We present a modelling framework to simulate probabilistic futures
of global cropland areas that are conditional on the SSP (shared
socio-economic pathway) scenarios. Simulations are based on the Parsimonious
Land Use Model (PLUM) linked with the global dynamic vegetation model
LPJ-GUESS (Lund–Potsdam–Jena General Ecosystem Simulator) using
socio-economic data from the SSPs and climate data from the RCPs
(representative concentration pathways). The simulated range of global
cropland is 893–2380 Mha in 2100 (
Land use and land cover change (LULCC) is a fundamental aspect of global environmental change, but large uncertainties exist in estimating the effect of multiple drivers on LULCC in the future (Brown et al., 2014). A range of different models and scenarios have been used to project future cropland areas to 2100 with estimates in the range of 930 to 2670 Mha (Alexander et al., 2016; Prestele et al., 2016). This compares with today's cropland areas of around 1530 Mha. The large differences in these projections reflect uncertainties in process understanding, the use of different models to represent these processes and the direction of development of multiple drivers, including food demand and agricultural productivity (Schmitz et al., 2014; Smith et al., 2010). The direction of socio-economic drivers is referred to as deep uncertainties, which are addressed through the use of scenarios (van Vuuren et al., 2008). Cropland projections at the high end of the projected uncertainty range for global cropland would have profound consequences for, for example, global carbon and nitrogen fluxes, the global water balance, biodiversity, and other ecosystem services (Lindeskog et al., 2013; Pereira et al., 2012; Zaehle et al., 2007). Hence, quantifying and understanding the inherent uncertainties in the drivers of LULCC has important consequences for policy responses to support sustainable development. However, the effects of uncertainties in the underlying scenario assumptions have not been systematically quantified for global cropland projections.
Scenarios are characterized by storylines that describe assumptions about key drivers and processes from which model input parameters are interpreted (Rounsevell and Metzger, 2010). These parameter interpretations are by definition deterministic within a scenario context, since they do not consider the uncertainties associated with the interpretation process itself. By contrast, probabilistic approaches examine system uncertainties by assigning probability distributions to input variables (reflecting uncertainties about scenario assumptions) to assess the influence of uncertainty on system outputs (van Vuuren et al., 2008). The conditional probabilistic approach combines the strength of scenarios in addressing deep uncertainties with the probabilistic approach that explores the uncertainties in the assumptions about model input parameters (O'Neill, 2005; van Vuuren et al., 2008). In this case, the probability distribution of each model input parameter is conditional on the internal logic and assumptions within the contextualizing scenario. Hence, conditional probabilistic futures are useful in exploring parameter uncertainty within and across scenarios (Brown et al., 2014; van Vuuren et al., 2008).
In this paper, we present probabilistic futures of global cropland that are
conditional on scenario assumptions. In doing so, we quantify the
uncertainties within these assumptions, as well as representing the deep
uncertainty across different scenarios. The assessment is based on the
following key questions:
How will cropland area evolve until 2100 in
response to socio-economic drivers and climate change? Will future ranges of
global cropland for different scenarios overlap due to differences in
socio-economic conditions and/or due to uncertainties in model input
parameters? How does the influence of the uncertainties in model input
parameters change through time?
We use a scenario framework based on the five shared socio-economic pathways (SSPs) and develop a scenario matrix combining the five SSPs and four representative concentration pathways (RCPs). This scenario matrix is filled with probabilities based on the assumption that a given SSP will correspond to a given RCP. For each SSP, we derive RCP-specific input (yields in this case) applying the scenario matrix. The resulting conditional probabilistic futures are named F1–F5, where the numbers 1–5 correspond to SSP1–5. The RCP–SSP scenario framework was used since it is the most recent scenario approach for global environmental change research (Ebi et al., 2014; O'Neill et al., 2016; van Vuuren et al., 2011, 2014). We apply these scenarios to a global-scale, socio-economic model of agricultural land use change (the Parsimonious Land Use Model, PLUM; Engström et al., 2016) in combination with crop yield time series derived from the dynamic global vegetation model, LPJ-GUESS (Lund–Potsdam–Jena General Ecosystem Simulator; Lindeskog et al., 2013; Smith et al., 2001). PLUM has been benchmarked against different models and scenario studies in a land use model intercomparison exercise (Alexander et al., 2016; Prestele et al., 2016) that has demonstrated its consistency in comparison with other global cropland simulations. Because of its rapid runtimes, PLUM can explore uncertainties across its input parameter space (Engström et al., 2016) and hence is appropriate for use in probabilistic simulations requiring multiple model iterations.
