A common finding in empirical studies of environmental impacts across countries is that they are tightly linked to economic consumption. Richer countries consume more, and therefore have a larger impact than poorer countries (York et al., 2004). Moreover, contemporary societies have a larger impact than they did historically when they had lower incomes. Most research also suggests that while environmental consequences tend to increase with income, the relationship between environmental impacts and economic consumption is less than linear (Haberl et al., 2020; Vadén et al., 2020; York et al., 2003b).

Several theoretical concepts have been introduced to relate economic growth to environmental impacts. One such model is the environmental Kuznets curve (EKC), which was first presented by Grossman and Krueger (1991) to describe a societal evolution in which the relationship between environmental impacts and affluence has an inverse U shape. It stipulates that the levels of environmental degradation are lowest in the societies that are the least and the most affluent. It is named in analogy to the Kuznets (1955) curve, which linked income inequality to economic growth. Relatedly, green growth, or the green economy, is a concept introduced by Pearce et al. (1989) to describe a case in which increased affluence is not related to increased environmental consumption. Among the explanations for the EKC and green growth are growth in the scale of the economy; changes in the composition of the economy, i.e., structural shifts from an agrarian to an industrial to an information-intensive service-based economy; as well as the adoption of novel environmentally-friendly production techniques, policies or investments (Antweiler et al., 2001; Coxhead, 2003; Panayotou, 1993; Shafik and Bandyopadhyay, 1992; Stern, 2004). In their background study for the World Development Report 1992, Shafik and Bandyopadhyay (1992) found strong support for the EKC (also discussed in Shafik (1994)). A number of subsequent studies have provided empirical support for the EKC in different contexts (Cole, 2003; Grossman and Krueger, 1995; Lean and Smyth, 2010), as reviewed in Tan et al. (2014). For example, in a study of five Association of Southeast Asian Nations (ASEAN) countries from 1980 to 2006, Lean and Smyth (2010) found that long-run estimates supported the EKC. However, as highlighted by Stern (2004), the EKC may not apply to all types of environmental impacts.

Researchers have made a distinction between a weak and a strong version of green growth or decoupling (Haberl et al., 2020). In the weak format, this notion simply means that as societies get richer, the relative environmental impact of increased affluence declines. Hence, the environmental impact increases less proportional to the rise in affluence. The stronger version implies that in richer societies, the absolute environmental impact is smaller. This can be conceptualised as a situation in which the elasticity between affluence and the environmental impact transforms fundamentally, not only from having a value between zero and one, but to having a negative value at some (high) level of affluence. The weak case can be described as relative decoupling, while the strong case can be seen as absolute decoupling. While there is considerable evidence of relative decoupling across many environmental dimensions, there is far less evidence of absolute, sector-wide decoupling at the global level. Likewise, the EKC is supported for specific forms of environmental stress in certain regional contexts (Haberl et al., 2020).

In particular, few or no studies have found evidence of absolute decoupling at the global level for either climate change or land use impact, although there are examples at the national level, especially in the former case (Vadén et al., 2020). For example, a recent study based on data from the 1960s to 2015 in the Nordic countries reported that an EKC was observed for per capita carbon dioxide (CO₂) emissions in Denmark, Finland, Iceland and Sweden (but not in Norway) (Urban and Nordensvärd, 2018). A literature review of research in 27 advanced economies reported that the majority (41 of 55) of the examined studies found support for the EKC hypothesis and for the absolute decoupling of CO₂ emissions and gross domestic product (GDP) at the national level (Al-Mulali and Ozturk, 2016).

Studies of the relationship between land use and affluence are rarer and tend to have a narrower scope. In this literature, the evidence presented often does not support the EKC. For example, Pontarollo and Serpieri (2020) studied residential built-up land in 42 Romanian counties from 2000 to 2014, and found an inverted EKC. These results are in line with an earlier study relating land consumption to per capita GDP in 20 Italian regions over the 1980–2010 period (Bimonte and Stabile, 2017), which reported an N-shaped curve with increasing impacts for very high levels of affluence.

As we will illustrate later in our IPAT models, some climate forecasting approaches imply absolute decoupling, while many other climate change and land use scenarios instead point to relative, but not absolute, decoupling.

