IIASA modeling framework
RCP8.5 was developed using the IIASA Integrated Assessment Modeling Framework that encompasses detailed representations of the principal GHG-emitting sectors—energy, industry, agriculture, and forestry. The framework combines a careful blend of rich disciplinary models that operate at different spatial resolutions that are interlinked and integrated into an overall assessment framework (Fig. 1). Integration is achieved through a series of hard and soft linkages between the individual components, to ensure internal scenario consistency and plausibility (Riahi et al. 2007).
The three principal models of the IA framework (Fig. 1) are MESSAGE–MACRO (Messner and Strubegger 1995; Rao and Riahi 2006), DIMA (Rokityanskiy et al. 2007) and AEZ–WFS (Fischer et al. 2007) (see below for further details). The three models are driven by a set of harmonized inputs at the regional, national, and grid (0.5 × 0.5°) level. For this purpose, the regional population and GDP scenarios of the A2r scenario (see Section 3.1) are disaggregated to the level of countries through a combination of decomposition and optimization methods. In a subsequent second step, national results are further disaggregated to the grid-cell level, which provides spatially explicit patterns of population and economic activities (Grubler et al. 2007). The latter indicators are particularly important for the spatially explicit modeling of emissions and land-cover changes in the forestry and agriculture sectors. They provide the basis for the estimation of comparable indicators (such as relative land prices or population exposures to pollutant emissions) that define e.g. the relative comparative advantages of agriculture- and forestry-based activities or the stringency of spatial pollutant emissions reductions.
The MESSAGE model (Model for Energy Supply Strategy Alternatives and their General Environmental Impact) stands at the heart of the integrated assessment framework. It is a systems-engineering optimization model used for medium-to long-term energy system planning, energy policy analysis and scenario development. The model maps the entire energy system with all its interdependencies from resource extraction, imports and exports, conversion, transport and distribution to end-use services. The model’s current version provides global and sub-regional information on the utilization of domestic resources, energy imports and exports and trade-related monetary flows, investment requirements, the types of production or conversion technologies selected (technology substitution), pollutant emissions, inter-fuel substitution processes, as well as temporal trajectories for primary, secondary, final, and useful energy. In addition to the energy system, the model includes a stylized representation of the forest and agricultural sector and related GHG emissions mitigation potentials. It is a long-term global model operating at the level of 11 world-regions and a time horizon of a century (1990–2100). For each scenario the model calculates the least cost solution for the energy system given a set of assumptions about main drivers such as energy demand, resources, technology performance and environmental constraints.Footnote 2
The AEZ–WFS (Agro-Ecological Zoning—World Food System) model framework projects alternative development paths of the agriculture sector using three components: (i) a spatially detailed agronomic module assessing crop suitability and land productivity (AEZ); (ii) an applied general equilibrium model of the world food system (WFS); and (iii) a spatial downscaling model allocating the aggregate WFS production levels and agricultural land use to spatial biophysical resources. AEZ simulates land-resource availability, crop suitability, farm-level management options, and crop production potentials as a function of climate, technology, economic productivity, and other factors (for further details see Fischer et al. 2002, 2009; Fischer 2009). Land is broadly classified as built-up land, cultivated land, forests, grass/wood land areas, including managed and natural grassland areas, and sparsely vegetated and other land. WFS is an agro-economic model (Fischer et al. 2005, 2009) that estimates regional agricultural consumption, production, trade and land use. Applying the AEZ-WFS framework, use and conversion of land is determined for food and feed production to meet the global demand in accordance with agronomic requirements, availability of land resources, and consistent with national incomes and lifestyles of consumers. Land for residential use and transport infrastructure is assigned according to spatial population distribution and density. The remaining land, i.e. part of grass/wood land, forest areas and sparsely vegetated areas, is further evaluated in the DIMA model (see below) for possible use in dedicated bioenergy systems and for forestry purposes (for additional details see Tubiello and Fischer 2007). Agricultural residue supplies based on the agricultural land use are also available for energy use and picked up where cost-effective. The delineation of pasture and unmanaged grasslands is based on the projections of livestock numbers computed in the WFS model.
