## Main

The Paris Agreement sets the framework for international climate action. Within that context, countries are aiming to hold warming well below 2 °C and pursue limiting it to 1.5 °C. How such global temperature outcomes can be achieved has been explored widely in the scientific literature1,2,3,4 and assessed by the IPCC, for example, in its Fifth Assessment Report (AR5; ref. 5) and its Special Report on Global Warming of 1.5 °C (SR1.5; ref. 6). Studies explore aspects of the timing and costs of emissions reductions and the contribution of different sectors3,7,8. However, there has been critique that, with the exception of a few notable studies9,10,11,12, the scenarios in the literature first exceed the prescribed temperature limits in the hope of recovering from this overshoot later through net-negative emissions13,14,15,16. Some pioneering studies10,11,12 have explored implications of limiting overshoot through, for example, zero emissions goals, or have looked into the role of bioenergy with carbon capture and storage (BECCS) in reaching different temperature targets9. All these studies have relied on one or two models and/or a limited set of temperature targets.

We bring together nine international modelling teams and conduct a comprehensive modelling intercomparison project (MIP) on this topic. Specifically, we explore mitigation pathways for reaching different temperature change targets with limited overshoot. We do this by adopting the scenario design from ref. 11 and contrast scenarios with a fixed remaining carbon budget until the time when net-zero CO2 emissions (net-zero budget scenarios) are reached with scenarios that use an end-of-century budget design. The latter carbon budget for the full century permits the budget to be temporarily overspent, as long as net-negative CO2 emissions (NNCE) bring back cumulative CO2 emissions to within the budget by 2100. This approach dominates the current literature and leads to a temporary overshoot of the associated temperature target. Importantly, the earlier introduced ‘net-zero budget scenarios’ limit cumulative CO2 to a maximum without exceeding the emissions budget. These scenarios thus keep global warming below a certain threshold (without exceeding it) and stabilize the temperature thereafter.

The new pathways fill important knowledge gaps. First, they cover the range of carbon budgets consistent with low stabilization targets in a systematic way and across a wide range of diverse global models. The pathways thus explore important uncertainties, including the attainable scenario space across different models and target definitions. This information is critical for international assessments, such as those by the IPCC17. Secondly, we explore the impacts of the country pledges from the post-Paris process for the attainability of overshoot and non-overshoot targets. Thirdly, we investigate salient temporal trade-offs with respect to mitigation costs; and finally we explore distinct differences in terms of the possible regional and global designs of net-zero CO2 emissions systems. The main narratives of the pathways and assumptions are provided in Table 1.

## Implications for emissions pathways

Reaching stringent temperature targets with limited overshoot, requires a pronounced acceleration of the near-term transformation towards net-zero CO2 emissions. Staying within a budget of 500 GtCO2 (consistent with a median warming of 1.44–1.63 °C), for example, requires CO2 emissions to reach net zero between 2045 and 2065 (range across models). When an ‘end-of-century’ carbon budget is used, the time of reaching net-zero CO2 emissions is delayed between 5 and 15 years (to 2060–2070). This delay, combined with the higher emissions over that period, results in 0.08–0.16 °C higher peak temperatures compared to scenarios that are identical in all but their allowance to overshoot the carbon budget.

A broad set of behavioural, biophysical, economic, geophysical, legal, political and technological factors render transformations to net zero more or less challenging18. The modelling exercise here informs primarily challenges related to economic, geophysical and technological feasibility. The lowest attainable net-zero CO2 emissions budget (limiting overshoot) is 400–800 GtCO2 across the models (assuming immediate implementation of ambitious policies and a middle-of-the-road socioeconomic development19). This budget range corresponds to a median peak warming during the twenty-first century between 1.42 and 1.72 °C. Weak near-term policies that result in higher GHG emissions over the next decade, such as those implied by the current nationally determined contributions (NDCs), will affect the lowest attainable carbon budget. We estimate that the NDCs (Methods) will lead to GHG emissions of 46.8–56.3 GtCO2e by 2030, which is substantially higher than the range of cost-effective emissions pathways consistent with 2 °C (25–48.6 GtCO2e), let alone 1.5 °C, by 2030 (19.4–35.3 GtCO2e). We adopt the definition of 1.5 and 2 °C goals from the SR1.5 (Methods). Assuming NDCs are not tightened and comprehensive climate policies are thus delayed until after 2030, the lowest attainable net-zero CO2 budget across the models is 500–1,200 GtCO2, which corresponds to a warming of 1.61 and 1.89 °C. Current NDCs thus put limiting warming to 1.5 °C out of reach on the basis of the biophysical, economic, geophysical, technological and economic feasibility dimensions reflected by the models applied here. Other feasibility dimensions, such as behavioural, legal, political or social aspects, can affect these ranges further, although this study does not explore their impact.

