Co-occurrence of climate-change induced and anthropogenic pressures in Central American key biodiversity areas

Central America hosts many key biodiversity areas (KBAs), areas which represent unique and irreplaceable ecosystems of global importance for species conservation. However, large extents of these areas are not under legal protection and could be threatened by pressures from land use change (e.g. deforestation and agricultural expansion), high human population density (e.g. population growth and urban sprawl) and climate-driven biome shifts. Here, we simulated future biome stability under the influence of climate change across KBAs in the Mesoamerican biodiversity hot spot and combined the results with projections of land use and population density up to the end of the 21st century. We applied four forcing scenarios based on two global climate models (GFDL-ESM4 and IPSL-CM6A-LR) and two shared socio-economic pathways (SSP1-2.6 and SSP3-7.0), which represent a range from low to high emission pathways. Our model projected decreased biome stability in 39%–46% of protected areas in KBAs, whereas this number even increased to 59%–60% for unprotected areas in KBAs (depending on the climate scenario). While human interferences in protected parts of KBAs are expected to be limited, large parts of unprotected areas in KBAs were projected to be pressured by multiple factors at once and are reason for concern. In particular, high human population pressures (>10 people km−2) emerged as a main threat over 30%–44% of the unprotected area in KBAs. These were largely accompanied by pressures from land use and sporadically reinforced by pressures from climate-driven biome shifts. Among the hot spots facing multiple high pressures are some of the last tropical dry and montane forest ecosystems in Central America, which stresses the need for urgent conservation action.


Introduction
The acceleration of global change is putting increasing pressures on earth's biodiversity (Watson et al 2016, Betts et al 2017, Potapov et al 2017. Against this prospect, the persistence of 'key biodiversity areas' (KBAs)-unique ecosystems hosting globally threatened, range-or biome-restricted speciescould be of paramount importance to avoid irreversible species losses (Eken et al 2004, Langhammer et al 2007, Donald et al 2019, Kullberg et al 2019. Within this context, Central America represents an outstanding case as it features a large number of such areas, covering more than 362 000 km 2 -or 25% of its terrestrial area. Despite the biological importance of KBAs, around 51% of their area within Central America is currently unprotected. This is of concern considering that the region has been and is continuing to be shaped by strong human pressures. These include deforestation, expansion of cropland and pastures, land use intensification, land degradation, population growth and urbanization (Weinzettel et al 2018, Hu et al 2021. Existing global future projections show alarming trends towards land use intensification and increasing population density with hot spots in Central America, while environmental pressures from climate change may additionally impact the region (Lyra et al 2017, Chen et al 2020. While these studies provide important general insights, the considered pressure factors need to be jointly evaluated to reveal overlaps and possible extreme hot spots. A few large-scale studies have already integrated climate change with land use projections (Asner et al 2010, Boit et al 2016, Hof et al 2018. However, the comparability of land use and climate change effects in modelling studies is typically limited due to data heterogeneity, scale mismatches and the lack of common units of measurement (de Chazal andRounsevell 2009, Titeux et al 2016). A recent study by Martens et al (2022) showed how this gap may be bridged by mapping multiple pressures on a common scale for protected areas in Africa. In contrast, for Central America there remains a lack of studies projecting and comparing multiple pressures in a common system and at a fine resolution suitable for conservation planning. Furthermore, since time and resources for conservation actions are limited, special focus needs to be put on the most critical areas (Hannah et al 2020, Jung et al 2021, Voskamp et al 2022. Particularly in view of unprotected KBAs and the need to strengthen efforts to achieve global biodiversity targets (IPBES 2019), information on potential pressures in areas of interest for biodiversity conservation is crucial to have a chance to mitigate them.
Here, we aimed to fill these gaps by investigating the potential co-occurrence of different pressures in a standardized reference system and at a high spatial resolution useful for conservation planning. Therefore, we first simulated vegetation dynamics for a broad range of climate scenarios. We then derived climate-induced pressures by estimating biome shift probabilities between the reference  and future period (2071-2100). In the next step, we combined these results with projections of the anthropogenic pressures land use (change) and population density by standardizing all pressures to a common scale (known as human footprint mapping). By overlaying the results, we identify hot spots of multiple pressures. Finally, we discuss implications for conservation planning with particular regards to prioritization schemes.

