Figure 1a indicates large aggregate global benefits from constraining warming by 2100 (with 66% probability) to 1.5 °C rather than 2 °C in terms of reduced exposure to drought and fluvial flooding, heat stress, disease and for crop yields. Error bars and ranges indicate the implications of using alternative global circulation model (GCM) patterns in downscaling. Avoided risks range in percentage terms from 10 to 44%, with largest benefits in percentage terms accruing for reduced exposure to fluvial flooding (44% reduction) and reduced exposure to drought and crop yields (~ 27% reduction). Benefits accruing in terms of cumulative land lost due to submergence and people at risk from coastal flooding tend to be smaller in percentage terms than other benefits owing to the time lag associated with the response of sea level rise to reductions in greenhouse gas emissions (known as the commitment to sea-level rise). Our projections of absolute additional levels of risk indicators globally at 1.5 °C and 2 °C warming relative to the observed baseline are given in Table S1.
Globally, water scarcity decreases slightly with increased climate change; however, this should not be interpreted as a benefit, since there are regional increases and decreases, and where there is reduced water scarcity, there is often an associated increase in runoff and consequently flood risk. One risk metric relating to the prevalence of malaria indicates a benefit from climate change in parts of Africa and South America: Here, a drying climate reduces the risk of infection in many areas, and limiting global warming may prevent this net benefit being realised; however, overall globally, population exposure to malaria and dengue is 10% lower if warming is constrained to 1.5 °C rather than 2 °C.
Benefits within sectors accruing from constraining warming to 1.5 °C rather than 3.66 °C are projected to be notably larger (32 to + 85%) than those accruing from constraining warming to 2 °C rather than 3.66 °C (26 to + 74%) (Figure 1b,c). Benefits of limiting warming to 1.5 °C rather than 3.66 °C reach 87% for population exposure to fluvial flooding and over 70% for population exposure to drought and reductions in crop yields. These numbers are similar to the estimates emerging from our parallel study (Warren et al. 2021) which also uses the same climate change scenarios and finds that economic damages at 1.5 °C warming are 92% lower than mean losses of 3.67% of GDP (range 0.64–10.77%) associated with global warming of 3.66 °C.
The fact that our pathways to 1.5/2 °C warming achieve this with a probability of 66% rather than 50% is an advantage, since it provides greater confidence that limits might be attained. Avoided risks due to following pathways that constrain warming to 1.5 °C rather than 2 °C by 2100, with 66% probability, are also projected to accrue already in the 2050s (Table S3) for aggregate economic damages, exposure to dengue infection and crop yields. Crop specific projections for crop yields are given in Table S4.
Figure 2 maps the differences between indicators of risk in 2100 for the pathway constraining warming to 1.5 °C rather than 2 °C, each with 66% probability, for each risk indicator examined. When aggregated regionally, benefits, expressed as avoided increases in risk indicators, or as avoided percentage changes in risk indicators, vary significantly (Figure S5a, b, Table S2). Figure S4b and c map the regional increases in risk indicators resulting from the combination of socioeconomic changes and climate changes by the year 2100, at 1.5 °C or 2 °C warming respectively for a subset of risk indicators, excluding those for which indicators are by definition close to zero in the baseline.
Figure S4a provides absolute total global projected risk indicators at each warming level in 2100, showing that the largest absolute benefits globally in terms of quantified reduction of risk indicators arising from constraining warming by 2100 (with 66% probability) to 1.5 °C rather than 2 °C are in the avoidance of population exposure to drought.
Figure S4d and e map regionally the percentage and absolute levels of risk indicators at 1.5 °C and 2 °C warming. While considerable spatial variation is apparent, there remains a predominant trend that risks tend to increase with warming.
Figure S5, S6 and S7 detail and compare regionally percentage and absolute differences between levels of risk indicators at 1.5 °C vs. 3.66 °C warming and also between 2 °C vs. 3.66 °C.