To construct the conditional probabilistic futures (F1–F5) we used qualitative and quantitative information from the SSPs directly and quantitative information from the RCPs indirectly as input to PLUM (Fig. 1).
The SSPs describe plausible, alternative societal development pathways over the 21st century in the absence of climate change or climate policies (O'Neill et al., 2016, 2013). The SSP-specific development of society and sectors, such as energy and land use, results in varying challenges for mitigation and adaptation to climate change (O'Neill et al., 2016). In SSP1, economic growth and technological development are strong, sustainable solutions are preferred and population growth is low, resulting in small challenges for mitigation and adaptation. By contrast, SSP3 presents great challenges for mitigation and adaptation because it is characterized by high population growth but low economic and technological growth, combined with resource-intensive lifestyles. SSP2 describes a world with medium population growth and technological and economic development, resulting in medium challenges for mitigation and adaptation. The remaining two scenarios (SSP5 and SSP4) present contrasting challenges for mitigation and adaptation. SSP5 is a fossil-fuel-based world that is focused on development (low population growth, high economic and technological growth) and presents a great challenge for mitigation, but a small challenge for adaptation. SSP4 presents a small challenge for adaptation but a great challenge for mitigation because inequality is high across and within countries, and various levels of economic and technological development benefit the global elite.
The SSP–RCP modelling framework. The RCPs and SSPs, as well as the scenario matrix (author judgement about the distribution of RCPs conditional on SSPs), are input to the model (indicated in blue). Models (indicated in orange; GCMs: general circulation models) use input or results of other models (intermediate results, indicated in green). The final outputs of the modelling framework are the cropland futures F1–F5 (indicated in red).
In the modelling framework presented here, the different socio-economic
pathways were combined with different levels of climate change associated
with the four RCPs. The RCPs are defined by their forcing targets from 2.6
to 8.5 W m
The SSPs and RCPs were combined within a scenario matrix to reflect
assumptions about the plausibility of an RCP arising from an SSP.
Theoretically, all cells within this matrix are possible, but not all
combinations of SSPs and RCPs are equally plausible and consistent (van
Vuuren and Carter, 2014; van Vuuren et al., 2014). For instance, van Vuuren
et al. (2012) indicate that emissions are only likely to be as high as
assumed under RCP8.5 with large-scale use of fossil fuels, driven either by
rapid economic growth (and little substitution towards less carbon-intensive
fuels) or very high population growth. By contrast, for the matrix cells at
the lower end of the RCP range (2.6 and 4.5 W m
The conditional probabilistic approach was implemented through the following
steps (van Vuuren et al., 2008):
identification of uncertain parameters; estimation of the conditional probability ranges associated with these
parameters, i.e. their probability density functions (PDFs); use of Monte Carlo sampling across the PDFs to undertake multiple
simulations; identification of the uncertainty ranges in model outcomes and the
determinants of model uncertainty. uncertainties in the PLUM socio-economic input parameters and uncertainties in climatic variation and sampling from the scenario matrix
for the crop yield simulations.
Thus, the uncertainty ranges of the global model output variables arise
from
The effect of a. was explored in combination with b. for global output
variables. We hypothesize that the effect of b. is smaller than the effect
of a. on the range of global cropland change. This hypothesis was tested
by undertaking model runs in which only the socio-economic parameter
uncertainties were explored. This stepwise approach is described in more
detail below (Sect. 2.3: steps 1–3 for a.; Sect. 2.4: steps 1–3 for b; Sect. 2.5: step 4).
PLUM simulates agricultural land use in terms of cropland for 160 The availability of input data determines the number of countries included.
In the evaluation version in Engström et al. (2016) 162 countries were
included, whereas 160 are used here.