The DIMA model (Dynamic Integrated Model of Forestry and Alternative Land Use; Rokityanskiy et al. 2007) is used to quantify the economic potential of global forests, explicitly modeling the interactions and feedbacks between ecosystems and land use related activities. Regional demand trajectories for timber and prices for carbon and bioenergy are major drivers for the relevant estimates. Food security is maintained by introducing an exogenous scenario-specific minimum amount of agricultural and urban land per grid cell as projected by AEZ-WFS (and used as input by DIMA). The DIMA model is a spatial model operating on a 0.5 × 0.5° grid raster. It determines for each grid and time interval, which of the forestry processes (afforestation, reforestation, deforestation, or conservation and management options) would be applied in order to meet a specific regional timber demand and how much woody bioenergy and forest sink potential would be available for a given combination of carbon and bioenergy prices. Main determinants of the land-use choices in each grid are assumptions about the costs of forest production and harvesting, land-prices and productivity, age structure of standing forest, and age-specific plant growth. Forest dynamics are thus a result of interactions between demand-pull (price of bioenergy and carbon as well as timber demand), and inertia on the supply-side (imputed through growth limitation of the forest). A schematic illustration of the main linkages between the three principal models is shown in Fig. 1.
In the sequel of Section 2 we discuss the main methodological improvements for the RCP8.5. We particularly focus on those aspects most relevant for the development of spatial land-cover and emissions projections, which serve as inputs to the climate modeling community (see also Hurtt et al. 2011 and Lamarque et al. 2011 in this SI).
Spatial land use and land-cover change projections
The spatial land cover information of the RCP8.5 builds upon the dynamic land-use projections available from the original A2r scenario as published in Fischer et al. 2007; Tubiello and Fischer 2007, and Riahi et al. 2007. The categories comprise 1) built-up land (residential plus infrastructure), 2) cultivated land (arable and permanent crops, separated by irrigated and non-irrigated land), 3) forests (separated by managed and unmanaged forests), 4) grassland/woodland/shrubland (GWS), and 5) other land (water, desert, rocks, and ice).
Major improvements of the RCP8.5 (compared to the original A2r scenario) include updates with respect to the representation of base-year land-cover statistics, updates in the AEZ resource inventory, as well as the split of the aggregated GWS category into pasture and natural grasslands. The latter was done specifically as input for the climate modeling teams of the IPCC-AR5 to represent dynamic land-cover changes in their future climate projections.
The base-year (2000) land inventory uses a continuous representation of different shares of land-uses at 5 min latitude/longitude, i.e. each 5 min grid cell is characterized by shares of the above classes.
Six geographic datasets were used for the compilation of an inventory of seven major land cover/land use categories: (1) GLC2000 land cover, regional and global classifications at 30 arc-seconds (JRC 2006); (2) IFPRI Agricultural Extent database, which is a global land cover categorization providing 17 land cover classes at 30 arc-seconds (IFPRI 2002), based on a reinterpretation of the Global Land Cover Characteristics Database (GLCCD 2001), EROS Data Centre (EDC 2000); (3) The Global Forest Resources Assessment 2000 of FAO (FAO 2001) at 30 arc-seconds resolution; (4) Digital Global Map of Irrigated Areas (GMIA) version 4.0.1 of (Siebert et al. 2007) at 5 arc-minute latitude/longitude resolution, providing by grid-cell the percentage land area equipped with irrigation infrastructure; (5) IUCN-WCMC protected areas inventory at 30-arc-seconds (http://www.unep-wcmc.org/wdpa/index.htm), and (6) Spatial population density inventory (30 arc-seconds) for year 2000 developed by FAO-SDRN, based on spatial data of LANDSCAN 2003, LandScanTM Global Population Database (http://www.ornl.gov/landscan/), with calibration to UN 2000 population figures.
An iterative calculation procedure has been implemented to estimate land cover class weights, consistent with aggregate FAO land statistics and spatial land cover patterns obtained from (the above mentioned) remotely sensed data, allowing the quantification of major land use/land cover shares in individual 5 arc-minute latitude/longitude grid-cells. The estimated class weights define for each land cover class the presence of respectively cultivated land and forest. Starting values of class weights used in the iterative procedure were obtained by cross-country regression of statistical data of cultivated and forest land against land cover class distributions obtained from GIS, aggregated to national level. The percentage of urban/built-up land in a grid-cell was estimated based on presence of respective land cover classes as well as regression equations, obtained using various sub-national statistical data, relating built-up land with number of people and population density.
When land is spatially allocated to various uses in the AEZ-WFS model sequence, first the conversion to built-up land is quantified, driven by changes in population numbers and density. Second, changes in agricultural land simulated in WFS are spatially allocated, simultaneously affecting the other land use types, except built-up land. Finally, other land use changes (not driven by agriculture or built-up conversion), mainly between forest and grass/wood land types, are accounted for. The conversion of agricultural land is allocated to the spatial grid for 10-year time steps by solving a series of multi-criteria optimization problems for each of the countries/regions of the world food system model.