Risk analysis carried out in this study using sWBGT finds hundreds of millions of people to be additionally affected by heat stress at each (successively higher) warming level. This result is consistent in magnitude with other recent studies, such as Matthews et al. (2017) who project 350 million more megacity region inhabitants to be exposed to deadly heat by 2050 for an end of century warming level of 1.5 °C. Andrews et al. (2018) also project hundreds of millions of people to be exposed to extreme heat for warming levels of 1.5 °C and above. As has been shown in other multi-sectoral impacts, studies which include humid heat metrics (e.g. Byers et al. 2018) projected heat exposure is most pronounced in the tropics, and as such, we identify benefits of reduced exposure associated with limiting warming in low-latitude regions. Our study focuses on applying targeted climate scenario data to calculate global (combined urban and rural) population heat stress using the sWBGT metric. While sWBGT can produce an overestimate of heat exposure risk during cloudy or windy conditions and vice versa, Willett and Sherwood (2012) argue that changes in solar radiation and wind speed are unlikely to impact significantly on global patterns. Population impacts of exposure to heat stress will depend on the activity of the person concerned and the choices that they make.
We estimate a continuous increase in global drought risk and find hundreds of millions of people to be additionally affected by drought at each (successively higher) warming level, which is well aligned with previous research (Prudhomme et al. 2014; Smirnov et al. 2016; Lehner et al. 2017; Arnell et al. 2018; Naumann et al. 2018; Liu et al. 2018). A direct comparison in terms of affected people and impact avoided with any of these studies, however, is complicated due to the use of different drought indices, metrics and population data. For example, Naumann et al. (2018) also use the SPEI-12, but they use event length and magnitude as a measure of impact, while Smironv et al. (2016) use the SPEI-24 but calculate the number of people affected. This shows that more studies are needed that use sets of indices, metrics and data that intersect with already existing ones to enable a better comparison. Other studies are based on SPI which is easier to calculate, but by definition excludes evapotranspiration meaning that the drought estimation is conservative as it fails to allow for the important role of temperature rise in contributing to potential evapotranspiration.
Here, we focus on a comparison with Arnell et al. (2018) because they use the same approach to calculate the number of people affected from drought frequency and the same socioeconomic scenario (SSP2). Their analysis is, however, based on the standardised runoff index (SRI). Arnell et al. (2018) report a median 39% (36–51% for the 10–90% range) impact avoided between 1.5 and 2.0 °C, with 630–1300 million people exposed to drought at 1.5 °C and 710–1600 million people exposed at 2.0 °C (Figure 1). Our mean values of 1406 million and 1752 million, respectively, are slightly above the upper end of their ranges, while our mean value of 26% impact avoided is below theirs. This might be due to the choice of SPEI compared to SRI. The SPEI might capture anomalies in which the projected proportional increase in precipitation is smaller than the proportional increase in PET. Such conditions are linked to increasing drought conditions (Naumann et al. 2018), and capturing those additionally could explain the higher numbers of people affected and the smaller proportion of avoided impacts. In a comparison of different drought metrics (including SRI and SPEI) in a single basin, SPEI-12 indicated a more severe drought than SRI-12 for more of the years 1970–2017 than the reverse (Dikici 2020). SRI is usually used as a measure of hydrological drought, while SPEI (and others) are used for agricultural, meteorological and hydrological drought. Thus, SPEI captures the water that may be captured in ways other than run-off and in food stocks. Differences between values of these projected metrics would therefore be expected.
Our regional projections indicate a mixed picture, similar to the results obtained by Arnell et al. (2018). Overall, we identify Africa, India and the Middle East as hotspots for increased exposure to drought. While the number of people exposed to drought conditions is expected to increase in most of Southern America, Europe and East Asia, we also find that there are regions which could see lower numbers of people affected in the future, particularly in Russia, China, Indonesia and parts of South America (Figure 3). When limiting global warming to 1.5 °C rather than 2.0 °C, most inhabited areas in the world would benefit, apart from a few isolated patches in Russia, Indonesia and South America (Figure 2), which is well in line with the analysis of Arnell et al. (2018).
