Final countdown for biodiversity hotspots

Most of Earth's biodiversity is found in 36 biodiversity hotspots, yet less than 10% natural intact vegetation remains. We calculated models projecting the future state of most of these hotspots for the year 2050, based on future climatic and agroeconomic pressure. Our models project an increasing demand for agricultural land resulting in the conversion of >50% of remaining natural intact vegetation in about one third of all hotspots, and in 2–6 hotspots resulting from climatic pressure. This confirms that, in the short term, habitat loss is of greater concern than climate change for hotspots and their biodiversity. Hotspots are most severely threatened in tropical Africa and parts of Asia, where demographic pressure and the demand for agricultural land is highest. The speed and magnitude of pristine habitat loss is, according to our models, much greater than previously shown when combining both scenarios on future climatic and agroeconomic pressure.

located in tropical developing countries that house more than a third of the global human population (Veech, 2003). These countries face strong challenges, such as high demographic pressure, food shortage, poverty and corruption (Williams, 2011). These problems are expected to multiply in the future (Bradshaw & Brooks, 2014). Current estimates indicate that in most hotspots less than 10% of natural intact vegetation (NIV) remains (i.e., nondisturbed pristine ecosystems, see Sloan, Jenkins, Joppa, Gaveau, & Laurance, 2014). Since such NIV is essential for the survival of many species, its potential rapid conversion into agricultural land is assumed to have severe consequences for biodiversity (Bradshaw & Brooks, 2014;Venter et al., 2016a,b). In addition to the complete destruction of NIV, its deterioration through the exploitation of natural resources such as timber harvesting and the collection of wood for domestic use (such as cooking) is leading to what is known as "empty forests" (Wilkie et al., 2011).
Recent modeling of future land use change impacts on habitats and biodiversity found that by the year 2100, global NIV will be reduced by a further 26-58%, with estimated species extinction of 0.2%-16% based on species-area models (Jantz et al., 2015). While this study focused on the impact of land use change on ecosystem health and biodiversity, and other studies examined the influence of climate change (Midgley, Hannah, Millar, Rutherford, & Powrie, 2002, Thomas et al., 2004, we here combine both factors and run joint models for 33 hotspots for the year 2050. While the changes in climatic conditions are highly dependent on the global circulation system, agroeconomic pressure strongly depends on local and regional factors, such as demographic pressure, land-use practices, human development, and societal factors like political stability (Roe & Siegel, 2011). However, both factors may also influence each other, antagonistically (when climate change causes lower productivity due to dryer conditions, and thus reduces agricultural output and diminishes land-use pressure, or synergistically (when increasing precipitation allows future improved agricultural conditions for some regions) (see Challinor et al., 2014). We analyze which hotspots are likely to suffer the most in the near future under multiple, global and local factors. Furthermore, we test if the susceptibility of a hotspot depends on properties of the hotspot itself, such as limited geographical size, the climate zone, or the continent where the hotspot is located.

MATERIAL AND METHODS
In all models conducted, we omitted the hotspots New Caledonia and Polynesia-Micronesia due to their limited size and data deficiencies. For the remaining 33 hotspots considered in our study, we compiled the following information: continent, climate zone, size, altitude, number of endemic plants, protected area, and remaining vegetation (Table 1).

Assessing the impact of climate change on biodiversity hotspots
We define climate change as potential future shifts in precipitation and temperature. To quantify the magnitude of climate change, we calculated the currently realized climate space within each hotspot and compared it to projected future climate spaces (01/2045-12/2054) based on different scenarios of climate change. Climate data were obtained from the Coupled Model Intercomparison Project (CMIP5) and comprise four IPCC greenhouse gas emission scenarios (i.e., representative concentration pathways, RCPs 2.6, 4.5, 6.0, 8.5; IPCC 5th assessment report). The RCPs stretch from a "best-case" (2.6) to a "worst-case" (8.5) scenario (van Vuuren et al., 2011). We randomly chose 8,522 grid cells (0.5 • × 0.5 • ) located within the hotspot areas, extracted 10 daily climate parameters for these cells from the comprehensive dataset, and conducted a principal components analysis in order to reduce the redundancy in the dataset. We then tested for each grid cell whether any value of the 14 PCs exceeded the baseline range; if that was the case, we assumed that climate changed significantly in this cell. The hotspot area "lost" to climate change is thus the ratio of the cells with a significantly changed climate to the total number of cells in a hotspot.

