Understanding how anthropogenic activities impact biodiversity and human societies is essential for nature conservation and sustainable development. Land-use and climate change are widely recognized as two of the main drivers of future biodiversity change (Hirsch and CBD, 2010; Maxwell et al., 2016; Sala, 2000; Secretariat of the CBD and UNEP, 2014), with potentially severe impacts on ecosystem services and ultimately human well-being (Cardinale et al., 2012; MEA, 2005). Habitat and land-use changes, resulting from past, present, and future human activities, as well as climate change, have both immediate and long-term impacts on biodiversity and ecosystem services (Graham et al., 2017; Lehsten et al., 2015; Welbergen et al., 2008). Therefore, current and future land-use projections are essential elements for assessing biodiversity and ecosystem change (Titeux et al., 2016, 2017). Climate change has already been observed to have direct and indirect impacts on biodiversity and ecosystems, which are projected to intensify by the end of the century, with potentially severe consequences for species and habitats, and, therefore, also for ecosystem functions and services (Pecl et al., 2017; Settele et al., 2015).

Global environmental assessments, such as the Millennium Ecosystem Assessment (MEA, 2005), the Global Biodiversity Outlooks (GBO), the multiple iterations of the Global Environmental Outlook (GEO), the Intergovernmental Panel on Climate Change (IPCC), and other studies have used scenarios to assess the impact of socio-economic development pathways on land use and climate and their consequences for biodiversity and ecosystem services (Jantz et al., 2015; Pereira et al., 2010). Models are used to quantify the biodiversity and ecosystem services impacts of different scenarios, based on climate and land-use projections from general circulation models (GCMs) and integrated assessment models (IAMs) (Pereira et al., 2010). These models include empirical dose–response models, species–area relationship models, species distribution models and more mechanistic models such as trophic ecosystem models (Pereira et al., 2010; Akçakaya et al., 2015). So far, each of these scenario exercises has been based on a single model or a small number of biodiversity and ecosystem services models, and intermodel comparison and uncertainty analysis have been limited (IPBES, 2016; Leadley et al., 2014). The Expert Group on Scenarios and Models of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) is addressing this gap by carrying out a biodiversity and ecosystem services model intercomparison with harmonized scenarios, for which this paper lays out the protocol.

Over the past 2 decades, IPCC has fostered the development of global scenarios to inform climate mitigation and adaptation policies. The Representative Concentration Pathways (RCPs) describe different climate futures based on greenhouse gas emissions throughout the 21st century (van Vuuren et al., 2011). These emissions pathways have been converted into climate projections in the most recent Climate Model Inter-comparison Project (CMIP5). In parallel, the climate research community also developed the Shared Socio-economic Pathways (SSPs), which consist of trajectories of future human development with different socio-economic conditions and associated land-use projections (Popp et al., 2017; Riahi et al., 2017). The SSPs can be combined with RCP-based climate projections to explore a range of futures for climate change and land-use change, and they are being used in a wide range of impact modeling intercomparisons (Rosenzweig et al., 2017; van Vuuren et al., 2014). Therefore, the use of the SSP-RCP framework for modeling the impacts on biodiversity and ecosystem services provides an outstanding opportunity to build bridges between the climate, biodiversity and ecosystem services communities; it has been explicitly recommended as a research priority in the IPBES assessment on scenarios and models (IPBES, 2016).

Model intercomparisons bring together different communities of practice for comparable and complementary modeling, in order to improve the comprehensiveness of the subject modeled, and to estimate uncertainties associated with scenarios and models (Frieler et al., 2015). In the last decades, various model intercomparison projects (MIPs) have been initiated to assess the magnitude and uncertainty of climate change impacts. For instance, the Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP) was initiated in 2012 to quantify and synthesize climate change impacts across sectors and scales (Rosenzweig et al., 2017; Warszawski et al., 2014). The ISI-MIP aims to bridge sectors such as agriculture, forestry, fisheries, water, energy, and health with global circulation models, Earth system models (ESMs), and integrated assessment models for more integrated and impact-driven modeling and assessment (Frieler et al., 2017).