The pathways feature net-negative emissions from a few megatons to ~500 GtCO2 across models, depicting a techno-economic potential for declining warming after its peak between 0.13 and 0.34 °C by 2100 (Fig. 1b). This temperature reversal is mainly driven by NNCE but can also be partially the result of reductions in non-CO2 forcers20 (Methods; see Supplementary Figs. 1.16,9,10 for the relationship between peak temperature, overshoot and NNCE).

The net-zero budget scenarios allow for the systematic quantification of the residual non-CO2 emissions consistent with different peak temperature levels (Fig. 1c). A large share of these residual non-CO2 emissions is caused by the agriculture, forestry and other land-use (AFOLU) sector, most prominently by enteric fermentation (CH4) and fertilizer use (N2O). The annual residual non-CO2 emissions in the second half of the century range from slightly above 3 to >10 GtCO2e highlighting once more the dual importance of CO2 and non-CO2 mitigation measures (Fig. 1c). We emphasize that while our net-zero budget scenarios exclude NNCE, for many policy goals, including those of the Paris Agreement21 or the climate neutrality target of the EU22, NNCE are needed to balance residual non-CO2 emissions and reach net-zero greenhouse gas emissions16.

## Upfront costs and long-term economic benefits

The IPCC AR5 emphasizes that mitigation costs would rise over time as a result of efforts to limit climate change5. These mitigation costs traditionally reflect the impacts on gross domestic product (GDP) while ignoring the benefits of mitigation due to avoided impacts5. Typically, relatively smaller mitigation costs are reported in the near term through to 2030 compared to the medium term (2050) or the very long term by 21004,5,11,23. This evolution is primarily a result of most integrated assessment model (IAM) studies focusing on targets for the end of the century, which, by design, favours postponement of mitigation action until later in the century11,24.

Scenarios that limit temperature overshoot (that is, the net-zero budget scenarios), pace mitigation actions differently, requiring much more rapid emissions reductions in the near term (Fig. 1 and Supplementary Fig. 1.1-8). Avoiding overshoot is thus associated with higher upfront investments and higher near-term mitigation costs. We find that GDP in the near term is 0.5–4.8% lower in scenarios that keep warming below 1.5 °C with no or limited overshoot and 0.1 to ~1.6% lower in scenarios that limit warming to 2 °C with no or limited overshoot (compared to end-of-century budget scenarios with overshoot).

Once net-zero CO2 emissions are reached, however, the mitigation effort in the net-zero budget scenarios with limited overshoot can be relaxed, since no further emissions reductions are necessary. This results in a slow-down, or even decline, of carbon prices while keeping CO2 emissions constant at net zero (Supplementary Fig. 1.1-6). During this phase (in the latter half of the century) the economy accelerates since lower mitigation expenditures are required and GDP growth is becoming higher in the net-zero budget scenarios with no or limited overshoot (compared to the end-of-century budget scenarios).

Perhaps most importantly, we find this GDP rebound in the long term to be by far larger than the upfront dampening effects on GDP due to efforts to limit temperature overshoot. In other words, the higher near-term GDP losses of limiting overshoot are fully compensated by higher GDP growth in the second half of the century (Fig. 2a). The absolute GDP levels in the long term (2100) are thus higher across all models and mitigation scenarios that limit the overshoot (Fig. 2a), which is consistent with the reduced stringency of the target at the end of the time horizon. This observation holds also on the regional level with relatively higher losses in the near term in fossil fuel-exporting regions (Supplementary Fig. 1.1-12). For a 1.5 and 2 °C target, the long-term GDP (2100) is about 1.2% higher (range 0.1–2.4%) in scenarios that limit overshoot. Similarly, the peak carbon prices over the course of the century—a relevant indicator measuring policy stringency and disruptiveness25,26—are substantially lower in most scenarios without overshoot (Supplementary Figs. 1.1-6 and 1.1-7). The difference between net-zero budget and end-of-century budget becomes smaller at weaker temperature targets and diminishes fully at high budgets where CO2 emissions do not need to become net zero over the course of the century (depending on the model this corresponds to a budget of 1,000–2,500 GtCO2).

Across all IAMs we find that accelerating the transformation towards net-zero CO2 emissions would have benefits for the long-term GDP, even without considering the benefits of avoided impacts that are traditionally not included in the type of scenario analysis presented here.