Assessing the impact of agroeconomic change on biodiversity hotspots
The factor agroeconomic change depicts the external pressure exerted on hotspots by changes in the human population and income and thus changes in demand for agricultural goods. Climate change effects are also indirectly considered in this factor via changes in crop yields. Future changes in national food demand are assumed to be proportional to changes in the human population, for which projections were taken from Shared Socioeconomic Pathway (SSP) scenarios: RCP 2.6 (SSP1), RCP 4.5 (SSP1), RCP 6.0 (SSP2), and RCP 8.5 (SSP3) (Riahi et al., 2017). We further assume that the demand also depends on income; a higher income leads to higher demand, but with decreasing marginal effects of income (Engel curve). Future changes in agricultural productivity were estimated with the climate-data driven EPIC model (Williams, 1995). The overall change in crop productivity was computed for each spatial unit as the weighted sum over all major crop management systems, (weight = estimated current agricultural land share of each system). The change in agricultural productivity is thus only driven by climate changes, not adaptation of agricultural management. Lastly, we calculated the area needed to compensate for changes in demand for agricultural commodities as well as agricultural productivity, thus yielding the change in pristine hotspot area due to agroeconomic change.
T A B L E 1 Characteristics of global biodiversity hotspots. Given are name of hotspot, running number (ID corresponds with Table and Fig. provided in the article), continent, climate zone, size (km 2 ), Altitude (m), number of endemic plant species, protected area (km 2 , and %), and an estimate of current remaining vegetation RV for each hotspot using a combination of automated and visual satellite-image analyses of land-cover classes and conditions as well as the mapping of major landscape disturbances from Sloan et al. (2014). Hotspots are listed in alphabetic order

Combined climatic and agroeconomic change
The combined pressure from climate change and agroeconomic change is given as the projected total change in pristine area from 2010 to 2050, calculated as the sum of the change in agricultural area in the hotspot and the area of remaining pristine vegetation, of which the ratio of area with significantly different climate in 2050 is subtracted. Figure 1 displays a summary of our modeling approach. A detailed description of the methods is provided in Appendix S1. Cropland and other land use F I G U R E 1 Graphical illustration of the derivation of agroeconomic, climate and combined pressure indices. The agroeconomic index depicts the loss of current pristine area (in percentages) to agroeconomic pressure, which is dependent on changes in agricultural production levels, population, and income development. The climate index depicts the share of hotspot area where climate will be significantly different in 2050 in comparison to today, given as loss of total area. The same share is then applied to the area of remaining vegetation calculated in step 1 (agroeconomic pressure), which yields the final area of remaining vegetation, of which the total loss of pristine area can be calculated (combined index)

Estimating endemic plant species loss based on area loss
We calculated total endemic species loss following the method described in Jantz et al. (2015). A species-area relationship is used to estimate potential extinction rates based on area loss. The number of species in a smaller area A is calculated based on the species numbers in a larger area A 0 : with S(A) as the number of species in subarea A within the larger area A 0 , z as the slope of the power function relating species numbers to area (Rosenzweig, 1995) and S 0 as the total number of species in A 0 . Values for A and A 0 were taken from Table 1, values for S 0 from Mittermeier et al. (2005) and values for z from Jantz et al. (2015). This is a simplistic model for predicting species loss, but without data for individual species, more sophisticated modeling (e.g., Fordham et al., 2016) is not possible.

Effects of hotspot characteristics on the susceptibility to change
The susceptibility of all hotspots to climate change, agroeconomic change and combined changes were analyzed with all characteristics of hotspots listed in Table 1 used as explanatory variables. Linear models were fitted for all three drivers of change separately. Explanatory variables were included in the order continent, size, remaining vegetation, climate zone, endemic plant species and altitudinal in full models, which were simplified stepwise removing terms of least significance first base of F-ration tests. Models were fitted using base function in R, version 3.3.0.

