Climate change research and action must look beyond 2100

Anthropogenic activity is changing Earth's climate and ecosystems in ways that are potentially dangerous and disruptive to humans. Greenhouse gas concentrations in the atmosphere continue to rise, ensuring that these changes will be felt for centuries beyond 2100, the current benchmark for projection. Estimating the effects of past, current, and potential future emissions to only 2100 is therefore short‐sighted. Critical problems for food production and climate‐forced human migration are projected to arise well before 2100, raising questions regarding the habitability of some regions of the Earth after the turn of the century. To highlight the need for more distant horizon scanning, we model climate change to 2500 under a suite of emission scenarios and quantify associated projections of crop viability and heat stress. Together, our projections show global climate impacts increase significantly after 2100 without rapid mitigation. As a result, we argue that projections of climate and its effects on human well‐being and associated governance and policy must be framed beyond 2100.


| INTRODUC TI ON
When climate models were first used in the 1980s and 1990s, the year 2100 was seen as a suitably distant horizon for climate projections. However, this benchmark is now just one human lifespan away, and opportunities to readily curb emissions in line with the Paris Agreement are dwindling (IPCC, 2018). Anthropogenic activity is already altering atmospheric carbon dioxide concentrations at a rate that generally exceeds those known in Earth archives (Burke et al., 2018;Kemp et al., 2015;Zeebe et al., 2016), generating changes deleterious for humans and ecosystems (Ford et al., 2019;IPBES, 2019;Pascale et al., 2020). Obtaining insights into anthropogenic effects on the Earth system that support human existence is therefore critical for designing governance and policy structures that can mitigate these effects, which are predicted to continue well beyond 2100 (Riahi et al., 2017).
Since 1990, the three Working Groups (WG) of the Intergovernmental Panel on Climate Change (IPCC) have produced periodic Assessment Reports, including the 2021-2022 publication of the sixth incarnation.
Central to future climate projections are socioeconomic scenarios, including estimates of future fossil fuel consumption, land-use change, industrial activity, and associated greenhouse gas and short-lived pollutant emissions (Riahi et al., 2017).
The core scenarios prepared for IPCC's Fifth Assessment Report (AR5) were termed Representative Concentration Pathways (RCPs) and covered four emissions trajectories. RCPs ranged from a global scale reduction on fossil fuel reliance and achievement of net-negative CO 2 later this century (RCP 2.6), to a high-emission scenario that included substantial new investments in fossil fuels and lack of global climate policy and governance (RCP8.5) . The newer Shared Socio-Economic Pathways (SSPs) include five development 'storylines' that capture emission scenarios and pair them with socioeconomic scenarios (O'Neill et al., 2020;Pedde et al., 2020;Riahi et al., 2017). The primary time horizon for both RCP and SSP scenarios is 2100.
However, it is now clear that without deep and rapid reductions in greenhouse gas emissions, climate change will continue for centuries into the future. Efforts to extend projections beyond 2100 exist but are limited. For example, emission and greenhouse gas concentration projections to 2300 are provided for each RCP scenario in CMIP5, which were further extended to 2500 by Meinhausen et al. (2011). Similar long-term projections exist to 2500 for Shared Socioeconomic Pathways (SSPs) in CMIP6 (Meinshausen et al., 2020). However, no complex climate model results from CMIP5 or CMIP6 are available beyond 2300.
Although several CMIP5 models ran projections to 2300, at present very few CMIP6 models have done so, requiring the IPCC's Sixth Assessment Report to base longer-term projections primarily on simpler models (Lee et al., 2021). Indeed, many studies that focus on time horizons beyond 2100 have used reduced complexity or intermediate complexity Earth System models (Goodwin et al., 2018;Palmer et al., 2020;Zickfeld et al., 2013)  modelling past 2100 is currently not focused on projecting aspects of ecosystem services of importance to human well-being, such as useable land not inundated by sea-level rise (Clark et al., 2016), habitable temperatures (Schwingshackl et al., 2021), agricultural change (Müller et al., 2021), and availability of freshwater (John et al., 2021).
In short, although 50 years have passed since the initial climate projections (Forster, 2017), our time horizon for coupled climate projections remains primarily at 2100 (though see, e.g. Goodwin et al., 2018;Palmer et al., 2020;Zickfeld et al., 2013). We therefore argue that climate and social projections beyond 2100 need to become more routine (Pearson, 2020;van Renssen, 2019). To make our case, we present climate projections modelled to 2500 under three emission scenarios representing strong, moderate, and weak global climate policy (RCP2.6, RCP4.5, and RCP6.0). We explore crop viability and heat stress after 2100 to highlight the necessity of socioeconomic planning on timescales beyond the next 80 years and propose a social-governance approach to account for longer-term climate dynamics. Our modelling exercises provide an initial framework and baseline for the assessment of longer-term anthropogenic effects on climate and Earth systems and highlight the need for further work in this area.