Efforts to link population, land use, productivity and wages go back to classical economic writings by legendary economists and demographers such as Malthus (1798). During the 1960s and 1970s, when human population growth reached its historical maximum (Lam, 2011), there was increasing concern that population was a major cause of environmental problems. The theoretical perspectives we apply in this article, primarily the IPAT equation, originate from this period. Many of the worst predictions from this time did not come true (Lam, 2011), although subsequent research has confirmed a link between population size and environmental impacts (York et al., 2003a).

Recent empirical and theoretical studies on the association between population growth and environmental impacts at the national, regional and global levels have generally found that the relationships are close to one (Rosa et al., 2004; York et al., 2003b); that is, everything else being equal, most environmental impacts are directly proportional to population size. It is important to note, however, that these calculations generally account for the effect of population net of the level of affluence, which means that an individual in a high-income society still contributes more to environmental problems than an individual in a low-income society. The evidence for relationships that diverge distinctly from one is rather limited (Rosa et al., 2004), even though elasticities can be theoretically expected to vary. Elasticities above one could apply if new, low-quality land is needed for a given unit of consumption after a certain amount of high-quality land has been used. Lower elasticities might be valid if higher population densities lead to more efficient societal organisation. Researchers have also argued that there is a danger in estimating population elasticities from longitudinal data because such coefficients may partly reflect omitted variable bias, and have instead suggested using population as a scaling factor (O’Neill et al., 2012). In line with most previous research, we find elasticities of population very close to one.

York et al. (2003a) related IPAT to three different social theories that address how consumption and population contribute to environmental challenges. First, the human ecology, or neo-Malthusian view, focuses on population growth as a key driver of anthropogenic environmental impacts. According to this perspective, environmental conditions determine human development, and in an IPAT model, this means there is a positive, linear relationship between population and the total environmental impact. This is close to how the IPAT was originally introduced by Ehrlich and Holdren (1971).

Second, modernisation, or environmental economics from a neo-classical perspective, asserts that environmental challenges can be addressed through existing social, political and economic institutions (York et al., 2003a). Accordingly, this view argues that current levels of economic growth, capitalism and globalisation can be maintained without fundamentally harming the planet. Hence, this theory posits a relationship between environmental quality and economic development that reflects the green growth model or the EKC. This implies that the IPAT models would generate values of I that are curvilinear with income; that is, above a certain level of income, I decreases even though A increases. While such relationships have been shown to hold for some pollutants, for instance for air pollution in the form of sulphur dioxide (Grossman and Krueger, 1991), there is less evidence of sector-wide decoupling, as discussed above.

Third, the political economy perspective holds that economic production is the most important factor in this relationship (York et al., 2003a). This position maintains that neither technological development nor political reforms will suffice to adequately mitigate environmental impacts. As producers develop technologies and other methods to reduce labour costs, they will increase their use of shared ecological resources, which means that environmental externalities are inevitable. Economic elites will not internalise the costs voluntarily, and they will use their political power to challenge any reforms that fundamentally change this system, as described further by York et al. (2003a). This perspective maintains that even technologies that reduce ecological damage will, in the end, increase it, because higher profits will be used to escalate growth (and will thus aggravate the effects of economic production). Hence, according to this view, the only solution is an end to economic growth. Thus, this position is consistent with the degrowth perspective and with calls for a fundamental restructuring of society. In the IPAT models, this means that for the total environmental impact (I) to decrease, affluence (A) must decrease as well.

In the early 1970s, Ehrlich and Holdren (1971) and Commoner (1971) introduced the IPAT equation to elucidate the relationship between population, affluence and environmental impacts (Chertow, 2000):

where I is total environmental impact, P is population, A is affluence and T is impact per unit of economic activity. While this equation has been challenged as a simplification because of interactions and dependencies between affluence, population and technology, it formalises essential components of the relationship between these factors. Nevertheless, any interpretations derived from IPAT will have to relate to interaction effects. Chertow (2000) provided a discussion of the strengths and weaknesses of various versions of IPAT in the literature. A further development of IPAT with respect to carbon emissions was proposed by Kaya and Yokobori (1997), in which energy use per unit of consumption and the carbon intensity of energy consumption were distinguished.