Study area
Our study area covered the Mesoamerican biodiversity hotspot (Myers 2003) between 4-24 • N and 73-100 • W with a main focus on areas defined as KBAs (figure 1). Inside this extent 440 KBAs can be found, that range from very small patches (2 ha) to extensive landscapes (20 918 km 2 ) with a median size of 193 km 2 . Overall, the region is of critical importance for biodiversity conservation: Of a total of around 16 000 species included in the IUCN red list  for Mesoamerica, approximately 5000 are endemic  to the region and around 3300-or 20%-are listed under the categories vulnerable, endangered or  critically endangered (IUCN 2021). Due to intense human modification, remaining natural landscapes are largely fragmented, leaving only a handful of large undisturbed forest landscapes. Most of these are tropical moist forests, the largest of which are the Maya Forest in Southeast Mexico, forests in the Mosquitia region in Northern Honduras and Nicaragua and forests along the Cordillera Central in Costa Rica and Panama. On the other hand, only few seasonally dry forests remain intact, although most of them are classified as KBAs (Portillo-Quintero and Sánchez-Azofeifa 2010).

Data
Bias-corrected and spatially downscaled (0.5 • ) daily climate forcing data (precipitation, average surface temperature and solar radiation) were retrieved from the ISIMIP3b data set (Lange 2019, Lange and Büchner 2021. To account for the uncertainty of climate scenarios, we selected four CMIP6 scenarios that span the range from low to high climate forcing. From the five global climate models (GCMs) included in the ISIMIP-3b dataset, we selected GFDL-ESM4 (GFDL), which generates a low climate forcing and IPSL-CM6A-LR (IPSL) which lies at the other end of the spectrum. At the same time, both GCMs show different characteristics with regards to the timing of changes and their temporal variability (figure S10). In terms of shared socio-economic pathways (SSPs) we included SSP1-2.6 (SSP126) which represents a low emission scenario and SSP3-7.0 (SSP370) which represents a high emission scenario. The latter was chosen since it resulted in almost equally strong temperature increases but even higher precipitation variabilities in the study region compared to the extreme marker scenario SSP5-8.5.
Nitrogen deposition data was obtained from Lamarque et al (2013) and data on atmospheric CO 2 concentration from Meinshausen et al (2020). Elevation data used for downscaling climate inputs was obtained from the SRTMGL1 dataset at 1 arcsecond resolution (NASA JPL 2013). Land use projection data was taken from the LUH2 dataset at 0.25 • spatial resolution (Hurtt et al 2019a(Hurtt et al , 2019b(Hurtt et al , 2020 and downscaled (∼1 km) projections of population density from Gao (2020Gao ( , 2017; based on Jones and O'Neill 2016). To harmonize the resolution of all pressure factors (ranging from 0.000 28 • to 0.25 • ) and avoid strong up-or down-scaling, we chose a target resolution of 30 arc-seconds (∼1 km) for the pressure analysis.
Shapefile data for protected and KBAs were obtained from the World Database on Protected Areas (UNEP-WCMC, IUCN 2020) and the World Database of KBAs (BirdLife International 2020), respectively. For the identification of undisturbed forests, we used the IFL 2020 dataset (Potapov et al 2017, IFL Mapping Team 2021. An overview of all datasets used in this study is given in table S1.

Dynamic vegetation modelling
To simulate climate change impacts on vegetation dynamics we used the Lund-Potsdam-Jena General Ecosystem Simulator v4.0.1. (Smith et al 2014(Smith et al , 2001, also see https://web.nateko.lu.se/lpj-guess/), which has been applied in numerous climate change impact studies before (e.g. Hickler et al 2012, Seiler et al 2015, Hof et al 2018, Martens et al 2021. This dynamic global vegetation model operates at the level of plant functional types (PFTs) and models main physiological and ecosystem processes in relation to climatic and edaphic drivers at stand scale (for a detailed description see supplement). The model was run for the time period covered by the climate data set (historical data: 1850-2014, forecast: 2015-2100) for all GCM-SSP combinations. As a basis for comparison between current and future conditions, we defined the last 30 years of the historical climate data  as the reference period. Equally, for the future time period we selected the final 30 years of the future climate projections (2071-2100).
LPJ-GUESS outputs were first aggregated to the target resolution of 30 arc-seconds. These data were then used to assign biomes based on the leaf area index and net primary productivity of all PFTs that were simulated to appear in a given grid cell (for details see supplement). By comparing yearly biomizations of the future period 2071-2100 against average biomes of the reference period 1985-2014, we then calculated the biome stability for each grid cell (=ratio of years with the same biome as the reference). For the subsequent pressure scoring, we subtracted this value from 100 to derive the biome shift probability as indicator for climate induced pressures. Finally, the result was divided by 10 to match the standardized value range from 0 to 10 of the human footprint mapping used for the other pressure factors (see below). As an extension to this, we further zoomed in and analysed the last remaining intact forest regions inside KBAs for trends of changing PFT shares over time.