To derive LPJ-GUESS country-level projections of actual and potential cereal
yield, LPJ-GUESS simulations were performed on a 0.5
Decreasing yield gap for Ukraine. The scenario parameters related to technological change determine how rapidly the yield gap decreases over time (the arrow only being symbolic, indicating the drivers of changing yield gap).
Quantitative uncertainty levels
(uncert.) for
PLUM input parameter groups and the influence of SSP scenario elements (from O'Neill et al., 2016). In PLUM several of the input parameters were grouped together conceptually, as indicated by the parameter group number (no.). For example, there are four input parameters that describe meat consumption trajectories of different income and cultural groups (meat1–meat4, see Table 1), which all belong to meat consumption, parameter group no. 3.
The country-level actual yields calculated in the previous step might differ
slightly from the country statistics from FAOSTAT on which PLUM is based.
Thus, as a final step, the yield calculated in PLUM was scaled to match
yields reported in the year 2000 (FAOSTAT, 2015). The scaling factor for the
year 2000 was applied throughout the simulation period. An example of the
yield calculations is given in Fig. 2, for Ukraine. This example uses the
socio-economic assumptions from SSP5 and the yield projections driven by
RCP6.0. Carbon fertilization has a strong effect on crop yields; e.g. for
Ukraine potential yields are simulated to increase from below 6 to above 8 t ha
The computational costs of PLUM are relatively low since it is a simple model that operates on the country scale, with global parameterization and a focus on aggregated global outputs. This allows a wide range of socio-economic input parameters to be tested. Thus, in addition to input parameters that affect cropland changes directly, we also analysed input parameters that affect other global output variables such as meat consumption and cereal demand.
The conditional probability ranges describe the uncertainty ( Throughout the paper, parameter names are
given in italics.
For each SSP, the mean of the parameters
The applied population and GDP projections are SSP and country specific but
retain uncertainty with respect to the interpretation of the underlying drivers,
model structures and country groupings. These uncertainties were explored
with the uncertainty levels of the global parameters
For the mean values of the other global model input parameters, we started
with the historic mean value of each parameter (Engström et al., 2016)
and assessed a baseline trend qualitatively (Table 1). The positive or
negative strength of the qualitative baseline trends were characterized with
symbols (
Low, medium or high uncertainty levels were attributed to each input
parameter and scenario. These uncertainty levels comprise several sources of
uncertainty: the understanding of the world characterized by a storyline,
the knowledge about the global average development of a driver, and the
heterogeneity and variability of the model parameter across and/or within
countries. The change in trend and uncertainty level (see Table 1) was interpreted for each model input parameter conditional on each SSP using the
scenario elements in Table 2. For example, we assumed that the scenario
elements of “technology development and transfer” and “agriculture”
(Table 2) influence the input parameters of yield development (for both
cereals and animal products:
For SSP5, technology development and transfer are described as being rapid
and agriculture is highly managed and resource-intensive with a rapid
increase in productivity (O'Neill et al., 2016). We interpreted this as
strong improvements in feed conversion ratios (5:
Some PLUM input parameters are not global but based on country groups to
reflect variability in local contexts, such as meat consumption (Table 1,
no. 3:
These qualitative estimates of changes in trend and uncertainty levels for
the PLUM input parameters in Table 1 were translated into quantitative
values (mean and standard deviation characterizing the PDF; see Sect. 2.3.3)
by sampling from an input parameter value matrix (input parameters in rows,
symbols
To assess the uncertainty in projected model output, Monte Carlo sampling
was used to create different sets of PLUM input parameters from PDFs
conditional on each SSP. We assumed a normal distribution for most PLUM
input parameters since it seems unlikely that extreme values would occur
frequently. Moreover, extreme values would be applied to all countries
simultaneously due to the nature of the global parameterization (i.e. in one
model run all countries have the same value). The choice of normal
distributions was also supported by the normal distribution seen in the
inter-country variability in historic data for global parameters of, e.g., meat and milk consumption. The land conversion parameters (nos. 7–9, Table 1)
are an exception, as their values only limit the internally calculated land
conversion rates. Each maximum value was thought to be equally plausible and
so we assumed the land conversion rates to be uniformly distributed. The
PDFs were constructed by using the mean and standard deviation (minimum and
maximum for land conversion parameters) derived for each SSP (Table B1, Appendix B).