As IPAT implies by design that the different factors contribute equally to environmental impacts, it does not allow for hypothesis testing of their respective contributions. To address this limitation, Dietz and Rosa (1997) developed a stochastic form of IPAT, Stochastic Impacts by Regression on Population, Affluence and Technology (STIRPAT). This model does not presume a priori functional relationships between P, A and T; instead, it considers that these associations can be estimated from data, following:

where a scales the model, b, c and d denote coefficients of P, A and T, respectively; and e is a random error term. The coefficients are similar to elasticities in economics, as they reflect the degree to which a percentage change in the explanatory variable generates a percentage change in impact. York et al. (2003b) introduced the notion of ecological elasticity as the responsiveness of an environmental effect to a change in any of the driving forces, specifically population elasticity of impact, b, and affluence elasticity of impact, c, in eq. 2.

STIRPAT implies that if the elasticity is null, the impacts are not affected by changes in population or affluence. If it is one, then there is a proportional relationship between factors; that is, a 1% change in population (b=1) or income (c=1) results in a 1% change in the impact, while higher values imply that the impacts grow faster than the driving factor (York et al., 2003a). For climate, this could apply if higher incomes increase the demand for products with a higher carbon impact, such as airfare. For land use, it may be valid if higher incomes lead to more demand for products and services that increase deforestation. Values above zero and below one indicate inelastic relationships, with impacts that are less reactive. This may happen if higher incomes increase the demand for products with lower environmental impacts, such as services. Values of b and c below zero mean that environmental impacts decrease when population and affluence increase. This would imply that increasing population and income levels enable disruptive innovations that fundamentally alter the way humans impact the environment. In IPAT, the coefficients and the error term equal unity.

STIRPAT has been applied to quantify the relationship between environmental impacts, population and affluence in a wide range of contexts (Dietz and Rosa, 1997; Rosa et al., 2004; Shi, 2003; York et al., 2003a). In addition to such historical assessments, STIRPAT-type models have been used for projections. For example, Liddle (2011) projected carbon emissions from transport and residential electricity in different countries in the Organisation for Economic Co-operation and Development (OECD) from 2010 to 2050. Several related studies have addressed China’s carbon emissions in the coming century (Fan and Lu, 2022; Li et al., 2016). Li et al. (2016) predicted China’s GHG emissions from 2015 to 2035, with parameters derived from a STIRPAT-based analysis of data from 1998 to 2014. They made projections based on three emissions scenarios, with variations in population, affluence, carbon emission intensity, urbanisation, energy consumption structure and economic structure (Li et al., 2016).

However, it could be argued that STIRPAT is too simplistic to model the relationship between greenhouse gas emissions and P, A and T, as it excludes well-documented and complex relationships between, for example, energy use and carbon emissions (Kaya and Yokobori, 1997), or models incorporating energy prices (Liddle and Huntington, 2020). The current study builds on these studies, although it evaluates two ecological challenges and three forecasting approaches. An advantage of STIRPAT is that we can use it across environmental dimensions.

Our first approach to predicting T is grounded in the assumption that future technological developments in the form of annually reduced environmental impacts per unit of production will follow historical trends. It postulates that changes in T in the past have been independent of changes in P and A, and that annual improvements in T will be constant throughout the 21^st century at a rate that reflects average historical rates. In this approach, projections of T are independent of changes in A and P (although the annual rate of change in T is derived from historical records of I, P and A).

For climate impact data, we use records for global GDP (International Monetary Fund [IMF], 2022) and emissions from all greenhouse gases (GHG) in units of gigatons (Gton) of CO₂-equivalents (CO₂e) (Climate Watch, 2022) (Table 2). This implies an average decline in emissions of 3.00% per year between 1990 and 2019, and we set the annual T reduction ahead to this rate. Note that this assumption is consistent with estimates in the literature. For example, the International Energy Agency (IEA, 2022) reported that the emissions intensity of GDP declined by approximately 3% per year in the US and the EU between 2010 and 2021, and that it was reduced by 40% in China between 2000 and 2021 (IEA, 2022), which also corresponds to approximately 3% annually.

