Projecting terrestrial biodiversity intactness with GLOBIO 4

Abstract Scenario‐based biodiversity modelling is a powerful approach to evaluate how possible future socio‐economic developments may affect biodiversity. Here, we evaluated the changes in terrestrial biodiversity intactness, expressed by the mean species abundance (MSA) metric, resulting from three of the shared socio‐economic pathways (SSPs) combined with different levels of climate change (according to representative concentration pathways [RCPs]): a future oriented towards sustainability (SSP1xRCP2.6), a future determined by a politically divided world (SSP3xRCP6.0) and a future with continued global dependency on fossil fuels (SSP5xRCP8.5). To this end, we first updated the GLOBIO model, which now runs at a spatial resolution of 10 arc‐seconds (~300 m), contains new modules for downscaling land use and for quantifying impacts of hunting in the tropics, and updated modules to quantify impacts of climate change, land use, habitat fragmentation and nitrogen pollution. We then used the updated model to project terrestrial biodiversity intactness from 2015 to 2050 as a function of land use and climate changes corresponding with the selected scenarios. We estimated a global area‐weighted mean MSA of 0.56 for 2015. Biodiversity intactness declined in all three scenarios, yet the decline was smaller in the sustainability scenario (−0.02) than the regional rivalry and fossil‐fuelled development scenarios (−0.06 and −0.05 respectively). We further found considerable variation in projected biodiversity change among different world regions, with large future losses particularly for sub‐Saharan Africa. In some scenario‐region combinations, we projected future biodiversity recovery due to reduced demands for agricultural land, yet this recovery was counteracted by increased impacts of other pressures (notably climate change and road disturbance). Effective measures to halt or reverse the decline of terrestrial biodiversity should not only reduce land demand (e.g. by increasing agricultural productivity and dietary changes) but also focus on reducing or mitigating the impacts of other pressures.


| INTRODUC TI ON
Global biodiversity is threatened by unprecedented and increasing anthropogenic pressures, including habitat loss and fragmentation, overexploitation, climate change and pollution (IPBES, 2019;Maxwell, Fuller, Brooks, & Watson, 2016;Tilman et al., 2017). This has prompted a proliferation of international commitments and agreements striving to halt biodiversity loss. Prominent examples include the Aichi biodiversity targets (CBD, 2010) and the more recent sustainable development goals (UN General Assembly, 2015), which encompass targets for biodiversity as well as human well-being, thus underlining that these are interlinked. To deliver on these ambitious goals, decision-making needs to be supported by a solid understanding of current trends in biodiversity as well as the effects of future changes in drivers and pressures. Scenario-based biodiversity modelling is indispensable to systematically evaluate the impacts of current and future drivers and pressures on biodiversity and assess the effectiveness of possible conservation measures (IPBES, 2016;Kok et al., 2017;Pereira et al., 2010).
The recently developed shared socio-economic pathways (SSPs) comprise a set of five diverging plausible future scenarios of human development and associated changes in the environment Riahi et al., 2017). The SSPs are a combination of qualitative descriptions ('narratives') and model-based quantifications of potential trends, such as expected human population growth or economic development. The narratives provide the logic and internal consistency of the scenarios and include possible trends in relevant drivers that are more difficult to project quantitatively, such as political stability, environmental awareness and lifestyle Riahi et al., 2017). The SSPs have already been elaborated in terms of, among others, energy, greenhouse gas emissions and land use (Popp et al., 2017;Riahi et al., 2017). Recently, a protocol has been developed to quantify the SSPs also in terms of biodiversity and ecosystem services, based on harmonized land use and climate change input data and a suite of complementary biodiversity and ecosystem models (Kim et al., 2018), for supporting the global assessment of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES, 2019). Following this protocol, we assessed the implications of three SSPs for terrestrial biodiversity intactness in 2050. To that end, we used an updated version of the GLOBIO model: a global model of biodiversity intactness, expressed by the mean species abundance (MSA) metric, as a function of multiple anthropogenic pressures on the environment (Alkemade et al., 2009).
An important strength of the GLOBIO model is the breadth of pressures it considers. Originally developed to quantify the impacts of infrastructure on biodiversity intactness (Nellemann et al., 2001), it was later extended to also include the impacts of climate change, land use (via both habitat loss and fragmentation) and atmospheric nitrogen deposition (Alkemade et al., 2009). GLOBIO quantifies biodiversity using the MSA metric, which is a measure of local biodiversity intactness conceptually similar to the biodiversity intactness index (Scholes & Biggs, 2005). For this scenario analysis, we have introduced several new or updated model features to the GLOBIO version as described by Alkemade et al. (2009) andSchipper, Bakkenes, Meijer, Alkemade, andHuijbregts (2016). Because current global-scale landuse models are relatively coarse-grained and tend to underestimate the spatial heterogeneity of land-use patterns (Hoskins et al., 2016), we have developed a routine to downscale land-use data to discrete global maps with a spatial resolution of 10 arc-seconds. This enhances the possibility to account for spatial heterogeneity and ecological effects that depend on the spatial configuration of the landscapenotably habitat fragmentation. Furthermore, the model now allows for quantifying the impacts of hunting in tropical regions, where it is a major pressure (Benítez-López et al., 2017). Finally, we used updated versions of the modules to quantify the impacts of climate change, land use, habitat fragmentation and atmospheric nitrogen deposition based on updated and extended datasets.
For the scenario analysis, we followed the recently developed biodiversity model intercomparison protocol as described by Kim et al. (2018) by coupling three SSPs (i.e. SSP1, SSP3 and SSP5) with three representative concentration pathways (RCPs; i.e. RCP2.6, RCP6.0 and RCP8.5 respectively), which describe different climate futures based on greenhouse gas emissions throughout the 21st century (van Vuuren et al., 2011). The combinations of SSPs and RCPs allowed us to explore a future with relatively low land-use change and climate change (SSP1xRCP2.6) as well as futures with high levels of land use or climate change (SSP3xRCP6.0, SSP5xRCP8.5; see Section 2 for further details on the scenarios). We used projections of climate change, land-use change and atmospheric nitrogen deposition corresponding to each of the three SSPxRCP combinations as inputs to GLOBIO in order to project the changes in terrestrial biodiversity intactness from 2015 to 2050 and identify the main pressures and drivers underlying these changes.