All PDFs were truncated by
The RCPs cover a wide range of emission and concentration scenarios: at the
low end with the mitigation pathway RCP2.6 and at the upper end with the
high-emission pathway RCP8.5 (van Vuuren et al., 2011). For a given RCP,
modelled global average temperatures between different GCMs can vary by up
to 1
A second source of uncertainty in future crop yield projections is that each SSP could, though with different probabilities, lead to different RCPs. To address this uncertainty, the likelihood of SSP–RCP combinations was estimated (in the absence of mitigation strategies) as described below.
Conditional probabilities (ranging from very low to very high) of SSPs resulting in RCPs based on authors' judgement.
The SSP–RCP probability judgements were combined with the interpretation of the SSP storylines (O'Neill et al., 2016) to estimate the conditional probabilities (van Vuuren and Carter, 2014) given in Table 3. The sustainability assumptions in the SSP1 scenario with respect to environmental and energy policies could curb emissions sufficiently to achieve RCP2.6, but it is more plausible for the SSP1 scenario to arrive at greenhouse gas concentrations consistent with RCP4.5 and RCP6.0 (medium probability).
Many medium reference scenarios result in forcing levels around 6–7 W m
The very high-emissions pathway of RCP8.5 can only be achieved with a combination of, for example, high economic growth and reliance on fossil fuels. The divergent development in SSP4 for the few elite and the many fewer privileged people is difficult to estimate. We assumed that SSP4 has a high probability of resulting in forcing around RCP6.0 or possibly lower. The latter is based on the moderate population growth and the original positioning of the scenario (small mitigation challenge). The majority of the population in SSP4 cannot afford a material-intensive lifestyle, making RCP8.5 forcing unlikely. For SSP5 we assumed that the material-intensive lifestyle combined with very high economic growth would lead to RCP8.5 with a high probability (comparable to the assumptions for the SRES (Special Report on Emissions Scenarios) A1F1 scenario; see van Vuuren and Carter, 2014).
The qualitative probabilities in Table 3 were translated into quantitative values in Table 4. We assumed that the qualitative notions of very high, high, medium, low, and very low probability translated into quantitative probabilities of 0.9, 0.75, 0.5, 0.25 and 0.1 respectively. The assigned probabilities were normalized so that the sum of probabilities for each SSP equalled 1 (see Table 4).
Scenario matrix translated to quantitative probabilities.
The probabilities of SSPs resulting in RCPs (Table 4) were combined with the
uncertainties arising from the climate variability of the different GCMs. To
do so, the aggregated country-level yield time series (described in Sect. 2.2) were calculated for each RCP–GCM combination. Yield time series
calculated for different countries vary, depending on the underlying GCM. To
account for this spatial variability, the deviations from the yields
averaged per RCP (
Cropland distributions with and without climate variability were analysed for
the 7200 and 3600 runs respectively to the year 2100 with respect to
convergence and or divergence across the five scenarios. We report the mean
development and ranges at 95 % confidence levels (corresponding to 2
standard deviations) for the global-scale, model outputs, i.e. population,
GDP, cereal consumption, meat consumption, feed demand, cereal demand, cereal
production, cereal yield and cropland. The word “likely”, based on the
IPCC's recommendation for uncertainty communication (Mastrandrea et al.,
2010), was used to report results that are probable at 68–100 %. A
global sensitivity analysis (GSA; Saltelli et al., 2008) was carried out to
quantify the contribution of the input parameters to the uncertainty of the
global cropland extent for all socio-economic model input parameters. The GSA
was implemented as previously described in Engström et al. (2016), with
Global cropland area increases initially for all scenarios but declines in
the simulations for F1 after 2015 (Fig. 3a). F1 continues to decline
to 963
Global population, GDP, cereal consumption, meat consumption, cereal
feed, cereal demand, mean cereal yield and cereal production for the five
scenarios F1–F5. Solid lines indicate the scenario mean; dashed lines
indicate the range based on the mean
The cropland distribution for F1 is skewed toward the higher end, which is due to the truncated distribution of uncertainties in the population projections. For the same reason, the distribution of F3 is slightly skewed toward the lower end. The cropland distribution in F3 is also peaked, indicating that the confidence in the model outcomes for F1 and F3 are the highest, despite the fact that these two scenarios show divergent global cropland development.