For land use impacts, we account for estimates by the Food and Agriculture Organization (FAO) regarding the amount of arable land used for crop production in 1961/1963 (1,372 million hectares (mn ha)), and in 2005/2007 (1,592 mn ha) with data from Alexandratos and Bruinsma (2012). Based on global GDP data obtained from the World Bank (2023), this implies an average decrease in land use per level of affluence of 3.2% per year (Table 3), which is the rate we assume in our land use models in Approach 1. This rate is relatively consistent with previous findings. For example, Lamb et al. (2021) reported that the land needed per unit of agricultural and forestry production declined by an average of 2.7% annually between 2010 and 2017.

In this approach, the calculation of the total land use impact is based on historic levels of population growth and affluence. Thus, an important assumption when fitting the impact to an IPAT framework and setting T=I/(A×P) is that the total environmental impact is a function of consumption in the past. This approach assumes that changes in GDP per capita have driven land use change historically, which is appropriate if human land use is proportional to economic activity. However, if humanity is instead viewed from a more ecological perspective in which all humans are assumed to have similar caloric needs, land use needs will tend to be very similar across all levels of affluence. The reality is most likely somewhere in between, and will vary for different types of land use. Another association between land use and affluence is reflected in the STIRPAT-based projections, which are presented below.

In the second approach, we use projections of technological development grounded in the STIRPAT framework (eq. 2). This implies that we assume that T varies explicitly with population and affluence levels. We use elasticities in STIRPAT that have been estimated under the assumption that the error term, e in eq. 2, and its coefficient, d, are included in T. This is consistent with the early STIRPAT literature, which assumed that it is not possible to operationalise T (Dietz et al., 2007; Dietz and Rosa, 1994, 1997; Rosa et al., 2004). However, note that more recent studies have argued that there are means to measure T, and a variety of factors have been proposed to reflect it. For example, McGee et al. (2015) found that impervious surface area (denoted terrestrial technology) is correlated with carbon outputs, arguing that it should be a measure of T in STIRPAT. Other factors that have been accounted for in the STIRPAT literature include urbanisation, financial development and trade openness, as well as renewable and non-renewable energy consumption (Jia et al., 2009; Usman et al., 2022). Nevertheless, as there is still a lack of consensus regarding how T should be operationalised, we opt for the standard method of keeping T in the error term. Specifically, we make projections of the total environmental impact based on the projections for population and affluence detailed above using literature estimates of elasticities (b and c in eq. 2). Then, we calculate T by the IPAT identity, according to:

When we examine changes over time, we then get:

If population and affluence increase monotonically, T/T0 slopes downward for elasticities below one (b<1 and c<1). For higher elasticities (b>1 and c>1), T/T0 instead increases over time. However, if b and c diverge, with one but not the other above unity, then T depends on the relative difference between P and A.

The elasticities for P and A in our projections are grounded in a review of previously estimated STIRPAT models. For climate impact, we use elasticities obtained from two literature reviews of CO₂ and GHG emissions (Liddle, 2015; Pottier, 2022). (This definition of climate impact varies slightly from the one that we use in Approach 1, which is based on CO₂e, as this method assumes that the general trends are consistent across these different definitions of climate impact.)

Regarding elasticities with respect to population, we use the median of the cross-national, inter-temporal STIRPAT studies listed in Liddle (2015). In this data set, we exclude short-run data when long-run data are available, and we use disaggregated estimates (per income level) rather than overall estimates when both forms of data have been published, implying N = 29 data points. Concerning elasticities with respect to affluence, we additionally consider the review by Pottier (2022), who presented income elasticities of GHG or CO₂ emissions from various countries and time periods. We assume the upper bound in cases in which no other data were listed by Pottier (2022), generating 36 data points. We then use the median of the entries in Liddle (2015) and Pottier (2022) (Table 4).

For land use impact, we assume the elasticities of population and affluence published by Rosa et al. (2004), in which STIRPATs were estimated using data from 142 countries (Table 5). As proxies for land use, we use the results for arable land and grazing in Rosa et al. (2004), ultimately using the mean of the two analyses. Because we depend on this single study to model land use impact, the elasticities that we use are less reliable than those we use to model climate impact, which reflect the results of two comprehensive literature reviews. This imbalance is undesirable but unavoidable, because the literature on STIRPAT has mainly focused on climate impact (e.g., Wang et al., 2011 and Xiong et al., 2019). Studies with land use as the dependent variable are still scarce, even though there are alternative approaches, such as those that focus on the ecological footprint, a general measure that aggregates over many types of environmental stress (Dietz et al., 2007; Jia et al., 2009; Usman et al., 2022).