| CLIMATE PROJEC TI ON S AND VEG E TATION B E YOND 210 0
Results of our preliminary climate projections (SI Methods) drawn from the HadCM3 atmosphere-ocean coupled climate model (Gordon et al., 2000) combined with the TRIFFID dynamic land surface model (Cox, 2001) clearly demonstrate the need for quantification of climate change effects past 2100. Global mean temperature, for example, continues to increase after 2100 under all but the lowemission RCP2.6 scenario. Under the moderate-high RCP6.0 emissions scenario (a realistic scenario with low mitigation; Hausfather & Peters, 2020), global mean warming is 2.2°C above present-day levels by 2100 ( Figure 1, Figure S1) but continues to rise to 3.6°C in 2200 and 4.6°C in 2500. Warming is unequally distributed, with greater warming over the land surface and in polar regions ( Figure 1).
Our projections compare well to previous assessments of warming past 2100. RCP projections are within the range of those from available CMIP5 models to 2300 (crosses in Figure 1). The recent IPCC Sixth Assessment Report, drawing evidence mostly from the SSP1-2.6 (a similar scenario to RCP2.6), which easily encompass our projections (Lee et al., 2021). Note that IPCC Sixth Assessment Report projections are taken from Table 4.9 in Lee et al. (2021), which are relative to 1850-1900, and 1.0°C is subtracted from these values to represent warming from 1850-1900 to 2000-2019 (Gulev et al., 2021).
The higher emission scenarios (RCP4.5 and 6.0) result in major restructuring of the world's biomes by 2500. For example, HadCM3 projects a severe dieback of Amazon rainforest under RCP6.0 and RCP4.5 by 2500 (Figure 2), congruent with previous research using the same model under a high-emission scenario . Conversely, the low-emission scenario (RCP2.6) reaches peak warming this century ( Figure 1) with stabilization of global mean temperature only 0.5°C above the 2000-2019 mean and limited long-term shifts in global vegetation ( Figure S2). Sea level, however, continues to rise long after warming has stabilized (Palmer et al., 2018(Palmer et al., , 2020, even in the RCP2.6 scenario, due to slow continued mixing of heat into the deep ocean (Oppenheimer et al., 2019) ( Figure 1; Table 1). The long-term impacts of 21st-century emissions are therefore likely to be felt for centuries to come, continuing even after greenhouse gas concentrations have reached equilibrium (2150 for RCP4.5 and RCP6.0).

| HE AT S TRE SS AND H UMAN WELL-B EING B E YOND 210 0
Heat stress can be fatal to humans when wet-bulb temperatures exceed 35°C for 6 or more hours (Buzan & Huber, 2020;Sherwood & Huber, 2010). Physiologically fit humans can tolerate higher dry-air temperatures, but such temperatures can still lead to high mortalities (Diniz et al., 2020;Varghese et al., 2020). These conditions also cause damage to critical infrastructure on which humans rely, such as electricity (Burillo et al., 2019), transportation (Villalba Sanchis et al., 2020), and agriculture (Anderson et al., 2020;Mehrabi, 2020).
We estimate changes in heat stress to 2500 using the Universal Thermal Climate Index (UTCI) (Błażejczyk et al., 2013;Jendritzky et al., 2012). UTCI is a measure of heat stress encompassing both TA B L E 1 Calculated contribution to sea-level rise (meters) from deep ocean heat mixing in 2100, 2200, and 2500 under three RCP scenarios  Figure S3). The regions that currently experience periods of very strong heat stress today tend to be deserts, but also include the

| AG RICULTUR AL CHALLENG E S AF TER 210 0
The effects of climate on agriculture are a major research area covering crop adaptation, migration, and food production (Anderson et al., 2020;Mehrabi, 2020;Stringer et al., 2020). Climate-driven crop migration and yield reductions have been observed already (Moore, 2020;Sloat et al., 2020;Zhang et al., 2017) and projected for the future (Ceglar et al., 2019;King et al., 2018), but are not typically examined beyond 2100 (Tigchelaar et al., 2018

| REG I ONAL C A S E S TUD IE S
The changes we have projected are likely to have profound effects on natural vegetation and on human society by altering the distribution of tolerable environments and by changing the feasibility of agriculture.
To explore the potential effects of these changes on human well-being, we highlight site-specific projections for three regions ( Figure S13) of global importance under RCP6.0: the North American 'breadbasket', the Amazon Basin carbon sink, and the densely populated Indian subcontinent. We use our results to inform artistic interpretations of these regions to highlight the profound changes they may face under a plausible medium-to high-emission scenario (RCP6.0) (Hausfather & Peters, 2020) after 2100 (Box 1). For results from additional RCP scenarios, see Figures S14 and S15.