The variability in yields arising from the five GCMs and sampling from the SSP–RCP matrix does little to change the shape of the global cropland PDFs (Fig. 3b, comparing solid lines (with yield variation) to dashed lines (mean yield)). For cropland futures with flat distributions (F2, F4 and F5), the distributions with climate variability (solid lines) are slightly less peaky than without climate variability (dashed lines). This indicates that the climate variability contributes more to the overall uncertainty of global cropland areas for scenarios with larger overlap of global cropland outcomes (F2, F4 and F5), compared to the cropland futures F1 and F3. Overall, the effect of climate change variability and sampling from the SSP–RCP matrix is very small. However, the inter-annual variability of yields due to variations in climate patterns is considerable (not displayed here).
The strongly overlapping cropland ranges for the F2, F4 and also F5 scenarios are caused by the assumed uncertainties in the trends of the different input parameters but also by the counteracting effect of different scenario drivers leading to similar cropland futures. Conversely, the distinct development of cropland for the F1 and F3 scenarios is mainly due to the reinforcing dynamics of drivers as described below.
In contrast to the other population scenarios, the total population size in
F3 does not peak in the 21st century but grows continuously to 12.1
For F5, despite a decline in total population size to 7.4
Likewise, strong economic growth and technological change result in high
global average cereal yields in 2100 for F1 and F5, 5.4
The uncertainty in input parameters contributes differently to the uncertainty of global cropland futures over time (Fig. 5). For F1, population projections and technological change dominate uncertainty, the latter being especially important during the first quarter of the simulation period. Similarly, for F2, uncertainties in technological change and consumption are at first important, but after 2025 cropland degradation contributes largely to the uncertainty of global cropland. By contrast, population projections and technological change are the major contributors to the uncertainty range of global cropland for F3. For F4, uncertainties in the extent of land degradation, but also population projections and consumption and technological change, contribute to uncertainties in global cropland. Consumption and technological change become less important over the 21st century, compared with land degradation and population. These trends are similar for F5.
Total importance of global cropland for the futures F1–F5 to uncertainty of input parameters, aggregated to input parameter groups as in Table 2 (for non-aggregated results, see Fig. C1 in Appendix C).
For F2, F4 and F5, the uncertainty distributions of global cropland overlap
greatly, with cropland changes over the 21st century within the range of
No climate change mitigation actions were assumed in this study, although for SSP1 this would be plausible and consistent with the storyline. The simulated decrease in cropland for F1 suggests that land-based mitigation options, such as bioenergy production, could be implemented on abandoned cropland without compromising food security or the provision of other ecosystem services. However, the global sensitivity analysis showed that for F1 to consistently achieve strong decreases in cropland areas, it is important to stay within the range of input assumptions. Among others, consumption patterns have to reflect the more resource efficient and environmentally friendly lifestyle that underlies this scenario. Achieving technological change and thus yield increase is important, as is decreased environmental degradation and thus decreased cropland degradation rates.
The LPJ-GUESS yield projections are at the higher end of the range of yield
projections compared with other models (Rosenzweig et al., 2013) and likely
overestimate the effect of CO
The sensitivity analysis showed that assumptions about cropland degradation were important for cropland development across all scenarios. Cropland degradation was assumed to lead to an average global production loss of between 6 % (F1) and 14 % (F5) in 2100. This compares with an estimated global average of 20–40 % loss of potential production on degraded agricultural areas only (Zika and Erb, 2009). Hence, the PLUM results of total global production (not only on degraded agricultural areas) appear to be of the right magnitude and the sensitivity analysis highlighted the importance of accounting for these uncertainties.
Global cropland was less sensitive to the uncertainties associated with the consumption input parameters, which, for example, describe the rate of increase or decrease in meat consumption for the four consumption country groups. PLUM represents cultural differences in consumption patterns between countries (based on four consumption groups), but this could potentially mask part of the total importance of the consumption input parameters because the correlation between the parameters of the four consumption groups was not considered. Additionally, cropland changes are likely to be underestimated in F5 because meat consumption increases strongly in countries currently defined as developing and global average meat consumption approaches 110 kg per person in 2100. This would probably be associated with the intensification of animal production, which currently is not included in PLUM. Since intensive meat production would lead to an increase in the feed share derived from cereals, cropland areas would increase.