Note that the empirical literature suggests that the impact elasticities of population are near unity for both climate impact (Table 4) and land use impact (Table 5). However, elasticities of income are generally below one, which signals relatively inelastic relationships in which environmental effects increase at a slower rate than GDP. The median elasticity of income for climate impact, 0.58 (Table 4), is only slightly larger than the elasticity for land use impact, 0.50 (Table 5). Moreover, b for climate impact is above unity (1.12) while c is not (0.58), which implies that T depends on the relative difference between changes in P and A. However, for land use, elasticities of both population and income are less than unity, which means that increases in P and A result in declining T over time.

Lastly, we consider forecasts in the literature regarding impact I, and we examine how they relate to T in the IPAT framework (Approach 3). For climate, we apply the IPCC’s (2022) most recent models, which use a scenario matrix architecture in which socio-economic patterns are reflected in five key SSPs. In these models, various climate mitigation strategies are represented by five distinct levels of radiative forcing (Fujimori et al., 2018), which reflect concentrations of GHG and other factors of climate warming in 2100, in units of watts per square meter (Supplementary material, Figure S.1, available at https://doi.org/10.1553/p-n5en-z38a). These levels reflect five distinct Representative Concentration Pathways (RCPs), which account for different behavioural trends and efforts to curb emissions relating to, for example, energy generation, novel technologies and transportation.

Specifically, our IPAT model for climate change is based on the IPCC’s (2022) middle-of-the-road projection SSP2 with RCP 4.5, which was developed by IIASA (Fricko et al., 2017; Riahi et al., 2017). The SSP2 assumes the continuation of current social and economic trends and moderate mitigation efforts. We operationalise climate impact as GHG emissions in terms of unharmonised emissions of Kyoto gases in units of megatons of CO₂e per year, with data sets obtained from the SSP Public Database (IIASA, 2023). Note that this conceptualisation varies slightly from the one we use in Approach 1, which is based on all GHGs in units of CO₂e, and in Approach 2, which reflects CO₂ and GHG emissions. Thus, it is an assumption in our models that patterns over time are consistent across these three specifications. To obtain a trajectory for T, we divide the projection of climate emissions with trajectories for population and per capita GDP in SSP2.

For land use, we also account for forecasts in the IPPC’s middle-of-the-road scenario SSP2, which assume that current trajectories in the land sector will continue, with medium levels of regulation and technological change and material-intensive consumption, as well as an increase in animal calorie share and ongoing tropical deforestation (Popp et al., 2017). These forecasts assume that the total cropland in 2005 was 1.5 bn ha, and that the use of cropland will increase by 231 mn ha in the period from 2005 to 2100 due to increased demand for food and feed (Popp et al., 2017). We account for this development in relation to the projection of GDP in the SSP2 (Dellink et al., 2017) to calculate the annual decrease in T (Table 6). The approach thus assumes that T responds to the inverse of P×A ahead.

Calculated this way, the annual reduction of T is very dramatic, because land use has been relatively unchanged over the last decades, while affluence and population have grown considerably. This approach thus assumes that the levels of agricultural intensification, mechanisation, capital investment and nutritional inputs to the land that have taken place since the 1960s (i.e., the Green Revolution) will continue over the 21^st century (Evenson and Gollin, 2003). For low-income countries, this reflects the adoption of agricultural intensification practices that have mainly been applied in more affluent countries. For high-income countries, it implies the continuation of agricultural intensification at the same spectacular rate as in the last half-century (Evenson and Gollin, 2003).

Note that it is a major simplification in our models that we do not consider that A, P and T are endogenously related to each other. This approach diverges from historical trends and forecasts in which these endogenous relationships were accounted for either explicitly, as in the IPCC’s models, or indirectly, through various known and unknown causal links reflected in historical records. Thus, our IPAT-type extensions are based on strong assumptions that likely do not hold regarding the independence of and the temporal invariance in the relationships we have studied. Therefore, our approach contrasts with those of prominent researchers who have theorised that P and T are interdependent (Boserup, 1965). However, we believe that our approach is still conceptually helpful because it uses IPAT as a theoretical tool to explore how different methods used to predict environmental impacts relate to one another.