RESULTS
Climate will change significantly on 25-33% of hotspot area on average, depending on RCP scenario (Table 2, and  Table S2 and Figure S2.1 in Appendix S2). A total of 2-6 hotspots are projected to experience a significant change in climate for more than 50% of their area, and 2-3 for less than T A B L E 2 Results from modeling the percentage loss of pristine hotspot area ("Remaining vegetation") under climate pressure (temperature, precipitation, "Clim"), and the need of land according agroeconomic pressure (yield, population, income, "Agr"), and combined pressure ("Com"), for a range over all RCP/SSP scenarios. For climate pressure we assume that if climate changes significantly from baseline conditions in the remaining pristine area of the hotspot, the area is "lost" for biodiversity conservation. Values higher than 100% (as e.g., in hotspot "Eastern Afromontane") indicate that even more area than is available would be needed to satisfy future demand for commodities. Hotspots New Caledonia and Polynesia-Micronesia were not considered as their geographic area is too restricted and thus do not allow the running of stable models. The endemic plant species loss was calculated based on the species-area relationship described in Jantz et al. (2015) and is based on scenario RCP4.5/SSP1 T A B L E 3 Summary of linear models of variation in the two drivers of change and the combined index of change and possible explanatory characteristics of the 33 biodiversity hotspots analyzed. Data in brackets were not significant and have been removed from the minimum adequate model given in bold in the order indicated by the superscripts. Trends are displayed in Appendix S3  10%. The main explanatory variable for changes in climate is the continent on which a hotspot is located and the size of the hotspot, with hotspots on the African continent and small hotspots suffering the highest change in climate space (Table 3, Figure S2.2 in Appendix S2). The pressure from agroeconomic changes is more severe: 9-13 hotspots are projected to lose all accessible NIV to agriculture, depending on RCP/SSP scenario. For these hotspots, the modeling indicates that to satisfy local and global food and energy demand, more agricultural area than is actually available as remaining NIV would be needed. Another 1-4 hotspots will lose 50-99% of NIV by the year 2050 (Table 2, Figure 2). The main driver of agroeconomic change is a change on the demand side, that is population growth and income increase, whereas changes on the supply side, that is climate-induced changes to crop productivity, are of secondary importance ( Table 2, Table S2 in Appendix S2). The main explanatory variable for agroeconomic pressure is the currently remaining vegetation, where hotspots with an already low remaining pristine area are more likely to suffer further degradation (Table 3, Figure S2.2 in Appendix S2). According to our model projections, hotspots on the African and Asian continent are expected to suffer particularly under future increased habitat transformation (Figure 2). In many, but not all cases, the combined climate and agroeconomic pressure is similar to agroeconomic pressure alone, indicating that climate change in these hotspots is slight. Seven to eight hotspots appear to be little affected by agroeconomic pressure and will only lose 5% or less of its remaining NIV. Only 1-2 hotspots (Western Ghats and Sri Lanka in all four scenarios, plus Caribbean Islands in scenario RCP4.5/SSP1 and Coastal Forests of Eastern Africa in scenario RCP8.5/SSP3) are expected to suffer under both, increasing agroeconomic pressure (>100% of losses of NIV due to increasing demand for agricultural area) and strong climate change (>50% loss of area under original climatic conditions). Of great concern is the fact that losses of NIV through agroeconomic pressure are projected to be especially high in hotspots which already have a low percentage of NIV remaining (Table 1, Table 2). The hotspots projected to lose all or large parts of the remaining NIV by 2050 include many with the highest diversity of endemic species, for example, the Philippines, Caribbean Islands, Mediterranean Basin, Madagascar and Indian Ocean Islands, and Atlantic Forest.
Based on the calculated area loss, we estimate that in at least 9 hotspots all endemic species will be committed to extinction, and overall 36.5-52.1% (depending on the z-value) of endemic plants will be committed to extinction across all 33 assessed hotspots.