The use of a global model with reduced complexity risks missing potentially important dynamics and feedbacks, which could affect the magnitude of change (e.g. intensification in the livestock sector, as highlighted above). A reduced complexity model could also widen or limit the uncertainty range in outputs (depending on the balance between introduced uncertainty and better overall model performance). A further limitation of this approach are the judgements of the uncertainties in global model input parameters and their assumed distributions. Assessing these uncertainties is challenging because of the high degree of variability in development across 160 countries, in particular for SSP4 with large inequalities within and across nations. Furthermore, the use of normal distributions in the sampling of the input parameters might result in an underrepresentation of extreme outcomes. Thus, in the absence of better knowledge a relatively conservative approach was adopted here based on transparency in the assumed parameter ranges and distributions. Overall, the conditional probabilistic approach applied in PLUM led to cropland area ranges that are consistent with those reported by other scenarios and model intercomparison studies (Alexander et al., 2016; Hurtt et al., 2011; Prestele et al., 2016; Schmitz et al., 2014), which provides confidence in the modelling framework. PLUM is based on cereal demand and assumes that changes in cereal land are a reliable proxy for food demand and cropland changes with free-trade contributing greatly to meeting demands. For example, a change in the demand of cereals compared to other crops driven by climate change (either directly, or by enhanced demand for bioenergy) will require a revision to the constant cereal–cropland ratio. Future model development will take bioenergy production into consideration. The global-scale projections with PLUM need to be interpreted under the assumption that the future agricultural system will not be fundamentally different from how we understand it today; an assumption that occurs in most global models. Clearly, in some scenarios the free-trade simplification might not be valid (e.g. in SSP3), a limitation that is balanced by PLUM having simple and transparent relationships. We argue that the possibility to perform rapid model runs outweigh drawbacks in the current model version that arise from less than perfect regional model performance.
High-end climate change impacts on yields (i.e. from solely applying RCP8.5) were not tested here, as the goal of this study was to create plausible and consistent cropland futures that address the uncertainties within each scenario rather than assessing the impact of each emission pathway. Excluding high-end climate change impacts on yields explains why the variability in climate change was found to have a relatively small impact on global cropland areas. Small differences in the climate change impacts on agricultural areas between RCP4.5 and RCP6.0 were found elsewhere (Wiebe et al., 2015), as well as the comparatively larger effects of RCP8.5. The approach used here, based on a matrix populated with probabilities, streamlined the total number of scenarios and simultaneously removed the need to compromise with single selections of SSP–RCP combinations.
Considering the simple supply and demand mechanism in the model (the use of cereals as a proxy for demand and area changes), the likely range of global cropland simulated in this study ranged from 893 to 2380 Mha in 2100. This was consistent with the range reported in the literature of 930–2670 Mha in 2100, although slightly skewed to the lower end of this range. This shows that uncertainties in input assumptions are equally important for output ranges as differences in the model structure and that the entire uncertainty of global cropland development is probably even larger if these sources of uncertainties are combined. Considering the uncertainties in input assumptions, we found that the deep uncertainties reflected in assumptions for the socio-economic scenarios contributed most to the total magnitude of the projected cropland range. The uncertainties in scenario interpretation widened the total projected future cropland range and led to overlap in the simulated cropland areas for three out of five scenarios. Cropland futures where the output PDFs did not overlap with other scenarios were found for the SSP1 scenario projections and the SSP3 scenarios, whereas the SSP2, SSP4 and SSP5 scenarios were found to have large areas of overlap. This was partly due to the compensating dynamics of drivers, e.g. strong yield development and increase in consumption in the SSP5 scenario, but also due to the larger uncertainties in scenarios with contrasting challenges for mitigation and adaptation (i.e. SSP5 and SSP4 scenarios). Uncertainties in population projections, technological change and cropland degradation were found to be the most important for uncertainty in global cropland projections, while uncertainties in consumption levels and production levels were found to be less important. When taking account of the uncertainty ranges at the 95 % confidence interval across all scenarios, there were fewer differences between the scenarios, i.e. there is overlap at some level of probability in all global cropland projections, except for projections based on SSP3. This leads us to conclude that very different worlds can result in very similar cropland futures on the aggregated global scale as long as they share low to medium population development.