The criteria used in the land conversion module depend on whether there is a decrease or increase of cultivated land in a region. In the case of a decrease the main criteria include demand for built-up land and abandonment of marginally productive agricultural land. In case of increases of cultivated land, the land conversion algorithm takes land demand from the world food system equilibrium and applies various constraints and criteria, including: (i) the total amount of land converted from and to agriculture in each region of the world food system model, (ii) the productivity, availability and current use of land resources in the country/regions of the world food system model, (iii) agronomic suitability of land for conversion to crop production, (iv) legal land use limitations, i.e. protection status, (iv) spatial suitability/propensity of ecosystems to be converted to agricultural land, and (v) land accessibility, i.e. in particular a grid-cell’s distance from existing agricultural activities.
The classification of GWS into areas that predominantly correspond to pastures vs. natural GWS is based on spatial calculations of fodder supply versus livestock feed requirements. For this purpose feed balance calculations were performed to compare estimated feed requirements of livestock in a grid-cell to estimated feed supply from grassland and cropland in each grid cell. Feed requirements were calculated as energy requirements per unit of a reference livestock times number of ruminants (cattle, buffalo, sheep, goat). Feed supply assumes a grass harvest index of 60% (on grass/wood land) and a harvest index of 30% crop residues on crop land in the grid cell. These calculations were done at 5 min latitude/longitude and aggregated later to 0.5 × 0.5° resolution of the RCP8.5. By doing so the global grass/wood land cover was classified into four different categories. For areas with no ruminants or a share of GWS <10% in a grid-cell, these grid-cells were assigned to class 1; class 2 comprises areas with a ratio of feed requirements over feed supply of less than 0.1; class 3 corresponds to calculated ratios of 0.1 to 0.5, and finally class 4 corresponds to ratios greater than 0.5. The resulting global map of grazing intensity is presented in Fig. 2.
Base year estimates and environmental legislation
For the estimation of air pollutant emissions we rely on detailed technology activity data and emissions coefficients from the Greenhouse Gas and Air Pollution Interactions and Synergies model (GAINS, Amann et al. 2008a, b) and the recent assessment of environmental legislation until 2030 (Cofala et al. 2007). The activity data including improvements of emissions coefficients due to legislation was subsequently aggregated and implemented into the MESSAGE modeling framework to derive projections for pollutant gases, including sulfur-dioxide (SO2), nitrogen oxide (NOx), carbon monoxide (CO), volatile organic compounds (VOCs), black and organic carbon aerosols (BC and OC). Details of the methodology describing the linkage between MESSAGE and GAINS are summarized in Rafaj et al. (2010).
The main sectors covered in our analysis include power plants, fossil fuel extraction, gas flaring, waste and biomass burning (deforestation, savannah burning, and vegetation fires), industry (combustion and process), domestic (residential and commercial sectors), and road transport. We separately include estimates of air pollutants from international shipping and aviation sectors, which have recently been identified as important sources of air pollutants. Projections of emissions from international ships are based on the methodology described in Eyring et al. (2005a, b) and reflect the implementation of recent updates of IMO standards (amendments to the MARPOL Annex VI regulations). Lee et al. (2005) is used to derive estimates of aviation fuel consumption and controls.
The main control policies and strategies for air pollutants until 2030 across different sectors in both OECD & Non-OECD regions are detailed in Table 1.Footnote 3
For the medium to long term trends of RCP8.5 (beyond 2030) we assume a further reduction in emissions intensity based on the assumption that higher environmental quality will be associated with increasing welfare. To mimic this behavior, the Environmental Kuznets Curve (EKC) theory is applied to derive changes in future emission coefficients (see e.g. Dasgupta et al. 2001). Based on empirical observations, the EKC assumes first an increase in emissions (with increasing economic activities) followed by a decrease. Many EKC studies assume an income level between 5000 and 8000 $/cap as the turning point for the introduction of stringent environmental controls. Recent evidence, however, suggests that in many developing countries controls of air quality are introduced at faster rates than suggested by the experience of industrialized countries in the past (see Dasgupta et al. 2001; Smith et al. 2005). Increased environmental awareness and accelerated technological diffusion are major contributors to this trend. The turning point of the EKC are likely to happen thus at lower GDP/capita levels than assumed earlier. Consequently, we use in the RCP8.5 analysis an income level of 5000$/capita as the threshold for increasing environmental consciousness triggering declines in emissions intensities.Footnote 4 For resulting development of emissions intensities and overall emissions trends see Section 3 on “results”.