| North American Midwest
The interior plains of the American 'Midwest' are a global breadbasket. Today, the Midwest is characterized by cold winters and warm summers (Angel et al., 2018). Under RCP6.0, mean summer temperatures increase from 28°C today to 33°C by 2100 and 36°C by 2500 ( Figure 2

| Amazon Basin
The Amazon Basin is home to one-third of Earth's known species (Heckenberger et al., 2007) and currently serves as a carbon sink for roughly 7% of anthropogenic CO 2 emissions (Brienen et al., 2015;Friedlingstein et al., 2019) ( Figure S13). The region is also culturally and linguistically diverse, home to more than 350 indigenous languages (Aikhenvald, 2015). Our modelling suggests that rising temperatures and disrupted rainfall patterns will render the Amazon Basin unsuitable for tropical rainforests by 2500 (Figure 2, Box 1), with consequences for the global carbon cycle, biodiversity, and cultural diversity. Initial declines in forest cover in the model lead to a positive feedback of reduced transpiration, further reduced rainfall, and further forest retreat. The HadCM3 climate model exhibits this feedback more than most climate models, especially in the Amazon Basin (Poulter et al., 2010;Sitch et al., 2008), but still has a plausible sensitivity (Cowling et al., 2004). The HadCM3 model projects a limited retreat of the Amazon rainforest by 2100, but in the following centuries, forest dieback feedback enhances forest loss, and high temperatures and low precipitation ( Figure S16) conspire to produce a barren environment in most of the Amazon Basin (Boulton et al., 2017). Amazonian forest cover declines from 71% in the present day to 63% in 2100, 42% in 2200, and 15% in 2500. The newer HadGEM2-ES model also shows Amazon dieback (though less severe), with freely evolving vegetation when run to 2300 CE under a high-emission scenario (Drijfhout et al., 2015).

| Indian subcontinent
The Indian subcontinent is one of the most populous regions on Earth ( Figure S13). The region already experiences extreme climatic conditions, with thousands of heat-stress-related deaths recorded between 2013 and 2015 alone (Mazdiyasni et al., 2017). Our modelling suggests that mean summer monthly temperatures could increase 2°C by 2100 and 4°C by 2500, suggesting the Indian BOX 1 Artistic comparison of potential changes in regional landscapes and human activity between 2020 and 2500 under RCP6.0. Three image pairs illustrate the potential scope of regional changes under RCP6.0 (Figure 2). Although technology in 2500 is essentially unknowable, we limited technological advancement for the purposes of making comparisons between 2020 and 2500. Indian subcontinent (e) 2020 and (f) 2500 under RCP6.0 scenario. A characterization of India in the present day and in 2500. We illustrate a conservative view of potential human adaptations based on similar technology today and from science fiction (Elson, 2016;Smith, 2008). Extreme heat may require protective personal clothing for outdoor activity -in this hypothetical case, a sealed helmet and a suit conducting water and coolants around the body. Outdoor agriculture in 2500 may be managed by automated drone-machinery.

US
subcontinent will experience even higher heat stress than that projected for 2100 (Im et al., 2017)