We also examine to what extent changes in different parameters affect I in the different models. Examining changes in P speaks directly to theories linking environmental effects to population, such as the human ecology view. We model population using four diverging trajectories, whereby our standard middle-of-the-road trajectory is the medium scenario in the WPP (2022). We first change the medium scenario with 10% in 2100 and assume a linear change up to that point. We also use the projections for low and constant fertility in the WPP (2022). The low scenario represents rapid fertility decline in Asia and Sub-Saharan Africa, while the (arguably quite unlikely) scenario of constant fertility characterises population trajectories with the fertility levels of 2022 extrapolated into the future. In practice, population policies aimed at reducing fertility could provide non-coercive support for family planning or less government support for childrearing, while scenarios aimed at promoting fertility could instead involve increased socialisation of childrearing (Kolk, 2021).

For variations in affluence, which are directly related to degrowth from a theoretical perspective, we similarly model per capita GDP based on a 10% increase and a 10% decrease compared to the 2100 level in our standard scenario. Examining the effect on I of such a reduction in income can be seen as an assessment of degrowth as a strategy to address environmental challenges. It is therefore associated with the political economy viewpoint.

We also explore the impact of a range of possibilities regarding technological development. First, we consider variations in T of 10% in the models based on extrapolations of historical data (Approach 1). Second, we consider the first and the third quartile of impact elasticities of population and affluence in the reviewed STIRPAT literature (Approach 2). Third, we consider different radiative forcings in 2100 in the SSP2 (IPCC, 2022) (Approach 3) (Supplementary material, Figure S.1, left panel). These levels imply varying T, since they assume that the socio-economic patterns (P and A) are similar while the climate behaviours, policies and technologies (T) differ. We therefore obtain different projections for T by dividing climate forecasts with population and GDP trajectories in the SSP2 (Supplementary material, Figure S.1, right panel). This assessment relates to the modernisation perspective, as it focuses on effects of developments in T. We cannot use the corresponding method for land use because we have found only one relevant forecasting scenario in the literature (Approach 3).

The developed IPAT models for climate and land use impacts are shown in Figure 1. The historical extrapolations (Approach 1) result in climate impacts at the end of the century that are about half of current levels (Figure 1, left panel). This finding is relatively similar to the IPCC’s forecasts (Approach 3) (Figure 1, right panel), which imply that the likelihood of peak global warming staying below 2 °C is only 8% (5% to 95% percentile: [2% – 18%]), and that the global mean temperature in 2100 will likely increase to around 2.7 °C (IPCC, 2022). Our model illustrates that even scenarios such as the SSP2 RCP4.5, which is likely to be insufficient to mitigate climate impact, still imply a T that is very close to zero in 2100 (T = 0.06) (Table 7). (It echoes the substantial increases in population and affluence that the IPCC expects by then, with levels of P and A that are 1.32 and 4.54 of current readings, respectively.) These scenarios indicate that the IPCC is very optimistic, as even their middle-of-the-road scenario implies that the world will commit to substantial climate mitigation efforts.

The models based on STIRPAT are more alarming (Approach 2) (Figure 1, central panel). The calculated effects in 2100 are closer to catastrophic climate impacts (e.g., SSP5 RCP 8.5), and project a much larger T than in Approach 1 (Table 7). Moreover, in sharp contrast to environmental targets, our land use projections based on STIRPAT (Approach 2) and forecasts in previous research (Approach 3) suggest that environmental impacts will increase rather than decrease in 2100. The latter approach indicates that land use impact will increase to 1.13 of current levels in 2100, despite estimates that T will decrease to about one-fifth of today’s levels (Table 7). This is an example of weak decoupling, since A is expected to rise considerably by then. The lack of evidence of strong decoupling is disconcerting, considering that many environmental studies argue that current levels of human impact on land systems and biodiversity are already unsustainable (Steffen et al., 2015; Dasgupta, 2021).