| General approach
The core of the GLOBIO model is a set of quantitative relationships that assess the impacts of anthropogenic pressures on biodiversity. Pressures included in GLOBIO are climate change, land use, roads, atmospheric nitrogen deposition and hunting (Figure 1a). Impacts are quantified based on the MSA metric, which is an indicator of local biodiversity intactness (Figure 1b). The metric is quantified based on data that describe changes in community composition in relation to particular pressures. MSA values are retrieved by dividing the abundance of each species found in relation to a given pressure level by its abundance found in an undisturbed situation within the same study, truncating the values at 1, and then calculating the arithmetic mean over all species present in the reference situation (Alkemade et al., 2009;Schipper, Bakkenes, et al., 2016). Increases in individual species abundance from reference to impacted situation are truncated to avoid the indicator being inflated by opportunistic or generalist species that benefit from habitat disturbance. The GLOBIO model combines the pressure-impact relationships with maps of the pressures (i.e.  (Table S1). If the land use impact is expected to dominate over other impacts, the overall MSA value equals the MSA value of the land-use class.
For example, it is assumed that there are no additional impacts of atmospheric nitrogen deposition in croplands, which are typically fertilized, and that there are no additional impacts or roads within urban areas. Alternatively, it is assumed that (a) pressures act independently, that is, an organism is lost from the community if at least one of the pressures is higher than its tolerance limit; where P x,s,i is the contribution of pressure x to the loss in MSA for species group s in grid cell i. The rescaling is applied to ensure that the sum of the pressure-specific losses in MSA equals the total loss of MSA in the grid cell. Subsequently, the cell-specific MSA losses (Equation 1) and pressure contributions (Equation 2) can be aggregated across the grid cells 1 to n in any larger region of interest, calculated as mean value weighted by the area of the grid cells.