| SUGG E S TIONS FOR GOVERNAN CE FOR LONG TIME SC ALE S
Human activity has already caused warming of ~1°C above average global pre-industrial levels (IPCC, 2018). Global mean temperatures will continue to increase until the point at which CO 2 emissions reach net zero (Allen et al., 2009;Matthews et al., 2009;Rogelj et al., 2019). Return to a pre-industrial climate is not possible without either removal of excess greenhouse gases added to the atmosphere or a sustained geoengineering programme (Carton et al., 2020). Such efforts appear unlikely given failures of governance around negative emissions technologies (Carton et al., 2020;McLaren & Markusson, 2020;Stevenson, 2021). As such, we argue that a longer-term, post-2100 perspective is critical for assessing the scope of climate change on Earth systems and on human well-being (van der Geest & Warner, 2020).
Our climate, heat stress, and agricultural projections parallel work that suggests climate increasingly drives global and regional human dispersal (Burke et al., 2021;Chen & Caldeira, 2020), especially from the heat-stressed tropics where habitability and crop suitability may be reduced. The scale of change we project over the coming centuries, especially under RCPs 4.5 and 6.0, will therefore necessitate more cooperative and collaborative approaches to global mobility to accommodate substantial human movement from less habitable regions (Adger et al., 2020). Meeting this challenge will require a major evolution in international relations away from national security and competition toward cooperation and integration (Beardsworth, 2020).
Our projections for crop viability also portend declines in ecosystem services after 2100. Even before 2100, projections of climate change suggest low-income (often tropical) countries are vulnerable to reduced crop suitability, and high-income countries face challenges with inward migration and converting climatically suitable land to agriculture (Chaplin-Kramer et al., 2019;Poeplau et al., 2019;Zabel et al., 2014). Such shifts also bring risks of soil carbon release, incursion into biodiversity hotspots, and threats to water security (Hannah et al., 2020;Poeplau et al., 2019). Over the long term, proposed strategies for food security, even those considered transformative such as meatless diets and urban farming (Fraser & Campbell, 2019;Stringer et al., 2020), may be insufficient if present agricultural areas fall out of production and technological advancements or landscape management (e.g. agroecology) prove unworkable at scale. The structure and function of the global food system will require reimagining, potentially via changes to property rights, use, and ownership (Healy et al., 2020) that mirror changes in productive climates, landscapes, populations, and technologies (Aguiar et al., 2020;Stringer et al., 2020).
The scope of projected future changes examined here will likely require long-term and adaptive integration of diverse cultural, knowledge, and governance structures that are global in scope and approach Fazey et al., 2020). For example, new knowledge-action synthesis efforts (Marien, 2007;Pedde et al., 2020) could inform governance institutions. These centres could be new organizations, such as long-range foresight groups (Burrows et al., 2018) or 'Ministries for the Future' (Robinson, 2020), that are independent of or tied to existing governance institutions (e.g. United Nations) and networks of local governments (e.g. 100 Resilient Cities; Papin, 2019).
These cross-cultural, trans-national organizations must evolve to keep ahead of observed and anticipated human migration, food production, disasters, and other climate and ecological challenges (Cleaver & Whaley, 2018;Schultz et al., 2015). Practically, this can mean using a rolling-baseline, Russian-doll approach to scenarios and decision-making, embedding subjectively short-term (0-50 years) local or regional assessments and actions inside medium (50-100 years) and longer-term global perspectives (>100 years) based on observed and modelled impacts and thresholds (O'Neill et al., 2020). The medium-and long-term approaches aim to anticipate, develop, and implement structures and technologies for Earth system governance that permits fair access to and allocation of resources to the global population under different impact trajectories (Biermann & Kim, 2020;Kalfagianni & Meisch, 2020). This nested anticipatory approach (Boyd et al., 2015;Muiderman et al., 2020) to adaptive governance would accommodate rapid events, such as floods and droughts, within slower-moving changes to temperature, sea-level, crops, and biodiversity. Projections of climate and Earth system changes beyond 2100 inform these longer-term approaches, helping to ensure changes to ecosystems and their resources are adequately managed to sustain human survival (Bennett, 2017;Burke et al., 2021).

| CON CLUS IONS
The year 2100 is one human lifespan away, and the window to readily curb emissions in line with the Paris Agreement is rapidly closing (Leach et al., 2018). Our projections past 2100 indicate that without rapid and significant reductions in greenhouse gas emissions, large areas of the Earth will change in ways that reduce their capacity to support large-scale human occupation. The long-term effects of 21st century warming will be felt for centuries to come, even if emissions are limited in the future (Figure 1). Efforts at mitigation (Kyoto; Paris Agreement; UNFCCC, 2015) may have slowed the growth of greenhouse gases in the atmosphere, but commitments still fall massively short of the 1.5-2.0°C goal (Roelfsema et al., 2020). Even if commitments are met, projections still show that we must contend with heat waves and other extreme events of unparalleled intensity and frequency (Dosio et al., 2018;Schleussner et al., 2016). We therefore need to understand and model these changes beyond the next 80 years. These longer-term projections are critical to preparing the way for a peaceful and habitable Earth in the coming decades and centuries.
Our projections and associated approaches to adaptation governance represent an initial attempt and have considerable uncertainty given their extended time horizon. These efforts are meant to highlight the need for more sophisticated climate and Earth system modelling beyond 2100, including a focus on aspects of ecosystem goods and services not considered here. Our work thus provides a framework and baseline for the assessment of longer-term anthropogenic effects on climate and Earth systems, and highlights the critical need for further work in this area.

ACK N OWLED G EM ENTS
The climate modelling was undertaken on ARC3, part of the High Performance Computing facilities at the University of Leeds, UK. The authors thank Deborah Sporton, Tom J. Webb, Claudia Di Napoli, David R. Williams, and the Past Global Environments (PAGES) network for their feedback and support at various stages of this work.
We thank two anonymous reviewers for help that improved our contribution.

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
All climate model outputs and associated code will be made available upon acceptance in open-access archives.