From a methodological perspective, it is important to emphasize that our results are not suggesting that avoiding overshoot is leading to lower ‘overall’ cumulative mitigation costs over the entire century. The perceived overall cumulative cost of each pathway depends critically on the discount rate and how one weights the near-term GDP losses against the long-term GDP gains24. To explore the impact of the discount rate on the overall cumulative costs we conduct an ex-post-sensitivity analysis, systematically varying the discount rate between 0 and 5% (and apply them to the existing cost pathways of the scenarios). We find that discount rates of less than about 2% would make the perceived cumulative costs of most 1.5 and 2 °C scenarios overall less costly without overshoot (see Fig. 2c for the cumulative GDP losses and Supplementary Fig. 1.1-13 for the net present value (NPV) of the carbon price). Assuming higher discount rates on the other hand would favour relatively delayed mitigation with overshoot. Perhaps most importantly, irrespective of the discount rate, we find long-term GDP in 2100 to be higher in scenarios with limited or no overshoot (Section 1.2 of the Supplementary Information gives a discount-rate sensitivity analysis).

Another important cost factor is the NDCs. Their modest mitigation effort in the near term leads to relatively reduced costs in 2030 (Fig. 2b). The NDCs, however, have negative economic effects from 2040 onwards, where the acceleration of the mitigation effort for limiting temperature to 2 °C would result in relatively higher GDP losses for the entire century (Fig. 2b).

## Net-zero CO2 emissions systems

Our study explores a range of diverse net-zero CO2 emission systems. The distribution of the emissions reductions across sectors, space and time depends critically on a number of factors, including relative abatement costs, the inertia of sectors against fundamental structural changes and the ability to reduce emissions in different sectors to zero or even further to NNCE. In a zero CO2 emissions system, some sectors and regions continue to act as sources of residual emissions, which are balanced by sinks in other sectors/regions that remove CO2 from the atmosphere to achieve overall net-zero emissions (Fig. 3).

The magnitude of the sinks differs across the assessed models, ranging globally from ~5 GtCO2 per year (REMIND-MAgPIE and GEM-E3 models) to >10 GtCO2 per year (POLES and WITCH, Fig. 3). Afforestation and reforestation, as well as BECCS (see also sensitivity analysis in Section 1.6 of the Supplementary Information), are responsible for the bulk of the gross negative emissions in the scenarios. Their contributions vary markedly though. AFOLU and energy supply sectors act as sinks, while the demand-side sectors (transport, buildings and industry) are primarily responsible for any of the remaining residual emissions sources. The results emphasize the importance of addressing the residual emissions in these demand sectors, which in turn would lower the pressure on supply-side transformations, including the need to enhance the anthropogenic sink. In some models (for example, REMIND-MAgPIE and GEM-E3), industrial processes, feedstocks and/or the buildings sector reach zero emissions or contribute smaller amounts of NNCE. Electrification, efficiency and demand reductions play a critical role across all demand sectors.

The sectors differ with respect to the timing of when they achieve net-zero CO2 emissions. Globally CO2 emissions reach net zero around 2050–2075 and 2055–2100 in 1.5 °C pathways with low overshoot and 2 °C pathways, respectively (Fig. 1d and Supplementary Fig. 1.1-4). However, in most scenarios, the AFOLU sector is fully decarbonized 10–40 years earlier and the energy supply sector often 10–20 years earlier (Fig. 3c). The demand-side sectors on the other hand (buildings, industry and transport), with many small dispersed and difficult-to-abate emissions sources, do in many instances not reduce emissions to zero throughout the century when considered as part of an integrated net-zero strategy (Fig. 3c). Across demand sectors, limiting demand through improved efficiency and behavioural change, as well as rapid electrification, play an important role. Avoiding non-CO2 emissions is critical in the agricultural sector where reductions of N2O and CH4 emissions are achieved. Carbon dioxide removal (CDR) plays three important roles in all scenarios (also in scenarios that avoid NNCE): (1) helping to accelerate emissions reductions early in the century, (2) offsetting residual emissions to achieve net-zero CO2 and (3) achieving net-negative emissions in the long term to reduce warming after the peak (if necessary). See also Section 1.3 on the role of CDR in the Supplementary Information.