| Pressure-impact relationships
To update the pressure-impact relationships in GLOBIO, we used spatially explicit data on species' abundances in relation to different levels or intensities of each pressure ( Figure S1). For climate change, nitrogen deposition, road disturbance and hunting, we used databases that were specifically collected for this purpose (Benítez-López et al., 2017;Benítez-López, Alkemade, & Verweij, 2010;Benítez-López, Santini, Schipper, Busana, & Huijbregts, 2019;Midolo et al., 2019;Nunez, Arets, Alkemade, Verwer, & Leemans, 2019). For land use and habitat fragmentation, we used the PREDICTS database, which includes spatial comparisons of species' assemblages among different land-use types and habitat patch sizes (Hudson et al., 2017). For each pressure, we collected or selected the data such that influences of other pressures could be considered negligible or equal between control and treatment. For each dataset and pressure level or intensity, we first calculated species-specific abundance ratios by dividing each species' abundance in the disturbed situation by its abundance in the for a given pressure if the set included zeros or ones (Smithson & Verkuilen, 2006). Because MSA is an assemblage-level metric, we weighted the observations based on the number of species sampled (square-root transformed to reduce the skewness in the data). In this study we included impact relationships for terrestrial plants and warm-blooded vertebrates (birds and mammals), because the majority of the monitoring data is on these two species groups. For each group, we included only the impacts assumed to be the most relevant, that is, we included climate change, nitrogen deposition and land use for plants, and climate change, land use, infrastructure disturbance, fragmentation and hunting for warm-blooded vertebrates ( Figure 2). Where possible based on the data available, we tested for the influence of potential moderators on the impact relationships (e.g. the influence of climate zone on the climate change impact relationships) and identified the most parsimonious model based on the Bayesian information criterion (BIC). We preferred BIC over alternative approaches to model selection (e.g. Akaike information criterion) in order to minimize the risk of overfitting. We performed all data processing and model fitting in the R environment (R Core Team, 2017), including the glmmTMB package for beta regression modelling (Brooks et al., 2017). Further methodological details on the fitting of the specific pressure-impact relationships are provided in Text section S1.

| Land-use downscaling
Because current global land-use models are relatively coarsegrained and tend to underestimate the spatial heterogeneity of landuse patterns (Hoskins et al., 2016), we have extended GLOBIO with a routine to downscale coarse-grain land-use data to discrete maps with a spatial resolution of 10 arc-seconds. The land-use downscaling procedure requires three inputs: regional totals or demands ('claims') of each land-use type; map layers quantifying the suitability of each grid cell for each land-use type; and a 'background' map defining the land cover or land use of cells that are not being converted for fulfilling the claims. Claims can be derived from national or regional statistics or from models that estimate demands for land based on socio-economic developments, such as integrated assessment models. All claims need to be expressed in terms of area (km 2 ).

| Scenarios
Following the recently developed biodiversity model intercomparison protocol (Kim et al., 2018), we used three SSPs associated with different levels of human pressure on the environment: SSP1 ('sustainability'), SSP3 ('regional rivalry') and SSP5 ('fossil-fuelled development'). The sustainability scenario is characterized by a relatively low population growth, low growth in consumption due to less resource-intensive lifestyles (e.g. less meat) and more resourceefficient technologies, increased regulation of land-use change due to expansion of the protected area network, and substantial improvements in agricultural productivity, allowing for reforestation. The regional rivalry scenario is characterized by high population growth, resource-intensive consumption, low agricultural productivity and limited regulation of land-use change, leading to continued deforestation. Finally, the fossil-fuelled development scenario is characterized by low population growth, strong economic growth, a consumption-oriented and energy-intensive society, and highly intensive agricultural practices leading to a decline in deforestation.
We combined the SSPs with climate projections according to the RCPs such that the combinations covered a broad range of land-use and climate change, following the biodiversity model intercomparison protocol (Kim et al., 2018

| Pressure input data
We retrieved the global mean temperature increase since 1970 (in °C) for 2015 and for each selected RCP for 2050 from the MAGICC climate model, which is part of the IMAGE model framework (Meinshausen, Raper, & Wigley, 2011;Stehfest et al., 2014). We retrieved nitrogen deposition data (kg ha −1 year −1 ; 0.5° resolution) for each scenario-year combination also from IMAGE. To compile the land-use maps, we used the newly implemented land-use allocation module. We first established suitability layers for urban land and cropland based on the distance to existing urban and cropland areas, for pasture based on livestock densities, and for forestry based on existing forest cover, elevation and distance to roads and rivers (Text section S2). The suitability of natural land cover within protected areas was set to zero in order to limit land-use expansion within these areas. Next, we compiled a land-  lack of information on new future settlements, we used the presentday settlement data also for the scenario projections. An overview of the extent to which the pressure input data are covered by the data used to fit the pressure-impact relationships is provided in Figure S2.

