Background
The most comprehensive synthesis of climate change knowledge is carried out by the Intergovernmental Panel on Climate Change (IPCC). Its activity synthesised in public reports, is based on an extensive review of literature and provides governments with scientific evidence about the state, impacts and risks of a changing climate, as well as about potential mitigation and adaptation options. Among all these dimensions, our study focuses on the assessment of mitigation options, which is a key responsibility of IPCC’s Working Group III (WG III). For this assessment, the IPCC opens a call for science-based literature and synthesises peer-reviewed projections submitted by researchers worldwide. Each projection is an “emissions scenario”, it simulates long-term emissions of greenhouse gases (GHG) up to 2050 or 2100, and is generated using models that incorporate a set of consistent quantitative and qualitative assumptions. In the latest IPCC report, referred here as AR6, the analysis relied on a database of over 2,000 emissions scenarios.
Population projections at a global and regional level, are key inputs to emissions scenarios. The modelling architecture reflects a socio-economic nexus observable in the real world, where human activities have driven the fastest ever growth of GHG emissions. Population size, age composition, and educational attainment through food and energy demand directly affect GHG emissions, whereas, income trajectories specific to each population cohort indirectly affect the decarbonisation patterns.
Different estimations of future population projections
To assess how population may change between now and 2100, models based on “cohort-component methods” are mostly used. These models present analytical descriptions for each factor affecting the growth of population in a certain projected time. These factors are: fertility (measured as number of births), mortality (measured as number of deaths) and migration patterns.
An important distinction is between scenario-based and probabilistic projections. In the former case, projections are developed from a central scenario applying a sensitivity analysis on key inputs. In the latter case, inputs follow probability distributions which tell us whether certain trends have more or less likelihood to happen.
The differences are explained briefly below.
Scenario-based projections
Scenario-based projections can be obtained from deterministic versions of the cohort-component method and represent the standard approach for projecting population. These projections define one likely path of the future population (a ‘central projection’) where the numbers of births and deaths are taken equal to the expectations of their distributions. In Figure 1, the deterministic projections developed by the United Nations (UN) are shown in their two extreme scenarios: a “High” (“Hi”) and a “Low” (“Lo”) fertility scenario, with fertility rates assumed to be 0.5 children higher and lower, respectively, than in the central projection.
Further approaches extend the standard cohort-component model to a “completed cohort fertility at age 50 year”, where 50 years of age are taken as a reference in the modelling of birth rates. This is the case of the population projections of the Institute for Health Metrics and Evaluation (IHME), parameterised considering the Global Burden of Disease Study. The trends shown in Figure 2 show the IHME projections in one of their latest publications, with a focus on sustainable growth.
Expert elicitation can inform the estimation of parameters in models. One of the biggest expert elicitation processes was a community effort expert opinion which, in 2014, led to the identification of a framework representing socio-economic futures, the so-called Shared Socioeconomic Pathways (SSPs). The SSP framework is composed of five pathways and is widely used in emissions scenarios, since a consistent set of socio-economic assumptions (or narratives) is needed for formulating the scenarios. SSP narratives include a sustainability-oriented growth future (SSP1), a “middle of the road” future where trends broadly follow their historical patterns (SSP2), a future of “regional rivalry” (SSP3), a future of inter- and intra-regional inequality (SSP4); and a future of rapid and unconstrained growth in economic output and energy use (SSP5). Compared to other methods, projections based on SSPs explicitly include the education attainment levels as a determining factor of population projections. In Figure 1, the SSPs (version 2.0) show at the lowest levels of global population, the sustainability-oriented SSP1. The fastest growth pertains SSP3. SSP2 locates in between these two extremes.
Probabilistic methods
Probabilistic projections are obtained from models where fertility, mortality, and migration are varied accounting for uncertainty in the past data via an inter-parameter stochastic correlation. These methods were first developed at the International Institute for Applied Systems Analysis (IIASA). Later, in 2015 the UN started releasing probabilistic projections, in addition to the deterministic ones. For instance, in the latest UN Population Prospects, published in 2022, population projections for each country were constructed from a set of trajectories of future outcomes of fertility rate and life expectancy at birth. A central projection for migration was applied to each set of future fertility and mortality outcomes. To fill gaps in parameter estimation, the UN projections used a Bayesian hierarchical model. Other contributions, such as the one by the Resources For the Future (RFF), integrated knowledge gaps on inputs and performed a revision on the outputs via expert elicitation methods.
Different projections underlie different modelling assumptions
Figure 1 and Figure 2 show how the global databases previously discussed would compare with one other.
Figure 1: Population projections from global databases (SSPs and UN)
The figure shows the probabilistic projections from the United Nation Population Prospects (pUN) included between the 5th and the 95th percentile; the highest and lowest deterministic scenarios developed by the United Nations, UN, (dUN), “Hi” and “Lo”; and the SSPs, the Shared Socioeconomic Pathways (SSP1, SSP2, SSP3, SSP4, and SSP5) developed by the International Institute for Applied Systems Analysis (IIASA) . We refer to the SSPs in their version 2.0.