Simulation results presented in Figs. 3 and 4 are available for download from
In comparison to the version described in Engström et al. (2016),
several minor alterations (Table A1) and one larger alteration were made to
the model. The larger alteration relates to the representation of the yield
development in PLUM, which is explained below. Assumptions are also made
within PLUM about the scenario dependency of the availability of potential
arable land (
The global parameter
Cereal yield cerealYieldC is calculated in the following way:
cerealYield
PLUM development, affected variable, rationale for development and implemented development.
Global cereal surplus – exporting countries are assumed to decrease their production by the minimum
of either their domestic surplus or their share of the global surplus
(including demand created by
Global cereal surplus, importing countries – no change is assumed; only if cereal self-sufficiency should be increased, are countries assumed to increase production by their domestic deficit.
Global cereal
deficit – exporting countries are assumed to increase their production by
their share of the global deficit (including demand created by
Global cereal
deficit – importing countries are assumed to increase their production by the
maximum of either their domestic deficit or the share of the global deficit
(including demand created by
Yield-related variables in PLUM.
Facilitating governance and institutional structures, cooperative countries, environmentally conscious societies and decreased inequalities contribute in this SSP to the progress towards a sustainable world, including lower population growth. Due to effective international institutions and good information flow between markets, governments and farmers and functioning global markets, agricultural areas are decreased rapidly in case of food overproduction. Countries rely on regional trade, and attempts are made to keep food stocks low in order to be resource efficient. The conversion of natural land to new cropland is well regulated in most countries to avoid substantial deforestation and biodiversity loss. Investments in agriculture and agricultural research stay high in high-income countries, and local, context-dependent agricultural best-management practices (including non-conventional practices, e.g., no tillage) are implemented in most countries. The investment in technology continues to result in more efficient animal protein production. An additional important factor for globally increasing yields is the technology transfer between countries and income levels. Equity and education are important in this scenario and contribute to yield improvements as well. The awareness for resource efficiency also decreases food waste and the consumption of refined products, which leads to a decreasing cereal consumption. The environmental awareness of consumers leads to a slowing down and an eventual decrease in the consumption of dairy and meat products. However, low-income countries moderately increase their animal product consumption until they reach consumption levels that are common among western countries. Environmental degradation slows down and the status of land improves, thanks to increasing implementation of holistic and sustainable management and afforestation programs.
In SSP2 trends observed during recent decades continue, including some reductions in resource intensity, but mostly large inequalities between countries and economies remain. Technological development is moderate and preliminary, concentrated in high-income countries. Due to limited technology transfer, low-income countries do not benefit from advances in agricultural management and yields remain rather low. Agricultural markets are partially functioning and globally connected, but trade barriers are only reduced slowly. Some countries with limited access to global markets focus more strongly on increased domestic production and self-sufficiency. In general food stocks are held at moderate levels and the abandonment of cereal land remains unregulated. For new cropland generation, high-income countries follow existing regulations, while in some low-income countries with rich natural resources unregulated deforestation for cropland generation continuous to be a problem. Environmental degradation continues at historical rates, as no serious efforts are made to achieve large-scale sustainable land management. Additionally the continuing increasing demand for animal products contributes to expansion and intensification of agriculture with some negative environmental impacts.
In the fragmented world, regional blocks form, with little international cooperation and protectionist policies of regions as a result. This leads to little reduction of land intensity, low technological development and generally slow economic growth but high population growth. However, in some areas wealth moderately increases, and so does technological development. The increasing efforts of regions to be more food self-sufficient reduce agricultural trade and increase the food overproduction within regions to ensure sufficient food supply in case of regional harvest shortcomings. Consequently, agricultural area is only abandoned at a very slow pace, even if a region is food sufficient. At the same time, weak governance and institutional structures do not provide any strong regulations reducing the conversion of natural land to cropland. Forests and natural grasslands are converted into cropland at faster rates to ensure regional food security. Food consumption, and in particular the consumption of animal products, continues to increase in most regions but at a slower pace for low-income countries. The increased demand for food and the non-regulated land use change result in serious environmental degradation.