Anthropogenic impact projection
Anthropogenic pressures were mapped by analysing projections for population density (which reflect the demand for resources and living space) and land use (which indicate the intensity of human use or conversions of (near-) natural land). The impact mapping was based upon the 'human footprint' framework (Sanderson et al 2002, which standardizes different human pressures to a common scale of 'pressure scores' ranging from 0 (no pressure) to 10 (maximum pressure) and enables their joint evaluation. For the estimation of pressures connected to human population, human population density data was scaled logarithmically with the formula from Venter et al (2016): 3.333 × log10 (population density + 1). Since the maximum pressure score is reached with 1000 people km −2 , higher population density values were directly converted to a pressure score of 10. Because the data was only available at a decadal temporal resolution, we used the mean of the first two time steps 2010 and 2020 for the historical period  and the mean of the time steps between 2070-2100 for the future period (2071-2100). For the estimation of land use pressures we followed the approach by Martens et al (2022), who grouped LUH2 classes to match the original pressure score classification by Sanderson et al (2002) and Venter et al (2016) and extended it by adding classes for secondary vegetation. While secondary vegetation has been found to positively influence species richness and abundance compared to other non-primary land uses, the effect highly depends on vegetation maturity (Newbold et al 2015). The adapted scheme by Martens et al (2022) accounts for this by differentiating between young, intermediate and mature secondary vegetation (pressure scores 1-3). A full overview over all classes and their assigned pressure scores used here is given in table S3. These scores were then multiplied with projected land use fractions from the LUH2 dataset. Finally, the land use pressure map was downscaled through cubic interpolation to match the 30 arc-second resolution of the other pressure factors. Since human pressures are most relevant in unprotected landscapes, we focused on unprotected areas in KBAs for this analysis.

Integration of pressure factors
To integrate the projections from all three pressure factors, we used an additive approach (compare de Chazal and Rounsevell 2009). Therefore, we computed the average across all factors to arrive at continuous results. To highlight areas with high potential pressures from multiple sources, we additionally binarized the continuous results for each pressure factor into 'high' or 'low' pressure. To this end we used thresholds of ⩾4 for land use, i.e. the value for pasture above which a landscape may be considered strongly influenced by humans (see Watson et al 2016); ⩾3.47 for population, which equals a population density of 10 people km −2 and has been considered a threshold between remote and residential areas by various studies (Tognelli et al 2005, Koh et al 2006, Ellis and Ramankutty 2008; and a threshold of ⩾5 for biome shifts, which signifies changed biomes in the majority of years (i.e. biome stability ⩽50%) in the future time period. To account for the uncertainties related to these threshold definitions, we additionally analysed the sensitivity of the results by raising/decreasing the threshold values by ±33%.

Biome shifts
Throughout all climate scenarios, approximately 50% of the area of the studied KBAs showed decreased biome stability. The GFDL-SSP370 scenario overall showed the strongest impacts with less than 50% biome stability in 19% of the area, followed by IPSL-SSP370 (∼14%), IPSL-SSP126 (∼13%) and GFDL-SSP126 (∼12%). However, these statistics differed markedly depending on the protection status of the KBAs. For protected areas in KBAs between 54%-61% of the area was projected to remain at full biome stability, whereas for unprotected areas in KBAs this only applied to 40%-41% of the area (figure S2).
In terms of the spatial distribution, the highest scores for pressures from biome shifts were found in the central mountain ranges of Mexico, Guatemala and Costa Rica, on the Yucatan peninsula (south eastern Mexico) and along the Pacific coast of the study region. These hot spots were consistent over all climate scenarios with small local variations. Noteworthy were hot spots in the seasonal and dry forests of the Maya Forest and Selva Zoque (both located in Southern Mexico)-two of eight remaining intact forest regions in Central America (figure S3). Similarly, most of the dry forest KBAs along the Pacific coast of Central America showed instabilities. A temporal analysis of these results further revealed, that large parts of these stability declines only occurred after 2071, for example as it was the case for the central mountain ranges and the Yucatan peninsula (figures S5 and S6).