Also the timing of when different regions reach net-zero CO2 emissions varies considerably (Fig. 3c). Regions with a larger low-cost CDR potential and large-scale availability of land resources, such as Latin America and the Reforming Economies including Russia, tend to decarbonize first and much earlier than the world average (Supplementary Figs. 1.1-14 to 1.1-16). This sequence in the timing of decarbonization is because the pathways describe a cost-effective response across regions, implicitly assuming that there is some degree of coordination and financial collaboration that allows regions to tap into mitigation options that stretch across regions (when needed). Regions with high projected economic catch-up and continued population growth in the future and/or lower CDR potentials, such as Africa, parts of Asia and the Middle East, thus tend to reach net-zero CO2 emissions relatively later. In some scenarios, these regions even maintain some residual emissions throughout the century. Generally, today’s rich economies of the Organisation for Economic Co-operation and Development (OECD) reach net-zero CO2 emissions domestically about the same time as the global average if climate change mitigation is to be achieved cost-effectively. In a world in which rich OECD economies aim at taking up a climate leadership position, or to reflect higher historic responsibility, their net-zero CO2 timing could well be set earlier.

## Discussion

We have shown that scenarios with an accelerated transition towards net-zero emissions avoid a systematic (discounting) bias in favour of temperature overshoot. Furthermore, we identify sectors and regions that may provide an entry point for rapid and deep cuts towards zero CO2 emissions and illustrate that avoiding overshoot would be associated with economic gains in the long term (even without considering benefits of avoided climate impacts). Our study uses a net-zero carbon budget design which is a close proxy for peak warming. Other scenario designs, for example, limiting global temperature directly or using different metrics for the temperature equivalence, are possible as well10,12 and would affect the substitution dynamics of different greenhouse gases.

Net-zero CO2 emissions systems imply the deployment of CDR measures with very different implications for the sustainability of the overall mitigation portfolio. BECCS may lead to possible trade-offs with sustainable development, depending on the scale of deployment, implementation practice and local context18,27,28. The CDR portfolio thus varies across models, providing policy flexibility with respect to technology choices. Some pathways rely on BECCS (for example, REMIND-MAgPIE), while other pathways rely more heavily on nature-based solutions or use more balanced approaches across these options (WITCH, POLES and MESSAGEix-GLOBIOM). The IAMs do not include all possible CDR options29. CDR can serve three purposes in mitigation pathways: it can help to accelerate early emissions reductions, thus supporting achieving net-zero CO2 emissions as soon as possible; it can offset residual emissions from sectors that might be difficult to decarbonize completely; and it can provide a long-term risk-hedging strategy to generate net-negative emissions and gradually reverse warming if desired. In all three instances, deep reductions in gross CO2 emissions remain crucial.

The importance of demand-side measures cannot be overemphasized30,31,32. Generally, efficiency, behavioural change and the deployment of granular and small-scale technologies is enabling rapid technology diffusion and substitution processes33,34,35. In addition, demand-side mitigation is key for reducing residual emissions. Bottlenecks include particularly the industry sector’s demand for carbonaceous fuels and the transport sector, as well as the materials and consumption goods sectors. Particularly, material substitution and options for demand-side electrification need to be represented in a more bottom-up and granular fashion in the models.

The regional scenario results indicate opportunities for mitigation and do not imply political feasibility, which would need to consider a diverse set of ethical and other considerations36. In fact, we find large differences across regions to reach net-zero CO2 emissions and the pathways suggest that, from an economic perspective, it will be most attractive if some regions act as sources while others act as sinks. Achieving such an effective solution, however, poses a major challenge because it requires international collaboration and markets for cross-regional policy frameworks. In this context, it is encouraging to observe that net-zero emissions targets in a number of key countries/regions, like China37, the European Union (EU)38, Japan39 and South Korea40 are broadly consistent with the pace of the transformation as depicted by our study.

## Methods

The nine IAM frameworks, drawn on in this study are AIM-Hub (refs. 41,42), COFFEE (ref. 43), GEM-E3 (refs. 44,45), IMAGE (ref. 46), MESSAGEix-GLOBIOM (ref. 47), TIAM-ECN (ref. 48), POLES (ref. 49), REMIND-MAgPIE (refs. 50,51) and WITCH-GLOBIOM (refs. 52,53). The models span a wide range from least-cost optimization to computable general equilibrium models and from game-theoretic to recursive-dynamic simulation models. Such diversity is beneficial for shedding light on those model findings that are robust to diverging assumptions and model structures. Of particular importance for the current study is that all models have a detailed coverage of the energy sector and seven out of the nine models in addition represent land-use changes and related mitigation measures in detail. All models, however, represent land-based negative emissions options related to either bioenergy production and/or reforestation. Some of the models consider in addition the possibility of negative emissions through feedstocks in industrial products (GCAM and COPPE) and three models (POLES, WITCH and REMIND-MAgPIE) in addition also considers direct air capture. Cost assumptions of different technological CDR options are summarized in Section 1.4 of the Supplementary Information and a sensitivity analysis on BECCS is provided in Section 1.6 of the Supplementary Information. In terms of macroeconomic representation, our study considers a number of general equilibrium models where price-induced effects on GDP and productivity are computed (for example, GEM-E3, REMIND-MAgPIE, MESSAGE-MACRO and AIM-Hub). These models assume an exogenous reference path for GDP as the basis of which price-induced and path-dependent GDP losses are calculated. The models account for the macroeconomic path-dependency in terms of shifts in capital stocks, investments, saving and consumption patterns.