The left panel in Figure 1 depicts the land use impact if historical trends continue and T declines by 3.2% per year (Approach 1). We complement this finding with a model that accounts for extrapolations of historical trends based on cereal yields, generating an annual T decrease of 2.1% per year (Popp et al., 2017) (Supplementary material, Figure S.2). The difference between these two models is that the first represents total production, while the second refers only to production in the form of crop yields, reflecting intensification technologies such as industrial fertilisers for a given land area (although crop yields will be higher at lower levels of land use as more productive land is used). These models are relatively consistent in that they both result in significant decreases in land use impacts in 2100. Approach 1 implies that the land use impact at the end of the century will be only 0.07 of current levels (Table 7).

We further explore how changing trajectories of population, affluence and technology would change environmental impacts as seen through an IPAT lens. Here, we want to illustrate the importance of the time horizon and the difference in T across the two environmental dimensions and the three modelling approaches (Table 1). These aspects influence the extent to which changes in P and A will actually affect I. Our models can thus be used to highlight the relevance of addressing environmental concerns through policies aimed at reducing population or affluence.

The overall interlinkages of P, A and T have been discussed extensively in the environmental literature from a number of different theoretical perspectives. It is our hope that our IPAT-type modelling can clarify how these different theories implicitly (and occasionally explicitly) put different weights on the various parts of this identity. The original proponents of IPAT highlighted the negative consequences of anticipated growth in P and A (Ehrlich and Holdren, 1971), and they were thus sceptical of improvements in T. Contemporary environmentalists who focus on degrowth are similarly sceptical of such improvements, but they concentrate almost entirely on A. In contrast, researchers proposing green growth put considerable emphasis on mitigation through T. Our models have highlighted how substantive such technology improvements must be in light of mainstream forecasts of global GDP, especially if one considers contemporary consumption to be (already) above sustainable levels – a view that is common in the environmental literature. Broadly, our models are consistent with previous research that reasoned that the IPAT framework highlights T as the most dynamic and important part of the equation (Chertow, 2000). They are especially helpful for pinpointing the situations in which large reductions in T are feasible, and P and A are therefore of less relative importance. Our models also illustrate the extent to which technological improvements underlie different actors’ forecasts.

In line with earlier research using this approach, our STIRPAT-type modelling has highlighted the difficulties in disconnecting affluence and population from environmental impacts. These models, which are derived from observed trends over time and across regions, consistently give the highest environmental impacts. They suggest that to reach global environmental goals, our economic, technological and political systems may need to work in fundamentally different ways than they have in recent decades. Business-as-usual scenarios likely imply limited decoupling of I from and P or A. In our STIRPAT-based models, the differences between projections across environmental dimensions are driven by variations in the impact elasticities of P and A, which imply alternative paths for T. Thus, our models highlight that in standard climate models such as the SSP2 RCP 4.5, substantial weight is given to future technological improvements, and the near-total decoupling of affluence and population from climate impacts is assumed. As such, the IPCC’s middle-of-the-road prediction clearly reflects a green growth perspective.

Another inference that can be drawn from our models is that population policies (“fewer people”) and degrowth (“less consumption”) are less relevant strategies for addressing climate impacts than land use impacts if the mainstream SSP2 trajectories are assumed to be realistic. This is because there are promising technologies for mitigating climate impacts, especially within the energy supply sector (see Davis et al., 2018). By contrast, for land use change, population policies or reduced consumption appear to be more relevant. This is underlined in our STIRPAT-based approach, which indicates that land use impacts will increase substantially from levels that many argue are already unsustainable. We are not aware of many empirical and analytical studies on land use in the scientific literature that would clearly contradict such a perspective.

Our models suggest that climate impacts may be decoupled from affluence (and, relatedly, from population). These findings are supported by existing technologies and policies that would substantially reduce climate impacts, as shown in the right panel in Figure 1. Still, it is important to note that achieving green growth in the context of climate change means that the impact per dollar spent needs to decrease steeply for I to go down sufficiently. The reduction in T must be larger than the anticipated increase in P×A; for example, if P = 1.2, and A = 5, then T must be 1/6 for a constant I, or 1/12 for reductions in I by half in a green growth scenario. Assuming access to clean energy, it is imaginable that the carbon elasticities of population can be reduced to near zero, which implies that there will be no additional climate impact of an extra person in the world. We have one IPAT trajectory that reflects this positive trend (Figure 2, right panel).