As a final step in the development of the regional projections of the RCP, the MESSAGE model results for all major air pollutant emissions and reactive GHGs were harmonized with the historical and current inventories as described in Granier et al. (2011). A simple harmonization algorithm was assumed, where emissions growth of the native MESSAGE results were combined with the base-year values from Granier et al. (2011). For some sectors, where the algorithm led to qualitative changes in the overall trends, a declining offset over time was employed for the harmonization.
Downscaling of pollutant emissions
In addition to detailed representation of air-pollution legislation, another important improvement of the RCP8.5 comprises the development of new downscaling algorithms for the spatially explicit projections of pollutant emissions. These spatial air pollutant projections are important inputs to the AR5 climate experiments, and related atmospheric chemistry models (Lamarque et al. 2011).
The vast majority of downscaling approaches have traditionally employed proportional downscaling (van Vuuren et al. 2010), where emissions of individual grid-cells are scaled following aggregate changes at the regional level. While proportional algorithms are simple to implement and easy to reproduce, they generally do not account for important local differences in efforts to reduce pollutant emissions. Empirical evidence, for example, shows that efforts to reduce air-pollution have generally been stronger where the returns in terms of health benefits have been the largest. In the past this has been particularly the case in cities of today’s industrialized countries, where dedicated urban air pollution legislation has successfully reduced exposure and thus health impacts for millions of people (WEA 2000).
This trend is likely to continue in the future, particularly in the developing world, where urban air quality is one of the prime concerns. We thus employ an exposure-driven spatial algorithm for the downscaling of the regional air-pollutant emissions projection. By doing so, we generate dynamic spatial maps at the resolution of 0.5 × 0.5° for all world regions and major pollutant emissions (SO2, NOx, CO, BC, OC, VOCs). As a surrogate proxy for the spatial distribution of exposure we compute “population x emissions” of each grid-cell. The weight of each individual cell in the aggregate regional exposure (i.e., the numerical sum of all exposure values of the cells in the region) defines the allocation of emissions reductions for each cell. As a result emissions are reduced most in those cells with the highest exposure. Vice versa, in cells with either very low population or low emissions density the reductions are comparatively smaller. Technically, we solve the problem by creating a rank-size distribution of each region from the cells with the highest exposure to those with lowest. We start reducing emissions first in those cells that have the highest exposure.Footnote 5 Following a review of Air Quality Monitoring Information of US cities (EPA 2008; see also UNEP and WHO 1996) we adopt a maximum rate of reduction of up to 80% emissions reduction per decade for each grid-cell.Footnote 6
Obviously, the exposure driven algorithm is applied only if emissions are reduced on the regional level due to increasing stringency of air pollution legislation. In the case of regionally increasing emissions, we use spatial changes of economic activity (GDP) as a proxy to allocate increasing emissions across grid-cells. I.e., we assume that emissions increase proportionally to where economic activity is accelerating the strongest. For the spatial distribution of population and GDP we rely on the downscaled projections of the original scenario (A2r) as described in Grubler et al. 2007 (data can be downloaded at http://www.iiasa.ac.at/web-apps/ggi/GgiDb/).
Figure 3 gives a schematic illustration of the effect of the exposure algorithm for SO2 emissions in the Centrally Planned Asia region (including China) between 2020 and 2100. The two important features are: 1) that top exposed cells corresponding to the Chinese mega-cities improve air quality by about two orders of magnitudes by 2050, and 2) improvements in cities are complemented by important distributional changes, shifting e.g. emissions intensive activities to surrounding neighborhoods of cells with lower population density. For a comparison see also resulting spatial maps of SO2 emissions in Fig. 11 (Section 3).
Scenarios considered in this paper
The main scenario described in this paper is the RCP8.5. As indicated in the introduction, however, we also use the MESSAGE model for the development of mitigation scenarios that use the RCP8.5 as a baseline. As targets for the mitigation scenarios we adopt forcing levels of 2.6, 4.5 and 6 W/m2 by the end of the century, which corresponds to the same radiative forcing levels as assumed by the other RCPs in this SI (see van Vuuren et al. 2011b; Thomson et al. 2011; Masui et al. 2011). For each mitigation scenario the MESSAGE optimization model computes least-cost pathways to stay below the specified target. This corresponds to the introduction of a cumulative GHG emissions budget and a globally uniform price vector for greenhouse gas emissions (assuming full temporal and spatial flexibility in emission reductions across regions and gases).