Annual runoff increases considerably with global warming in large parts of Russia, Canada, India, north-east China, northern and central Europe, West Africa and some small parts in central and eastern Africa. Runoff drops as the warming increases from 1.5 to 2.0 °C in the southern USA, central region of South America, large parts of central Africa, south-east China and the Mediterranean. We focus on a comparison with Hirabayashi et al. (2013) which utilises the same CaMa-Model and finds that on average, the global flood exposure to (present-day) Q100 floods in the RCP8.5 scenario for 2071-2100 is projected to lie between 37 and 163 million. Similarly, we projected global flood exposure to Q100 floods in 2086–2115 with 3.66 °C warming obtaining a range of 81 to 195 million people. The findings are of the same order of magnitude and overlap in range, but are not identical as is to be expected given differences in the selection of the time period, population data sets and regional climate change projections used.
Population exposure to water scarcity is most evident in western India and northern region of West Africa. It is worth noting that the population in the northern region of West Africa is project to experience both fluvial flooding and water scarcity, less so at 2 °C than at 1.5 °C global warming. However, like Gosling and Arnell (2016), we also found relatively small differences in water scarcity between the 1.5 and 2 °C warming scenarios when compared with differences arising from uncertainties in regional climate change projection.
The additional exposure to fluvial flooding risk (Figure 2) is mostly evident in West Africa, India and parts of central East Africa, aligning with the identification of hotspots with multi-sector risk in West Africa and South Asia. Our study also shows that large areas of inhabitants in sub-Saharan Africa and southern Asia would be exposed to Q100 floods at the higher degree warming. Thus, our findings agree mostly with previous studies such as Piontek et al. (2014), Gosling and Arnell (2016) and Byers et al. (2018) who projected that the poor and vulnerable populations in Africa and southern Asia would be disproportionately impacted by multi-sectors impacts.
The constantly reducing crop yields we obtain under increasing global temperature align well with results in the existing literature using both process-based models and statistical models (see SM), although compared to these studies, our results appear to be conservative. Projections for regional changes in crop yields are consistent with a previous study (Schleussner et al. 2016) identifying Africa, SE Asia, and C&S America as hotspots for projected declines in yield and indeed for projected avoided risks if warming is limited to 1.5 °C rather than 2 °C (Figure 2). Equally, they are well aligned with the hotspots identified by Byers et al. (2018) and Piontek et al. (2014). Similar to Arnell et al. (2016), our results indicate that reductions of maize yield in all regions and soybean yields are projected to potentially increase in Europe, North America and Australasia but to decline in other regions. In case of wheat, we project declines in all regions, while Arnell et al. (2016) obtained mixed results. We suspect that this is due to the different types of wheat that were analysed. For rice, our projections indicate strong losses in Africa and South-East Asia but increasing yields in Europe and Australasia. Limiting global warming to 1.5 °C rather than 2 °C would provide benefits for most regions across the globe, particularly in the Americas, Europe and Africa (Figure 2), which is also in line with the findings of Arnell et al. (2016). Overall, our results suggest an inequality in risk of crop yield loss between the Northern and Southern hemispheres and especially tropical and non-tropical regions. The main limitation of the models used here is that they are based on unevenly spaced national data and that the area harvested was assumed to remain constant so that potential future land use change is not accounted for.
Our estimates of future climate-driven malaria risk are in line with previous research indicating that such risk is confined to specific regions (Piontek et al. 2014; Caminade et al. 2014; Ryan et al. 2020). Our results suggest that climate change is likely to increase the risk of dengue transmission across vast areas in Latin America in agreement with previous studies (Colón-González et al. 2018; Messina et al. 2019; Watts et al. 2019).