However, we have found little evidence of the corresponding pattern for land use change. Our models have thus illustrated that the anticipated technological and behavioural developments in carbon intensity do not necessarily translate into improvements in land use systems. Instead, our models have shown that while achieving green growth may be a realistic goal in the context of climate change, it is less applicable to land use. Consequently, the political economy view that calls for a stagnant economy (degrowth) is likely to be more relevant for mitigating land use impacts than for addressing climate impacts. In parallel, our results suggest that population policies, which are related to the human ecology perspective, may have a larger effect on the likelihood of reaching global goals for land use change than for climate change. Policymakers need to be aware of the different approaches that will be required to address these different environmental impacts.

Our models are generalisations of the world, and therefore suffer from several limitations. They are based on several theoretical simplifications, the most important of which is probably that we have largely avoided examining potential endogeneities between P, A and T. By harmonising the different types of models into one framework, we have thus excluded some of the causal ways that these aspects may be linked; hence, our models should primarily be seen as putting different approaches into a similar scale and a similar theoretical perspective to allow for comparisons. The models should be considered in this light, rather than as precise projections of environmental impacts. Consequently, we see this article as mainly representing a theoretical contribution. Any single IPAT-type model we have presented can be discussed, as different operationalisations or competing models in the secondary literature could have been chosen and may have been equally plausible. Nevertheless, we have found that this exercise has been helpful in interpreting the radically diverse perspectives found in the environmental literature.

Moreover, it is important to keep in mind that we have relied solely on experts in this study, as our modelling inputs are grounded in a literature review. Thus, the study has assumed that this literature is sufficiently rich to capture key dynamics. While the STIRPAT literature on climate impacts is relatively extensive, it has been more challenging to find related studies on land use. Moreover, we have assumed elasticities to be temporally constant, and it could be argued that our STIRPAT models for climate change are simplistic compared to models that also include factors such as energy use and future energy prices (Kaya and Yokobori, 1997; Liddle and Huntington, 2020). Our method is a simplification, which also relates to variations across income levels. There is only limited research on how elasticities change over different levels of affluence, but the few studies on this topic have suggested that the estimates are relatively robust (see e.g., Liddle, 2015). The modelling in this study is based on our own assessments of what constitutes relevant empirical inputs. For example, we have assumed that the slightly different definitions of climate impact that we adopt in the three modelling approaches (Table 1) do not have a large influence on the findings in general. Thus, our modelling assumes that variations in climate impact over time are not fundamentally different across these different specifications.

Another related limitation is that our STIRPAT-based climate impact model (Approach 2) uses the median of the listed studies in two literature reviews (Liddle, 2015; Pottier, 2022), even though Liddle (2015) argued that published studies in this field are sometimes flawed. Nevertheless, we note that the values that we assume in our baseline model are within a reasonable range based on Liddle’s (2015) recommended estimates, as discussed in the subsection Technology in the sensitivity analysis. Furthermore, the results of our sensitivity analysis that specifically accounts for Liddle’s (2015) recommended values (Supplementary material, Figure S.7) largely mirror our baseline results (Figure 1, central panel), although they also suggest that the climate impacts in 2100 may be even more severe than in our baseline projection. We refer to Liddle (2015) and Liddle and Huntington (2020) for more thorough discussions on the wide-ranging estimates for carbon emissions elasticities of population and affluence in the literature, for example, relating to cross-sectional dependence and variations over time.

Another potential concern is that we use a single set of trajectories for P and A in our IPAT modelling, while the various forecasting scenarios that we base our models on use different, but related, trajectories. The most important difference between these approaches is that we use the UN WPP (2022), while many SSP-based projections have used population forecasts from the Wittgenstein Centre for Demography and Global Human Capital and IIASA (KC and Lutz, 2017). The harmonisation of a diverse set of scenarios was necessary in our approach, which aimed to synthesise a sprawling literature and a large set of models with different assumptions regarding P and A. This highlights the importance of interpreting our results primarily as a theoretical tool for understanding conceptual differences, rather than as competing forecasts of environmental problems.