This world is characterized by high inequality, within and across countries, as well as between economies. In all countries, including low-income countries, few very privileged people steer all political, economic and industrial activities. This includes agriculture, which is strongly divided into highly industrialized large-scale monoculture agriculture steered by the privileged and small-scale farming performed by a large group of poor people. Investments in agricultural development of the industrial agriculture are large, but no technology transfer occurs to the small-scale farming, and here yields remain low. If the industrialized agricultural production is not profitable, cropland is abandoned at fast rates, while at the same time natural land is converted at fast rates to new cropland without considering environmental and social effects. The absence of sustainability regulations leads to serious environmental degradation, affecting the poor and making them even more vulnerable. The global food trade is dominated by the industrial agricultural businesses with very limited access for small-scale farmers. Small-scale farmers therefore rely more on self-sufficient agricultural systems. While the privileged society increases its consumption of animal products, the large group of poor people cannot afford large increases in meat and dairy consumption. The overall demand for food production does therefore not proportionally increase with the high population growth, as most of the world's people cannot afford an expensive diet in times of economic uncertainty.
In this world economic, resource-intensive development is prioritized, and while this leads to eradication of extreme poverty, it comes at environmental costs. Developing countries are pushed in their development, and soon all countries share a resource-intensive lifestyle, including high levels of animal product consumption. The high demand for these and other agricultural products is fulfilled by highly engineered agricultural systems. Investment into agricultural technology is very high. Increasing agricultural specialization of countries is common too; however, it is often connected to very resource-intensive production, both in terms of water and fertilizers. Agricultural area also expands into natural areas at faster rates if necessary. Solutions to environmental problems do not tackle the problem's roots, but only its symptoms. However, the global food market functions well and keeps the total food stock decrease slow.
An example of the importance of being explicit about baseline trends is the
following: in SSP5 the very strong trend of technology improving agricultural
management (Table 1; SSP5;
For each scenario and each input parameter, quantitative values were derived by sampling from Table B1 based on the scenario and input parameter's qualitative notions in Table 1.
This matrix (Table B1) was populated by first placing the (baseline) mean
value (Engström et al., 2016) based on the quantitatively estimated
baseline trend within the matrix. Secondly, we identified minimum and maximum
values for each input parameter, based on statistical analysis of historic
data or the authors' judgement as described in Engström et al. (2016).
Thirdly, these values informed the extremes (
The values for a change in trends (mean) and uncertainty value (standard
deviation) were used to create the probability distribution function for each
input parameter. We assumed a normal distribution for all input parameters,
except parameters 7–9. These parameters are maximum values, and it seems
more plausible that they are equally likely (uniform distribution). For
population, we truncated the distribution because the very low population
projections of SSP1 (peak of global population size in 2050–2055 at
8.5 billion, a decline in total population size to 6.9 billion in 2100)
requires a very stringent decline in fertility rates, and even for SSP1 with
a high focus on education (the most important driver for changes in fertility
rates), it seems very unlikely to us that the projections will be lower by
Matrix with quantitative values for changes
in trend from
Total importance of cropland to uncertainty of input parameters conditional on SSPs.
Kerstin Engström, Mark D. A. Rounsevell, Almut Arneth, Stefan Olin and Dave Murray-Rust designed the study. Kerstin Engström and Sara Brogaard assessed the uncertainties for global PLUM parameters. Detlef P. van Vuuren, Kerstin Engström, Mark D. A. Rounsevell and Peter Alexander estimated the qualitative and quantitative probabilities of the scenario matrix. Kerstin Engström and Stefan Olin developed the model code and performed the simulations. Kerstin Engström prepared the paper with contributions from all co-authors.
This study was carried out under the Formas Strong Research Environment grant to Almut Arneth, Land use today and tomorrow (LUsTT; dnr: 211-2009-1682). The authors would also like to thank J. Lindström for discussion of statistical methods for the yield sampling. Mark D. A. Rounsevell, Detlef P. van Vuuren and Almut Arneth acknowledge support by the FP7 project LUC4C (grant no. 603542). Edited by: J. Dyke Reviewed by: two anonymous referees