A limited number of studies have analysed the impacts or costs of sea-level rise specifically at 1.5 °C and 2.0 °C at a global scale, including Schleussner et al. (2016), Brown et al. (2018), Nicholls et al. (2018), Rasmussen et al. (2018) and Jevrejeva et al. (2018), whilst others, such as Hinkel et al. (2014), Vousdoukas et al. (2018) and Yokoki et al. (2018), have analysed impacts following the Representative Concentration Pathway scenarios. Whilst each study has produced different sea-level rise scenarios, the exposure and impact metrics produced are of a similar order of magnitude, taking account differences such as adaptation. For example, Nicholls et al. (2018) projected 33–117 million people/year at risk from flooding globally for the 1.5 °C scenario (0.24–0.54 m of sea-level rise) and 42–132 million people/year at risk from flooding globally for the 2.0 °C scenario (0.31–0.65 m of sea-level rise) in 2100, assuming no additional adaptation from the baseline (1995). In comparison, our results, indicate 41–88 million people/year at risk from flooding globally for the 1.5 °C scenario (0.24–0.56 m of sea-level rise) and 45–95 million people/year at risk from flooding globally for the 2.0 °C scenario (0.27–0.64 m of sea-level rise) in 2100, assuming no additional adaptation from the baseline (1995). Our projections of human exposure coastal flood risk are similar in magnitude to earlier studies including Nicholls et al. (2018), with differences emerging from the use of a mean across Shared Socioeconomic Pathways (SSP) 1–5, in that study as compared with SSP2 here. No comparable results are available for land loss due to submergence. Regionally, the greatest number of people at risk from coastal flooding, who stand to benefit the most from limiting warming to 1.5 °C rather than 2 °C, are in east and south Asia, especially China and India (Figure 2, SM4b,c). Additionally, land potential subject to submergence is projected in lower lying northern latitudes including North America and many of the world’s delta regions.
Our projected SLR is higher than some other projections (e.g. Goodwin et al. 2018; Geiges et al. 2020), but still within a plausible range. The difference here between the < 1.5 °C and < 2.0 °C scenarios is 0.06 m in 2100. This is less than the best estimate that Hoegh-Guldberg et al. (2018) conclude from a range of studies, of approximately 0.1 m difference in sea-level rise between the 1.5 and 2.0 °C scenarios, but within their uncertainty range. The sea-level rise projections here do not include the non-linear Antarctic ice sheet dynamics effects on sea-level rise proposed by DeConto and Pollard (DeConto and Pollard 2016; Edwards et al. 2019) that are projected to accelerate ice melt and sea-level rise in the latter half of the twenty-first century for high-warming scenarios (Hope 2013). Our projected estimates of avoided sea-level rise are thus lower than in Schleussner et al. (2016) due to a linear treatment of ice-sheet dynamics in our study, and also the effect of a small mid-century overshooting of 1.5 °C by of 0.1 °C mid-century in our mitigation scenario which is not present in Schleussner et al. (2016) (other models used in this study do not reflect the dynamics of temperature change during the twenty-first century, and hence the implications of this small overshoot do not affect our other estimates of risk). There is evidence that this non-linear ice sheet dynamics effect is not required to reproduce past sea level observations and reconstructions (Edwards et al. 2019), and if it does occur, it is not thought to significantly affect sea-level rise for warming pathways around 1.5 °C (Hope 2013). However, potential non-linear ice sheet dynamics effects imply that our projections of sea-level rise for higher warming scenarios (above 1.5 °C) may be conservative. The simulations presented here assume that defences have not been upgraded as sea-levels rise. In reality, defence standards are likely to increase following common practice. The results for land loss due to submergence and the number of people flooded assuming upgrades with adaptation are shown in the Supplementary Results (Tables S5, S6, S7, Figures S8, S9).
For further discussion of sector specific projections and further comparison with other literature, see SM.