Here, we present the methodology used to carry out a BES-SIM in both terrestrial and freshwater ecosystems. The BES-SIM project addresses the following questions. (1) What are the projected magnitudes and spatial distribution of biodiversity and ecosystem services under a range of land-use and climate future scenarios? (2) What is the magnitude of the uncertainties associated with the projections obtained from different scenarios and models? Although independent of the ISI-MIP, the BES-SIM has been inspired by ISI-MIP and other intercomparison projects and was initiated to address the needs of the global assessment of IPBES. We brought together 10 biodiversity models and six ecosystem functions and services models to assess impacts of land-use and climate change scenarios in the coming decades (up to 2070) and to hindcast changes to the last century (to 1900). The modeling approaches differ in several respects concerning how they treat biodiversity and ecosystem services responses to land-use and climate changes, including the use of correlative, deductive, and process-based approaches, and in how they treat spatial-scale and temporal dynamics. We assessed different classes of essential biodiversity variables (EBVs), including species populations, community composition, and ecosystem function, as well as a range of measures on ecosystem services such as food production, pollination, water quantity and quality, climate regulation, soil protection, and pest control (Pereira et al., 2010; Akçakaya et al., 2015). This paper provides an overview of the scenarios, models and metrics used in this intercomparison, and thus a roadmap for further analyses that is envisaged to be integrated into the first global assessment of the IPBES (Fig. 1).

All the models included in BES-SIM used the same set of scenarios with particular combinations of SSPs and RCPs. In the selection of the scenarios, we applied the following criteria: (1) data on projections should be readily available, and (2) the total set should cover a broad range of land-use change and climate change projections. The first criterion entailed the selection of SSP-RCP combinations that are included in the ScenarioMIP protocol as part of CMIP6 (O'Neill et al., 2016), as harmonized data were available for these runs and they form the basis of the CMIP climate simulations. The second criterion implied a selection of scenarios with low and high degrees of climate change and different land-use scenarios within the ScenarioMIP set. Our final selection was SSP1 with RCP2.6 (moderate land-use pressure and low level of climate change) (van Vuuren et al., 2017), SSP3 with RCP6.0 (high land-use pressure and moderately high level of climate change) (Fujimori et al., 2017), and SSP5 with RCP8.5 (medium land-use pressure and very high level of climate change) (Kriegler et al., 2017), thus allowing us to assess a broad range of plausible futures (Table 1). Further, by combining projections of low and high anthropogenic pressure on land use with low and high levels of climate change, we can test these drivers' individual and synergistic impacts on biodiversity and ecosystem services.

The first scenario (SSP1xRCP2.6) is characterized by a relatively “environmentally friendly world” with low population growth, high urbanization, relatively low demand for animal products, and high agricultural productivity. These factors together lead to a decrease in the land use of around 700 Mha globally over time (mostly pastures). This scenario is also characterized by low air pollution, as policies are introduced to limit the increase in greenhouse gases in the atmosphere, leading to an additional forcing of 2.6 W m-2 before 2100. The second scenario (SSP3xRCP6.0) is characterized by “regional rivalry”, with high population growth, slow economic development, material-intensive consumption, and low food demand per capita. Agricultural land intensification is low, especially due to the very limited transfer of new agricultural technologies to developing countries. This scenario has minimal land-use change regulation, with a large land conversion for human-dominated uses, and a relatively high level of climate change with a radiative forcing of 6.0 W m-2 by 2100. The third scenario (SSP5xRCP8.5) is a world characterized by “strong economic growth” fuelled by fossil fuels, with low population growth, high urbanization, and high food demand per capita but also high agricultural productivity. As a result, there is a modest increase in land use. Air pollution policies are stringent, motivated by local health concerns. This scenario leads to a very high level of climate change with a radiative forcing of 8.5 W m-2 by 2100. Full descriptions of each SSP scenario are provided in Popp et al. (2017) and Riahi et al. (2017). The SSP scenarios excluded elements that have interaction effects with climate change except for SSP1, which focuses on environmental sustainability. Thus, SSPs describe futures where biodiversity is not affected by climate change to allow for the important estimation of the climate change impact on biodiversity (O'Neill et al., 2014).

A consistent set of land-use and climate data was implemented across the models to the extent possible. All models in BES-SIM used the newly released Land Use Harmonization version 2 dataset (LUH2, Hurtt et al., 2018). For the models that require climate data, we selected the climate projections of the past, present, and future from CMIP5/ISIMIP2a (McSweeney and Jones, 2016) and its downscaled version from the WorldClim (Fick and Hijmans, 2017), as well as MAGICC 6.0 (Meinshausen et al., 2011a, b) from the IMAGE model for GLOBIO models (Table 2). A complete list of input datasets and variables used by the models is documented in Table S1 of the Supplement.