Human pressure trends
The projections land use pressures showed moderate disparities at a rather large scale (figure 3). While these generally did not exceed a score of 3 in the central and southern part of the study region, the northern part-and particularly the central regions of Guatemala and Mexico-largely showed scores of 4 and above. Compared with the average of 1985-2014, these northern regions were subject to high pressure score increases of up to +3 points for SSP126 and up to +6 points for SSP370 (figure 2, for IPSL scenarios see figure S1). Mild pressure releases of about −1 were found for all scenarios over parts of Mexico. In terms of pressure trends over time, we found roughly linear increases from 2011-2100, again with stronger increases under SSP370 than under 126.
For population pressures, scores differed markedly across the unprotected areas in KBAs ( figure 3). Again, Guatemala appeared as a main hot spot region with pressure scores of around 6-7 for SSP126 and 8-10 for SSP370. Similar scores were also found for the Puebla state in central Mexico and along the foothills of the Costa Rican Central Valley (Greater metropolitan area around the capital San Jose). Compared to the reference period, most areas showed mild pressure increases of up to +2 points with hot spots of change in Guatemala, Honduras and parts of Costa Rica. Notably, these high scores were already present for the time period 2011-2040 and only showed minor increases until 2100 for SSP370. Under SSP126 on the other hand, population pressure hardly changed over time ( figure 2). There, most regions-save for Guatemala-showed mild pressure releases.

Multiple pressures
The results of overlaying biome stability projections with pressures from land use and population density as well as a comparison of the individual pressure factors for the time period 2071-2100 are presented in figure 3 (for additional information see figures S4-S6). Overall, the highest average scores were found for Guatemala and scattered hotspots in Mexico and Costa Rica. These hotspots were similarly distributed under all climate scenarios but showed higher pressure scores for the SSP126 scenarios (up to 6) than for the SSP370 scenarios (up to 8). In general, high population pressures contributed most strongly to the formation of the hot spots. This was followed by land use, which largely coincided with high population pressures. Only for the dry forest KBAs covering the Selva Zoque (pacific coast of Mexico) and the Nicoya peninsula (northwest Costa Rica) pressures from biome shifts were the dominant factor.
The binarization of each pressure factor into 'high' and 'low' pressure revealed further insights regarding the overlap of multiple high pressures (figure S7). Across the range of thresholds tested, all variants showed co-occurrences of high population and land use pressures for Guatemala. For the central threshold variants, high population density impacts prevailed as main pressure source across large parts of the area (30%-44%), followed by land use (16%-27%) and climate-induced pressures (15%-21%). For SSP126 a majority of the area (∼59%) was projected to experience low impacts from all pressures. This figure however decreased to 44%-46% in SSP370. The largest overlaps between pressure factors occurred between population density and land use (13%-24% of the area), whereas high pressures from population density and biome shifts only co-occurred in 1%-4% of the area. Finally, overlaps of high pressures from all three factors at once covered 2%-5% of the area, which covered parts of the Selva Zoque, central parts of Guatemala and the Nicoya peninsula. When lowering the thresholds by 33% these hotspots were similar but even more extended. When increasing the thresholds by 33%, almost no areas highly pressured by all factors at once were left. Still, high pressures from one or two factors remained, even though their area was considerably smaller under SSP126 compared to SSP370.
In order to identify possible increased threats for particular ecosystem types, we further analysed the distribution of high pressures by underlying biomes (as projected for the period 1985-2014). For the central threshold variants, savannas and grasslands emerged as most extensively pressured, being affected by at least one pressure factor in more than 80% of their area under most scenarios (figure 4). On the other hand, the highest overlaps between multiple pressures were found for both montane biomes and in particular for SSP370, where up to 53% of their area may be affected by two or more pressures. Increasing the thresholds by 33% still resulted in extensive pressures for grasslands and savannas, yet showed very little area of overlapping high pressures for all biomes but tropical montane forests. Lowering the thresholds on the other hand resulted in widespread overlaps of at least two pressure factors for all biomes. For tropical montane forests, tropical seasonal forests, savannas and grasslands the share of areas where high pressures from all factors cooccurred even reached about 50% in some scenarios.

