrations of endemic species, areas with high irreplaceability, gaps in protection for both species and ecosystems, and ecological systems where these endemic species reside. Our focus on a variety of vertebrate and plant groups underlines the variation in spatial distribution patterns among different taxa. The geographical extents and levels of current protection of the ranges of species, endemic areas, irreplaceable areas, and key ecological systems also vary widely.
Mapping species distributions is inherently limited in terms of a true representation of biodiversity. As a one dimensional map of potential habitat based on climate, elevation and vegetation, the distribution modelling omits species interactions such as predation and competition, effect of human edges along habitat, and the effects of climate change [63, 81]. However it is a large step forward for this region where current conservation analyses are obliged to rely upon generalized hand-drawn maps of species ranges, or species lists for very large multi-country geographical units (e.g. Hotspots or Ecoregions) that were not intended nor appropriate for regional or landscape level applications . Our mapping of ecological systems, for example, identified ~90 ecological systems; the same area is covered by parts of 12 ecoregions (sensu).
The locations of high endemism (Figure 4) agree with past studies for taxa that have been examined previously, yet earlier studies were conducted with much less data availability and at much coarser spatial resolution. The high levels of endemic bird richness found in the northern part of the study area are consistent with previous work [36, 40, 83]. However, our study revealed previously unrecognized areas of bird endemism in Peru: the southern Huánuco region, the western Cordillera de Vilcabamba, and the region along the Río Mapacho-Yavero east of Cuzco (Figure 4, 7; see  for details). This study is the first to reveal detailed patterns of endemic species for mammals and amphibians (see  for location descriptions), and therefore few comparisons with past studies can be made. However the areas of high endemic mammal richness in Peru corroborate the one regional study of similar scope  and the mid-elevation concentration of endemic amphibians coincides with the less spatially explicit suggestions of  and . Centres of plant endemism varied among groups and families, yet the pattern for one group (Ericaceae) did correspond to a previous study . Other existing analyses use such coarse resolution (e.g., the 1°×1° Flora Neotropica grid ) that comparisons are too general to be meaningful. For most plant groups, this study is the first to assess spatial patterns of endemism in the eastern Andean basin of Peru and Bolivia.
Despite the increased level of detail in spatial scale that our dataset provides, continued work needs to focus on refining these biodiversity data to even finer spatial scales (e.g. 1:100,000) and higher levels of accuracy. The dataset and analyses we have produced are tied to the time of specimen collections and to the quality of available data. As more specimen locations are collected in the future with increasingly accurate locational and elevational information (using a precise global positioning system), distribution models could be re-run and models validated. Geographical collection bias, a problem for presence-only distribution models could be addressed in future modelling efforts by the selection of pseudo-absence data having similar bias as the presence data . More precise geographical climate data could refine the spatial resolution of model predictions; there will be an increasing prevalence of 'downscaled' geographical climate data thanks to higher spatial resolution digital elevation models (SRTM and ASTER). However the overall limitation is the lack of adequate meteorological stations in the region. Other layers that would be useful to incorporate upon their refinement would be a characterization of soils or geology. We successfully modelled all endemic vertebrates yet, additional models of plant species distributions should be realized. Considering there are over 5000 endemic plant species in the country of Peru (of which approximately 3200 fall within the altitudinal range of our study area) , our 435 species represents a small fraction of endemics to Peru and/or Bolivia in the Amazon watershed.
Our country wide analysis could be refined to department scale using land tenure information and local to regional protected areas and resource concessions. Current maps of forest deforestation and degradation would aid in calculating the remnant ranges for each species as well as ecological systems. Further analysis could be made in terms of the complementarity of species assemblages and their relationship to ecological systems and levels of protection, whose results could further guide priorities. However, the greater battle for biodiversity conservation lies in managing elements beyond our datasets and analyses, as described below.
The geographical patterns of endemism, irreplaceability, and ecosystems revealed here pose several challenges for conservation planning in the region (Figure 7). The most obvious challenge is the geographic configuration of the locations of endemic or irreplaceable areas. Although we mapped only a small subset of the biodiversity that occurs in the region, we found striking geographic differences in endemic species concentrations across taxonomic groups. The difficulty of using surrogates of one species group for another has been recognized [15, 52, 53], and our findings underscore the need for a large portfolio of protected areas and other protection mechanisms to conserve diverse elements of biodiversity.
Second, the gap analysis demonstrates that many areas where concentrations of endemic species occur remain unprotected today. Considering ongoing threats in the region from infrastructure development , oil extraction , gold mining [90, 91], illicit crops , and the continually advancing agricultural fronts, more carefully situated protected areas and novel land use regulation strategies will be necessary to safeguard substantial amounts of biodiversity.
Third, although we use protected area coverage to evaluate conservation coverage, we acknowledge that protection status does not necessary translate into actual protection on the ground. Indeed, resource extraction and degradation is continuing in many legally protected lands in the study area . Nevertheless, these reserves have the potential to protect important segments of endemic and irreplaceable areas, suggesting that strengthening the capacity of relevant authorities to improve protection is an important and continuing challenge.
Fourth, large reserves will probably be insufficient to maintain all biodiversity. Although large reserves often provide the best means for maintaining well-functioning ecosystems , the pattern of endemism we document, in which microendemic species are scattered across the landscape and not always concentrated geographically, will require multi-pronged conservation efforts. Restricted-range species that occur far from the major areas of endemism or irreplaceability, such as the two primates in the Bolivian Beni, would benefit from a wider network of smaller reserves, perhaps established by departmental, provincial, or municipal governments or private entities. Current trends toward the decentralization of responsibility for natural resource management to provincial governments may provide a useful institutional context for the establishment of some of these smaller, but nonetheless critical reserves .
Our finding that highly endemic areas disproportionally occupy a handful ecological systems presents yet a fifth challenge. Ecological systems characterize broad, integrated units of biodiversity and can be used as a coarse filter for conservation. While maintaining representation of all systems in landscape-level protection plans , planners may need to balance the need to protect endemic species with the need for a representative sample of ecosystem type and function as well as other targets such as endangered species or carbon sequestration. On the other hand, these particular ecological systems could be considered surrogates for areas of high endemism. The systems are advantageously close together in the Yungas region, are relatively limited in extent (totalling 7% of the study area), and have individual ranges that are < 35% protected.
A final challenge is continued climate change. We know that because of climate change, the ranges of many species will shift across the landscape and possibly out of protected areas [96, 97]. Evidence is accumulating that along the Andean slope, species shifts are already occurring [98, 99]. Yet the variation in projections of future South American climate makes assessment of the effects on species' distributions difficult . The steep elevation (and therefore climate) gradients in the Andes, where most endemic species are located, suggest that such displacements may take place over relatively small distances. Extinctions are most likely in species inhabiting the highest-elevation habitats, which occur above our study area . Nevertheless, planners should consider adding upslope buffers to conservation areas designated using current distributions of endemic species, and future research could model these species distributions under future climate scenarios.
To complement the further creation and effective management of protected areas, other alternative approaches, which will result in the maintenance of key ecosystems, should expand and continue. These approaches include, strategic conservation on private lands and brokering conservation agreements with private companies, effective land use planning and possibly carbon accounting at the regional government level for both public and private lands, and payments for ecosystem services (e.g. water provision, ecotourism recreation, carbon storage through forests: Reducing Emissions from Deforestation and forest Degradation, REDD). However priority areas for ecosystem services concessions may not necessarily overlap with priorities for biodiversity conservation (e.g.).