A common scenario design and modelling protocol was implemented by all models (Supplementary Information Section 2 on modelling protocol). For the mitigation scenarios, the models explored the full scenario space of cumulative CO2 emissions limits of <3,000 GtCO2 (2018–2100) in 100 GtCO2 increments (Supplementary Tables 2.1-1 and 2.2-2). We thus assess the lowest attainable budget for each model. In scenarios with no NNCE, sources and sinks across sectors and regions may balance each other out but total CO2 emissions are not allowed to become net negative. Mitigation of non-CO2 GHGs follows the same equivalent carbon price as for CO2 (driven by the cumulative CO2 emissions budget constraint). GHG mitigation on the land sector will hinge on appropriate policy designs that avoid competition over land for food or other basic ecosystem services, water resources and/or biodiversity54,55,56,57. To account for such possible trade-offs, the models in this study limit land-based mitigation and cap the GHG price effect on the agricultural sector to <US$200 per tCO2e. Some models include, in addition, explicit biodiversity protection constraints (MESSAGEix-GLOBIOM). Peak and decline of temperature due the reduction of non-CO2 emissions is between 0 and 0.14 °C across the models by 2100 (blue dots in Fig. 1b). In contrast to the CO2-induced temperature overshoot, the effect of non-CO2 on overshoot is relatively limited. The NPi (baseline) scenario broadly incorporates middle-of-the-road socioeconomic conditions based on the second marker baseline scenario from the Shared Socioeconomic Pathways (SSP2)4. It also assumes that climate, energy and land-use policies that are currently ratified are implemented (cut-off date 1 July 2019). The NDC scenario builds on the NPi and assumes that the NDCs (both unconditional and conditional NDC actions) as submitted by April 2020 are implemented by 2030. In addition, we have explored a sensitivity analysis with an update of the NDCs for big emitting countries as submitted in December 2020 (China, EU and Brazil) and find the implications for the emissions and the long-term results to be very small (see Supplementary Information Section 1.5 for a sensitivity analysis). For the NPi and NDC scenarios, a continuation of effort in the long term was assumed. This was implemented by extrapolating the ‘equivalent’ emissions reductions or carbon price in 2020/2030 (see Supplementary Information, Section 2.2 on NPi and NDC extrapolation methods). We have not considered the impact of the COVID-19 pandemic in a comprehensive way, effectively assuming a full recovery without substantial impact on long-term global emissions58. Sensitivity analysis based on selected scenarios indicate only a small impact on mitigation (Supplementary Information Section 2.1). The scenarios explored here, however, can inform governments that aim for ‘green’ recovery packages59, by illustrating the required pace and contribution of key mitigation sectors to reach net-zero CO2 emissions. The wide range of mitigation costs reflect parametric and structural differences across the models and their resulting marginal abatement cost curves. A classification of the models with respect to abatement costs is provided in ref. 60. Note that the marginal abatement costs increase rapidly when approaching the (model-specific) attainability frontier and thus reported carbon prices increase considerably (>>1,000 US$ per tCO2).

GHG emissions here always refer to the gases of the Kyoto basket (that is, CO2, CH4, N2O, HFCs, PFC and SF6, aggregated with 100-yr global warming potentials from the IPCC AR5.

The GHG emissions resulting from the different scenarios by the IAM models were fed into the probabilistic reduced-complexity carbon-cycle and climate model MAGICC for the estimation of global mean temperature projections consistent with the scenarios. MAGICC61,62 is used in a setup that captures the IPCC AR5 climate sensitivity uncertainty assessment61,63,64, as used in the IPCC SR1.5 (ref. 6). If not otherwise specified, the definition of the temperature goals follow the IPCC SR1.5, that is, limiting the exceedance probability to <0.34 for 2 °C and limiting the exceedance probability for 1.5 °C (with low overshoot) to <0.67 for the peak temperature and <0.34 for the year 2100. Through this methodology we assess the resulting global warming of different pathways and the corresponding peak warming that is associated with the cumulative emissions (budgets) of the scenarios.