Pressure impacts
Our projections showed increased climate-induced pressures over half of the studied KBAs and particularly in unprotected areas (figure S2). Without considering anthropogenic factors this already poses a severe challenge for the expansion of conservation efforts. Under consideration of all pressure factors,  a large share of unprotected areas in KBAs experienced high pressures from at least one pressure factor. The majority of these impacts were based on projections of growing population pressures which largely coincided with land use pressures. Climate-induced pressures on the other hand showed distinct hot spot regions, but may regionally coincide with pressures from land use and population density.
In comparison, Boit et al (2016) simulated land use and climate change impacts on Latin American biomes under SSP1 and SSP5 and attributed a large fraction of biome shifts until 2099 to cropland expansion. Nevertheless, under SSP5, Boit's study also projected regional biome shifts along the Pacific coast and on the Yucatan peninsula which were largely attributed to climate change. Global studies on climate and land use change impacts on biodiversity come to similar conclusions regarding high impact regions, yet with differing attributions (Hof et al 2018, Newbold 2018). Regardless of their impact ranking, however, the interplay of these factors needs to be noted, which may lead to mutual reinforcement. For example, human pressures may contribute to the degradation or homogenization of landscapes, which could in turn reduce climate change resilience of ecosystems. To account for such interactions, a technical integration of climate with land use change in ecological models would be highly beneficial. While issues like data and scale mismatches, high model complexity and propagation of uncertainties currently limit such applications, they remain an important target for future research (de Chazal and Rounsevell 2009).
Beyond the above described general trends, the fine resolution of our results allows a view to specific KBAs within our study region. In this respect, it is concerning, that strong impacts were projected for dry forest KBAs, which are already largely fragmented through human land use and thus particularly vulnerable. While most intact forest landscapes did not count among the pressure hot spots, considerable fractions of their neighbouring areas did, which could lead to edge effects and connectivity issues.