| Projected biodiversity changes
For 2015, we estimated a global area-weighted mean MSA of 0.56 (Table S2). Future projections resulted in an overall decrease in MSA for all three scenarios (Figure 3; Table S2). The global area-weighted mean MSA value was projected to decline by 0.02 in the sustainability scenario (SSP1xRCP2.6), by 0.06 in the regional rivalry scenario (SSP3xRCP6.0) and by 0.05 in the fossil-fuelled development scenario (RCP5xRCP8.5) (Table S2) Tables S2 and S3). On average, we found the largest projected declines for East Africa, Central Africa and Southern Africa in the regional rivalry scenario. The smallest declines occurred in North-East Asia (sustainability and regional rivalry scenarios) and North-America (sustainability scenario). Spatial patterns for plants and warm-blooded vertebrates were largely similar, although in the regional rivalry and fossil-fuelled development scenarios we found larger declines in MSA for plants than for vertebrates particularly in the boreal and Arctic regions (Figures S3 and S4).   Table S2; 5th and 95th percentiles per IPBES region are given in Table S3 of land use and nitrogen deposition showed clear spatial variability, with impacts decreasing in some scenario-region combinations and increasing in others ( Figure 5). Large increases in land-use impacts were projected for Central Africa, East Africa and Southern Africa in the regional rivalry scenario. In the sustainability scenario, the majority of the regions was characterized by a decline in land-use impacts.

| Global trends
Our projections indicate that biodiversity intactness will decline from present-day to 2050, even in the most optimistic scenario evaluated. These declines comply with a mid-term analysis of progress towards the Aichi biodiversity targets for 2020, which revealed that pressure indicators were mostly on a continuing increasing trend, while biodiversity indicators pointed to a continuing decline (Tittensor et al., 2014). Projected area-weighted global mean losses in MSA were similar for the regional rivalry (SSP3xRCP6.0) and fossil-fuelled development (SSP5xRCP8.5) scenarios. Compared to the fossil-fuelled development scenario, the regional rivalry scenario is characterized by a larger increase in global human population and a smaller increase in agricultural production efficiency (KC & Lutz, 2017;Popp et al., 2017), leading to an overall higher demand for agricultural land (Table S5). However, climate change impacts were larger in the fossil-fuelled development scenario, due to larger projected increases in global mean temperature, leading to similar overall biodiversity losses in the two scenarios, albeit via different pressures ( Figure 5). The results of the sustainability scenario (SSP1xRCP2.6) indicate that biodiversity declines may slow down in response to a decreasing demand for agricultural land. Values represent the mean across plants ( Figure S3) and warm-blooded vertebrates ( Figure S4). For visualization purposes, the maps were resampled to a resolution of 0.25 degree based on the mean across the underlying values The sustainability scenario is characterized by a global decline of ~5% in agricultural land area in 2050 (Table S5) (Table S5), which is characterized by higher MSA values than the agricultural land-use types (Figure 2), thus yielding a partial restoration of biodiversity intactness. Interestingly, the global human population projected for 2050 is highly similar between the sustainability scenario (8.5 billion people) and the fossil-fuelled development scenario (8.6 billion people; KC & Lutz, 2017), whereas the latter is characterized by an increase rather than a decline in agricultural land area (+2.5% worldwide; Table S5). The comparison between these two scenarios thus highlights the importance of changes in both production and consumption of agricultural prod-  Table S4 4