Figure 2: Population from global databases (SSPs, UN and IHME)The figure overlays the Shared Socioeconomic Pathways (SSP) range, with the “Hi” and “Lo” scenarios, and with the IHME scenarios. the scenarios developed by IHME (Institute for Health Metrics and Evaluation) which projected scenarios of faster or slower achievement of education and Sustainable Development Goals (SDG) (“Faster Met Need and Education” (“Faster”), “Fastest Met Need and Education” (“Fastest”), “Reference”, “SDG Met Need and Education” (“SDG”), “Slower Met Need and Education” (“Slower”))
Interesting trends are visible in the international database projections
The fastest projections show a continuous growth through to 2100 (such as the highest UN percentiles , and the scenarios “Hi”, SSP3, and IHME “Slower”).
Most forecasts peak then they either decline or stabilise. The earliest peak characterises the most sustainable scenarios (SSP1, “Lo”, and “SDG”) and sets global population ranging between 8.4–8.9 billion by 2050.
Scenarios conceived to represent medium trends do vary considerably. The IHME “Reference” and SSP2 peak either in 2060 (9.7 billion) or 2070 (9.4 billion), whereas the UN median peaks at 10.4 billion in 2090.
Although SSPs cover a relatively wide population range, compared to alternative lines of evidence, they tend to be more dense in the conservative ranges of population growth.
One of the most relevant input assumptions triggering the different trends is the effectiveness of policies on achieving sustainable development goals related to education attainment, a key fertility-determining factor.
Different estimations of future population projections
Comparing the IPCC assessment in the latest report (here referred as the “AR6 database”), with alternative lines of evidence selected from the previous sections, we focus on the projections of the UN and on the SSPs (version 2.0). The former are important because of their international relevance and known accuracy. The latter are important because of their relevance for framing a socioeconomic narrative in the emissions scenarios submitted to the IPCC.
The inter-quartile global population values of AR6 lie between the 5th and 20th percentile of the UN distribution (Figure 3, left panel). We observe a trend, which is already visible in the previous IPCC report, of progressively smaller interquartile intervals in population projections. This is the result of a strong focus of the scenarios submitted to the IPCC on the SSP2 pathway (Figure 3, central panel). The observation is confirmed by the metadata, which contain a description of the scenarios submitted to the IPCC. Cross-validating our results with the frequencies of the SSP families as calculated from the metadata, we obtained the following share of the scenario narrative: 3.7%, 82%, 2.4%, 1.5%, and 2.1, respectively, for SSP1, SSP2, SSP3, SSP4, and SSP5. The remaining share belongs to narratives unspecified in the metadata. This demonstrates a staggering dominance of the representation of SSP2 assumptions in the database.
Figure 3: Global population in the database used by IPCC in AR6 and in alternative lines of evidenceThe figure shows the population scenarios in the database used in AR6. In the panel on the left, the population scenario ensemble is represented in box plots, overlaid by continuous lines showing the 2022 revision of the UN Population prospects (the 5th, 20th, 50th, 80th, 95th percentiles (pUN) and two extreme fertility scenarios “Lo”, Low variant, and “Hi”, “High variant”). In the central panel, the population scenario ensemble is represented in box plots, overlaid by continuous lines representing the SSPs. In these two panels, the boxes show the quantiles, and the whiskers extend to the rest of the distribution excluding the outliers. Outliers are shown in dots beyond the whiskers. The panel on the right shows a density plot of the population scenario ensemble, taking as representative the 2100 milestone year.
Key findings
The comparative assessment of the scenarios submitted to the IPCC with alternative lines of evidence shows a insufficient level of diversity in the representation of future population at a global scale. Although not shown here, the same analysis replicated at a regional scale leads to the same conclusion of an overrepresentation of SSP2.
The almost exclusive representation of SSP2 is a problem. The comparison with probabilistic approaches proposed in alternative literature reveals that SSPs may cover only a few of the future possible outcomes. If compared with the population projections developed by the United Nations, SSP2 population level appears to be less than the 5th percentile of the global population in 2050 and to be less than the 20th percentile in 2100. Interestingly, the recent update of SSPs in 2024 (version 3.0), not only shows an improved calibration of population trends with historical values for the years 2020 -2023, but also an upwards shift towards higher population projection values compared to the version 2.0. In the updated SSP release, SSP2, for instance, would be closer in values to the UN median. Notwithstanding this important update, we stress the importance for modelling teams submitting to the next IPCC database to explore more extreme pathways, such as SSP3.
The choice of using more extreme scenarios would be not only justified by the outcomes of probabilistic demographic projections, but also required by the need for robust policies to face the climate crisis. Assigning probability outcomes to the different futures, probabilistic projections offer a complimentary view compared to the SSPs. Policy-makers should in fact consider the outcome probabilities in addition to the narrative of scenarios in their negotiations to prioritise among climate change mitigation and adaptation strategies.
Useful links
An extended version of the post contents can be found on our paper published on npj Climate Actions, entitled “Underestimating demographic uncertainties in the synthesis process of the IPCC”. The journal article develops the demographic analysis at a regional scale in addition to the global one presented here.
The code used for supporting the journal paper conclusions are available on GitHub “Climate_Scenario_Data_Science”.
The data used for the calculations are available on a database published on Zenodo, at the following website: “MANET: uncertainty in demographics – data on population projections”.