Biodiversity and ecosystem services models at the global scale have increased in number and improved considerably over the last decade, especially with the availability of biodiversity data and advancement in statistical modeling tools and methods (IPBES, 2016). In order for a model to be included in BES-SIM, it had either to be published in a peer-reviewed journal or adopt published methodologies, with modifications made to modeling sufficiently documented and accessible for review (Table S2). Sixteen models were included in BES-SIM (Appendix A, details on modeling methods in Table S2). These models were mainly grouped into four classes: species-based, community-based, and ecosystem-based models of biodiversity, and models of ecosystem functions and services. The methodological approaches, the taxonomic or functional groups, the spatial resolution and the output metrics differ across models (Appendix A). All 16 models are spatially explicit, with 15 of them using land-use data as an input and 13 of them requiring climate data. We also used one model, BIOMOD2 (Thuiller, 2004; Thuiller et al., 2009), to assess the uncertainty of climate range projections without the use of land-use data.

Given the diversity of modeling approaches, a wide range of biodiversity and ecosystem services metrics can be produced by the model set (Table S2). For the biodiversity model intercomparison analysis, three main categories of common output metrics were reported over time: extinctions as absolute change in species richness (N, number of species) or as proportional species richness change (P, % species), abundance-based intactness (I, % intactness), and mean proportional change in suitable habitat extent across species (H, % suitable habitat) (Table 4). These metrics were calculated at two scales: local or grid cell (α scale, i.e., the value of the metric within the smallest spatial unit of BES-SIM which is the grid cell) and regional or global scale (γ scale, i.e., the value of the metric for a set of grid cells comprising a region). For species richness change, some models project the α metrics at the grid cell level (e.g., species-based and SAR-based community models), while others average the local point values of the metrics across the grid cell weighted by the area of the different habitats in the cell (e.g., PREDICTS, GLOBIO). In addition, some models only provided α values while others provided both α and γ values (Table 4). For the models that can project γ metrics, both regional-γ for each IPBES regions (Table 1 in Brooks et al., 2016; UNEP-WCMC, 2015) and a global-γ were reported.

The species diversity change metrics measured as absolute number or percentage change in species richness show species persistence and extinction in a given time and place. Absolute changes in species richness and proportional species richness change are interrelated and may be calculated from reporting species richness over time, as Nt=St-St0 and P=Nt/St0, where St is the number of species at time t. Most models reported one or both types of species richness metrics (Table 4). The abundance-based intactness (I) measures the mean species abundance in the current community relative to the abundances in a pristine community. This metric is available only for two community-based models: GLOBIO (where intactness is estimated as the arithmetic mean of the abundance ratios of the individual species, whereby ratios >1 are set to 1) and PREDICTS (where intactness is estimated as the ratios of the sum of species abundances). The habitat change (H) measures cell-wise changes in available habitat for the species. It represents the changes in the suitable habitat extent of each species relative to a baseline, i.e., (Ei,t-Ei,t0)/Ei,t0, where Ei,t is the suitable habitat extent of species i at time t within the unit of analysis. It is reported by averaging across species occurring in each unit of analysis (grid cell, region, or globe), and is provided by the species-level models (i.e., AIM-biodiversity, InSiGHTS, MOL) (Table 4). The baseline year, t0, used to calculate changes for the extinction and habitat extent metrics, was the first year of the simulation (in most cases t0=1900; see Table 5).

For ecosystem functions and services, each model's output metrics were mapped onto the new classification of Nature's Contributions to People (NCP) published by the IPBES scientific community (Díaz et al., 2018). Among the 18 possible NCPs, the combination of models participating in BES-SIM was able to provide measures for 10 NCPs, including regulating metrics on pollination (e.g., proportion of agricultural lands whose pollination needs are met,  % agricultural area), climate (e.g., vegetation carbon, total carbon uptake and loss, MgC), water quantity (e.g., monthly runoff, Pg month-1), water quality (e.g., nitrogen and phosphorus leaching, PgN s-1), soil protection (e.g., erosion protection, 0–100 index), hazards (e.g., costal vulnerability, unitless score; flood risk, number of people affected) and detrimental organisms (e.g., fraction of cropland potentially protected by the natural pest relative to all available cropland, km2), and material metrics on bioenergy (e.g., bioenergy–crop production, PgC yr-1), food and feed (e.g., total crop production, 109 KCal) and materials (e.g., wood harvest, KgC) (Table 6). Some of these metrics require careful interpretation in the context of NCPs (e.g., an increase in flood risk can be caused by climate change and/or by a reduction of the capacity of ecosystems to reduce flood risk) and additional translation of increasing or declining measures of ecosystem functions and services (e.g., food and feed, water quantity) into contextually relevant information (i.e., positive or negative impacts) on human well-being and quality of life. Given the disparity of metrics across models within each NCP category, names of the metrics are listed in Table 6, and units, definitions, and methods are provided in Table S3.