.2 | Pressure contributions
We found that land use is currently the dominant pressure on terrestrial biodiversity, exceeding the present-day impacts of hunting, climate change and pollution. This is line with other recent analyses that ranked pressures affecting community composition and species' populations (IPBES, 2019;Maxwell et al., 2016;Newbold, 2018).
We note that our assessment may underestimate the present-day impacts of hunting because there might be more settlements and other relevant hunters' access points than included in our input data.
Moreover, we assessed the hunting impacts only for the tropics, due to a lack of data to include other regions. On the other hand, the pressure-impact relationship for hunting may overestimate the impacts because the underlying observations are biased towards mediumand large-sized species, which comprise the majority of our data (Text section S1), and which are more heavily hunted than small-sized species (Benítez-López et al., 2017;Ripple et al., 2016). Impacts of fragmentation might be underestimated because our pressure-impact relationship assumes that such impacts are absent in natural habitat patches larger in size than 10,000 ha (see Text section S1), due to insufficient biodiversity monitoring data including larger reference patches. Although 10,000 ha might be large enough to fulfil the minimum area requirements of small and herbivorous bird and mammal species, it is likely too small for minimum viable populations of large carnivores (Pe'er et al., 2014), and hence a fully intact community.
This implies that the effects of fragmentation, and thus land use as one of the underlying causes, could be larger than assessed here.
Our scenario projections suggest that land use will also be the most important cause of biodiversity loss in 2050. This is consistent with the projections of Sala (2000), but in contrast to studies indicating that impacts of climate change on biodiversity may have exceeded land-use impacts halfway this century (Di Marco et al., 2019;Newbold, 2018). It is notoriously difficult to quantify the effects of future climate change in comparison to the impacts of other threats, reflecting model as well as data limitations (Newbold, 2018;Tingley, Estes, & Wilcove, 2013). The pressure-impact relationships used in this study are based on relative species richness estimates retrieved from bioclimatic envelope modelling results rather than observational data of MSA (Text section S1), due to a lack of local biodiversity monitoring data across sufficiently wide climate gradients. Moreover, we considered global mean temperature increase only, thus ignoring the possible changes in seasonality or extremes as well as latitudinal differences in the magnitude of climatic change. More research on the effects of climate change on biodiversity intactness is urgently needed to further improve the GLOBIO model. We further note that our projections do not account for possible increases in future hunting impacts due to the establishment of new settlements, in absence of a settlement expansion model. Similarly, impacts of future roads might be underestimated because we could not account for the future construction of new roads. Although improvements and increased use intensity of existing roads, as assumed in our projections, typically precede the construction of new roads (Dulac, 2018;Kerali, 2003), future increases in road network length are also expected. Recent projections for 2050 suggested increases in 14%-23% of the global road network, as a function of country-specific estimates of human population and gross domestic product according to the SSP framework (Meijer et al., 2018). More work is needed to more accurately assess the biodiversity impacts resulting from future hunting pressure as well as road expansion. Further work is also required to enable the GLOBIO model to account for possible synergistic or antagonistic interactions between different pressures, which may lead to larger or smaller pressure contributions than expected based on their individual impacts (Brook, Sodhi, & Bradshaw, 2008;Darling & Cote, 2008). As an example, hunting impacts may be exacerbated by habitat loss and fragmentation, because remaining fragments are more accessible to hunters and isolation may reduce the recolonization from non-hunted source populations (Peres, 2001).

| Regional differences
Our results showed large spatial variation in local biodiversity in- typically co-occur, with spatial patterns primarily driven by the suitability of land for agriculture (Venter et al., 2016). Our projections revealed that further biodiversity declines are expected in some regions irrespective of the scenario, notably in sub-Saharan Africa. In contrast, the sustainability scenario projections suggest that losses might be substantially lower or even halted in other regions, for example in North-East Asia (Table S2), mainly due to considerable decreases in land demand. The differences in pressures and projected biodiversity changes among different world regions point towards the need for a more differentiated approach to improve large-scale scenario analyses, in particular when it comes to target-seeking rather than exploratory scenarios (Rosa et al., 2017). Differential targets may be needed depending on the feasibility to reduce anthropogenic drivers and pressures in different contexts. For example, reversing trends of biodiversity loss might be feasible in Europe, where human population is projected to decline (KC & Lutz, 2017).
However, targets for sub-Saharan Africa may need to be different (for example, no or limited further loss) in order to ensure feasibility, given the considerable projected increase in human population (KC & Lutz, 2017) and other sustainable development goals to be attained (UN General Assembly, 2015). Similarly, region-specific measures could be proposed. For example, measures to reduce food waste could be targeted at final consumers in wealthier regions, while a focus on reducing on-field post-harvest losses could be more attainable in sub-Saharan Africa and South and Southeast Asia (Kok et al., 2018). In addition, impacts need to be quantified for different complementary dimensions of biodiversity, particularly because responses to environmental change may differ among metrics and scales (McGill, Dornelas, Gotelli, & Magurran, 2015;Santini et al., 2017;Schipper, Belmaker, et al., 2016). For example, the MSA metric in GLOBIO does not account for spatial differences in species richness and may therefore miss out on the disproportional impacts in tropical regions in terms of numbers of species lost (Barlow et al., 2018). Similarly, aspects of spatial turnover (beta diversity) are not included in GLOBIO; hence signals of biotic homogenization or heterogenization are not picked up (Socolar, Gilroy, Kunin, & Edwards, 2016). A suite of complementary biodiversity models, combined with scenario settings better tailored to the regional and local context, is needed to further improve scenario-based biodiversity modelling (Kim et al., 2018;Rosa et al., 2017). Ensemble and probabilistic modelling approaches are recommended to account for model and parameter uncertainties, which were not accounted for in the present study, including the significant uncertainties in underlying climate and land-use projections (Stehfest et al., 2019;Thuiller, Gueguen, Renaud, Karger, & Zimmermann, 2019). These improvements are urgently needed in order to better inform decision-making aimed at safeguarding biodiversity.

ACK N OWLED G EM ENTS
This study was part of the GLOBIO project (www.globio.info). We