Implications for conservation
Even with information on potential threats to forest ecosystems at hand, the questions of (1) which areas to prioritize and (2) how to enforce protection remain.
As to the question of prioritization, the answer may largely depend on available funds, potential economic trade-offs and the prevalent conservation philosophy. While prioritization criteria are still highly debated in literature (Asaad et al 2017, Maxwell et al 2020, Mokany et al 2020, in the following some of the main concepts shall be discussed for our case. One strategy in this respect aims at the identification of 'safe havens' or 'refugia' , i.e. areas that are projected to remain in a stable condition over an extended period of time (Keppel et al 2012, Morelli et al 2016. Particularly in view of the difficulty of mitigating climate change effects on a local scale, the designation of protected areas in climatically stable regions can represent a useful approach to minimize the risk of biome shifts and thus secure habitat functionality in the long-term. From this perspective, the areas with the lowest overall score in figure 3 (or the grey areas in figure S7) may deserve a high conservation priority.
Another line of argument follows the idea, that particularly strongly pressured areas should be prioritized due to their high exposure (especially regarding anthropogenic impacts) and potential co-benefits resulting from their conservation due to avoided emissions (e.g. Kehoe et al 2017, Williams et al 2021, Sreekar et al 2022. In this respect, particularly areas showing high pressures from land use and population (turquoise areas in figure S7) may represent strong candidates for the expansion of the protected area network. The feasibility of such a conservation scheme, however, largely hinges upon socio-economic considerations. Relatedly, a study by Brancalion et al (2019) investigated both benefits and feasibility of restoration (in relation to possible opportunity costs) across tropical forests and found similar hot spots in Mesoamerica as the hotspots presented in figure 3.
Beyond this, prioritization strategies may also follow the objective to minimize biodiversity loss by focusing on the rarest habitat types, ecological representativeness and the complementarity of expansion areas with existing areas (Dinerstein et al 2020, Hanson et al 2020, Mammides et al 2021. In this respect, our analysis revealed that multiple high pressures may co-occur particularly in montane forest ecosystems (figure 4). This is concerning since the total extent of tropical montane forest within KBAs is already quite small with around 24 000 km 2 . Although tropical dry forests showed lower shares of high pressures, this biome may still count among the most threatened taking into account that only around 27% of their extent in KBAs is currently under protection-the lowest share across all biomes. This is also in line with a study by de Albuquerque et al (2015) who analysed the representativity of ecosystems included in the Mesoamerican protected area network and discovered large gaps, particularly in tropical dry broadleaf forests and regionally in Costa Rican moist forests. The selection of priority sites for conservation may further be narrowed down based on criteria of naturalness, such as the intact forest landscapes classification, and habitat connectivity. Based on our findings (figure S3), large parts of Guatemala, the Maya Forest, Selva Zoque and dry forest KBAs like the Nicoya Peninsula would represent high priority conservation sites also with regards to the above raised points.
While most countries within the study region have met their national Aichi target 11 of protecting at least 17% of their land area by 2020 (with the exceptions of El Salvador with 8.6%, Mexico with 14.5% and Colombia with 16.6%), efforts for expanding protection have considerably slowed down since 2010 (Álvarez Malvido et al 2021). With a view to current global ambitions on protecting 30% of the planet's land and ocean area by 2030 (UNEP 2022), an extension of current conservation efforts is urgently necessary. On the one hand this needs to be facilitated through the identification of priority sites (as done by this study) whereas on the other hand also implementation barriers need to be tackled.
Regarding the latter question, a main difficulty lies in enforcing effective measures. While the legal protection of areas per se attempts to exclude direct anthropogenic pressures, the reality might look different. On the one hand, political inconsistencies and economic pressures may lead to a reversal of the protection status through downgrading, downsizing or degazettement. Even in iconic national parks these phenomena have been observed in the past (Qin et al 2019. On the other hand, forest disturbance and illegal resource extraction through activities such as logging, mining, harvest of non-timber products or hunting may significantly reduce the effectiveness of protected areas (Laurance et al 2012). In Central America, around 23% of protected areas within KBAs correspond to high protection categories (IUCN I-II), 22% fall into the lowest category VI-which allows sustainable resource use-and a large share of 45% were either not applicable or not reported (UNEP-WCMC, IUCN 2020). Respectively, regulations on access and use are vague and not controlled in many cases. Therefore, it is no surprise that numerous examples for disrespected logging bans, wildlife trade or even 'narcodeforestation' (forest clearing for cocaine production and trafficking) are known for the study area (Navarrete et al 2011, Clerici et al 2020, Tellman et al 2020, Wade et al 2020, Gluszek et al 2021. For example, in a case study for the Guatemala Maya Forest Reserve 15%-30% of forests were transformed to agricultural land within only 15 years, largely driven by illegal cattle ranching as part of drug trafficking activities (Devine et al 2020).
On top of this, protected areas have traditionally rather been designated in remote areas with low economic value or poor infrastructure ('rock and ice' , Joppa and Pfaff 2009). Consequently, establishing protected areas in regions with high human pressures may be 'inconvenient' and stimulate land-use conflicts. Any expansion of protected areas to currently unprotected KBAs thus needs to be planned both with regards to monitoring and the enforcement of rules, as well as by engaging and integrating the needs of local communities (Porter-Bolland et al 2012).
In contrast, climate-driven pressures leading towards biome shifts pose a different set of challenges to conservation. The expansion of protected areas into regions with high climatic pressures may diminish potential conservation benefits over time. Nevertheless, releasing additional human pressures from such areas-for example from uncontrolled urban sprawl or extensive land use-could at least avoid rapid transitions. In this respect it should also be noted, that the footprint of human activities can extend far beyond the boundaries of settlements (Hansen and DeFries 2007). Therefore, the introduction of buffer zones around protected areas may be beneficial to reduce 'spill-over effects' from surrounding residential areas, yet this needs to be integrated with protected area management (DeFries et al 2010, de Almeida-Rocha and Peres 2021). While, for example, in Costa Rica large-scale deforestations of the 1980s had been reversed by the early 2000s, increasing monocultures and urban sprawl at the same time reduced overall ecological connectivity by 13% and thus undermined this possible success story (Montero et al 2021). In the end, particularly the connectivity of landscapes may become of utmost importance for conservation planning, both to allow climate-induced shifts and mobility between habitats and to reduce the risk of 'sink habitat' (Baumbach et al 2021).
In conclusion, the protection of natural ecosystems remains a key challenge under climate change induced and anthropogenic pressures. Here, we show that until the end of the 21st century these pressures may increase, overlap and lead to high impacts within biologically important but yet unprotected KBAs. Since many of these areas represent 'the last of their kind' , we call for special focus and prompt conservation actions, which should especially consider questions of ecological connectivity and enforcement of protection status, e.g. by engaging local communities. This study both highlights global change hot spots and provides a basis for prioritization of conservation areas. Therein, particularly dry forest KBAs along the pacific coast and on the Yucatan peninsula and tropical montane forests represent strong candidates.

Data availability statement
All input data is freely available from the sources indicated in the data section of this paper.
The data that support the findings of this study are openly available at the following URL/DOI: https:// zenodo.org/record/7837543 (Baumbach et al 2023).