DISCUSSION
Our results show that agroeconomic pressure will continue to be the main driver of NIV loss in hotspot areas. Contained in this driver are supply and demand side changes for food crops, with the demand side affected by demographic and income changes. Demographic pressure is extraordinarily high in most hotspot regions (Williams, 2013), leading to an increasing demand for food crops and the expansion of agricultural area. Furthermore, a higher wealth in societies is generally accompanied by a higher caloric intake per capita, particularly in developing and newly industrialized countries (Davidson & Andrews 2013). To produce the additional calories, agriculture needs to be intensified and cropland expanded (D´Amour et al., 2017), even more so if the higher demand is mainly for meat (Green et al., 2005). The exploitation of landscapes may also have an impact on the pressure onto pristine ecosystems: Soil quality is low in most tropical regions, so that rapid soil nutrient depletion after land conversion creates further need for new, fertile land (Habel et al., 2015).
Macroeconomic processes such as increasing global demand for meat, cash-crops and bioenergy plants may also cause large-scale transformation and cultivation of remaining NIV, as projected for hotspots in Brazil, West Africa, and parts of Asia, particularly the Sundaland hotspot (Koh, 2007). The growth of the palm oil industry in Malaysia and Indonesia adds little food benefits to the local community, but has resulted in large-scale loss of natural habitats (Koh & Ghazoul 2010). Increasing demand for meat in industrialized countries frequently results in the production of soybeans to feed EUbased livestock but also in outsourcing of meat-production to developing countries (Green et al., 2005), where it may seriously impact NIV, including in hotspots. Thus, habitat transformation for agriculture has been identified repeatedly as the most important driver of biodiversity loss worldwide (Maxwell, Fuller, Brooks, & Watson, 2017).
We recognize that some habitats and species may persist even in largely anthropogenic landscapes. Some hotspots are too steep, too difficult to access and/or have soils with low fertility and are unlikely be cleared for agriculture. Thus, there is no homogenous pressure across hotspots as assumed by our models and some endemic species may be less endangered than estimated. Furthermore, an average of 13% of hotspot area has some form of protected status, designating an area where species may survive. However, as previous studies have shown, most remaining pristine habitats are small and isolated and not able to hold viable populations over long periods. Environmental and demographic stochasticity and reduced genetic variability may lower individual fitness and species persistence (Melbourne & Hastings 2008). On the positive side, there is emerging evidence that in some tropical hotspots of rural depopulation (e.g., Philippines, Andes, Caribbean), habitats regrow and biodiversity recovers (Posa, Diesmos, Sodhi, & Brooks, 2008), but this has not been examined systematically yet.
The loss we estimate should be interpreted as an upper boundary, as we do not consider potential adaptation, neither on the sides of human beings, nor on biota. There may evolve various improvements in respect of agricultural production, food processing, and environmentally friendly intensification. Species may also adapt to persist in anthropogenic landscapes, that is, novel ecosystems, even after NIV has become modified (Pereira, Ziv, & Miranda, 2014;Newbold et al., 2015). We recognize that our projections for some places, such as the Mediterranean Basin, where the area of pristine vegetation cover is currently increasing, do not provide a realistic picture. In these regions, the negative trend of land-conversion is slowing down or is reversed either because of more sophisticated land management regimes or because the countries housing these hotspots import large quantities of food.
Our results indicate that the impacts of climate change are surprisingly small given the dire predictions from many other researchers. The time period to 2050 is relatively short compared to other projections, and climate changes will be less advanced than in projections to 2100 (e.g., Maxwell et al., 2017). However, climate change will and is having severe direct and indirect effects on pristine habitats and biodiversity, of which some aspects may not be well addressed in our models. Direct effects are particularly relevant for geographically restricted hotspots, as most endemic species will not be able to respond to climate change by colonizing new suitable areas and thus vanish entirely. Even in more accessible areas, many endemic species are adapted to and dependent on specific interactions with other taxa, resources, niches and environmental conditions, and do not possess high levels of phenotypic plasticity (Huey et al., 2012).
In addition to these direct effects, climate change may affect biodiversity indirectly through its impact on crop yields. According to our model, improved climate conditions may in some hotspots lead to a slight increase in yields, potentially encouraging farmers to clear more area to reap the benefits of higher returns for the same investment. Dryer and hotter climatic conditions, on the other hand, may reduce agricultural yields in specific regions and reduce the local pressure while shifting the burden of producing food to other, more productive areas (Schneider et al., 2011).
The major driver of biodiversity loss is the rapid destruction of pristine habitats across the globe, particularly the African and Asian continent (see also Maxwell et al., 2017;Tilman et al., 2017). Thus, there is a pressing need to anticipate and prevent further losses of the remaining NIV in hotspots. The situation is much more acute than previously expected and not evenly distributed, with some hotspots like the Philippines, Caribbean Islands, Madagascar and Indian Ocean Islands and the Atlantic Forest requiring further urgent conservation action, as they contain a large number of endemic species and are projected to face high pressure in future. In addition, only a small proportion of the land surface in these hotspots is under protection. Factors driving this demand for land are acting at various spatial scales, from local, regional, country-wide to global, mainly driven by increasing demographic pressure and life standards. As previous studies have suggested (Maxwell et al., 2017;Tilman et al., 2017), new national and international measures are needed to seek ways in which the demand for land can be reduced and current agricultural practices improved. While much effort has been invested in new policies to mitigate climate change, our study shows that threats from land conversion are much more urgent. The world's governments have committed to reduce the demand for land and for improved agricultural practices through the Sustainable Development Goals (SDGs) but this needs to be done in an integrated way that delivers across all SDGs including those on food security and climate change, as well as on biodiversity.

AUTHOR CONTRIBUTIONS
JCH, LR, UAS, and NES developed the idea of this study; LR, UAS, ES, NT, and RSD calculated agroecological models; and JOE and DR performed climate models. All contributed with writing and the interpretation of data.