The simulations for BES-SIM required a minimum of two outputs from the modeling teams: present (2015) and future (2050). Additionally, a past projection (1900) and a further future projection (2070) were also provided by several modeling teams. Some models projected further into the past and also at multiple time points from the past to the future (Appendix A). Models that simulated a continuous time series of climate change impacts provided 20-year averages around these mid-points to account for inter-annual variability. The models ran simulations at their original spatial resolutions (Appendix A), and upscaled results to 1∘ grid cells using arithmetic means. In order to provide global or regional averages of the α or grid cell metrics, the arithmetic mean values across the cells of the globe or a certain region were calculated, as well as percentiles of those metrics. Both 1∘ rasters and a table with values for each IPBES region and the globe were provided by each modeling team for each output metric.

To measure the individual and synergistic impacts of land-use and climate change on biodiversity and ecosystem services, models accounting for both types of drivers were run three times: with land-use change only, with climate change only, and with both drivers combined. For instance, to measure the impact of land use alone, the projections into 2050 were obtained while retaining climate data constant from the present (2015) to the future (2050). Similarly, to measure the impact of climate change alone, the climate projections into 2050 (or 2070) were obtained while retaining the land-use data constant from the present (2015) to the future (2050). Finally, to measure the impact of land-use and climate change combined, models were run using projections of both land-use and climate change into 2050 (or 2070). When models required continuous climate time-series data to hindcast to 1900, data from years in the time period 1951 to 1960 were randomly selected to fill the data missing for years 1901 to 1950 from the ISIMIP 2a IPSL dataset. Models that used multi-decadal climate averages from WorldClim (i.e., InSiGHTS, BILBI) assumed no climate impacts for 1900.

Reporting uncertainty is a critical component of model intercomparison exercises (IPBES, 2016). Within BES-SIM, uncertainties were explored by each model reporting the mean values of its metrics, and where possible the 25th, 50th, and 75th percentiles based on the parameterization set specific to each model, which can be found in each model's key manuscripts describing the modeling methods. When combining the data provided by the different models, the average and the standard deviations of the common metrics were calculated (e.g., intermodel average and standard deviation of Pγ). In a parallel exercise to inform BES-SIM, the BIOMOD2 model was used in assessing the uncertainty in modeling changes in species ranges arising from using different RCP scenarios, different GCMs, a suite of species distribution modeling algorithms (e.g., random forest, logistic regression), and different species dispersal hypotheses.

The existing SSP and RCP scenarios provide a consistent set of past and future projections of two major drivers of terrestrial and freshwater biodiversity change – land use and climate. However, we acknowledge that these projections have certain limitations. These include limited consideration of biodiversity-specific policies in the storylines (only the SSP1 baseline emphasizes additional biodiversity policies) (O'Neill et al., 2016; Rosa et al., 2017), coarse spatial resolution, and land-use classes that are not sufficiently detailed to fully capture the response of biodiversity to land-use change (Harfoot et al., 2014a; Titeux et al., 2016, 2017). The heterogeneity of models and their methodological approaches, as well as additional harmonization of metrics of ecosystem functions and services (Tables 6, S3), are areas for further work. In the future, it will also be important to capture the uncertainties associated with input data, with a focus on uncertainty in land-use and climate projections resulting from differences among IAMs and GCMs on each scenario (Popp et al., 2017). The gaps identified through BES-SIM and future directions for research and modeling will be published separately, as well as analyses of the results on the model intercomparison and on individual models. As a long-term perspective, BES-SIM is expected to provide critical foundation and insights for the ongoing development of nature-centred, multiscale Nature Futures scenarios (Rosa et al., 2017). Catalyzed by the IPBES Expert Group on Scenarios and Models, this new scenario and modeling framework will shift traditional ways of forecasting impacts of society on nature to more integrative, biodiversity-centred visions and pathways of socio-economic and ecological systems. A future round of BES-SIM could use these biodiversity-centred storylines to project dynamics of biodiversity and ecosystem services and associated consequences for socio-economic development and human well-being. This will help policymakers and practitioners to collectively identify pathways for sustainable futures based on alternative biodiversity management approaches and assist researchers in incorporating the role of biodiversity into socio-economic scenarios.