In summary, using mean estimates taking into account a range of alternative regional climate projections, our process-based models estimate that constraining warming to 1.5 °C rather than 2 °C would variously avoid 10–44% of the increases in indicators of climate change risk that would otherwise accrue in various disparate categories by 2100. A parallel analysis of global aggregate economic damages arising from these same climate change scenarios with PAGE09 (Warren et. al. 2021) indicates avoided economic impacts of 20% in the global aggregate from constraining warming to 1.5 °C rather than 2 °C, corresponding to a net present value of damages of 39 (range 13–108) US $trillion rather than 61 (range 15–140) US $trillion. Avoided damages (Warren et al. 2021) reach 87% (74–91%) from constraining to 2 °C rather than 3.66 °C, when damages reach 484 (60–1590) US $trillion, and 90% (77–93%) when warming is constrained to 1.5 °C rather than 3.66 °C. This implicitly assumes that the results from the scenarios are perfectly correlated, which is a reasonable assumption as high values for the transient climate response (TCR) and impact function parameters will give high global risks in all scenarios, and vice versa. Thus, estimates of the percentage of risks avoided by limiting global warming to lower, rather than higher, levels are comparable if an independent economic model is used to simulate aggregate climate change damages, with the mean estimate across assessed risk indicators of 20% risks avoided falling in the range projected from the physically based models.
Benefits vary across regions and between sectors, being greatest for fluvial flooding, with 38–55% risks avoided depending on regional climate projections used. However, some regions are exposed to a range of risks simultaneously and may become ‘hotspots’ of climate change risk.
Hotspot analysis (Figure
Early studies explored annual mean (Giorgi 2006) climate change hotspots, but focused entirely on the hazard component of risks related to projected annual average changes in precipitation and temperature, and identified the Mediterranean, the northern hemisphere high-latitude regions and Central America as the most prominent hotspots. Giorgi et al. (2006) went on to include socioeconomic factors, and also sea-level rise, but did not include projected increases in population: This approach highlighted China and Bangladesh as the most threatened nations on a gross population basis, followed by India, Russia, Brazil, and the USA. Later, Diffenbaugh and Giorgi (2012) conducted a statistical analysis in multi-dimensional climate space, including changes in seasonality but excluding sea-level rise or socioeconomic factors, but based on new CMIP5 projections, this identified the Amazon, the Sahel and tropical West Africa, Indonesia and the Tibetan Plateau as persistent regional climate change hotspots throughout the twenty-first century in a range of forcing pathways. In addition, areas of southern Africa, the Mediterranean, the Arctic, and Central America/western North America at higher levels of forcing.
Our regional analysis of the regional distribution of additional risks arising from the combination of climate change and socioeconomic change by 2100 (Figure S4b,c) and the associated hotspots map highlights India, West Africa and North America as regions at particular risk, at both 1.5 °C (Figure 3) and 2 °C (not shown, but almost identical): The overall distribution of risk is very similar at both warming levels, but as previously mentioned, there is a large percentage reduction in risks levels at 1.5 °C compared with 2 °C. This is broadly consistent with earlier work on hotspots that included aspects of exposure and vulnerability, most notably the distribution of human populations, in that it generally indicates that South Asia and Africa become hotspots of climate change risk, especially for water related indicators (Arnell et al. 2018; Byers et al. 2018). This is despite the fact that Byers et al. (2018) use different sets of climate change risk indicators (including the use of different thresholds in cases where indicators are threshold-based); cover different sectors (our study includes coastal flooding and excludes nitrate leaching); aggregate the indicators using an expert scoring procedure as compared with uniform weighting in our study; and are based on the outputs of different risk projection models/data sets and the use of different socioeconomic scenarios.
There are some notable differences however between the various studies. Byers et al. (2018) project larger risks in China than in our own study, probably because of the inclusion of a greater number of land-based climate change risks than in our own study. Similarly, Arnell et al. (2016) estimate heat events and hydrological drought irrespective of human exposure, identifying Australasia as an area at particular risk, which does not emerge from our study. This is because our study focused heavily on population exposure to hazards, and hence areas experiencing large increases in drought or water scarcity but with low human population do not emerge strongly in our analysis.
Changes in risk associated with different levels of warming will also depend strongly on the socioeconomic pathway (Arnell et al. 2018; Byers et al. 2018), which in this study was set to a ‘middle of the road’ scenario in which population rises to 9.4 billion in 2070 and declines thereafter. For example, for larger (smaller) increases in population, the absolute levels of populations exposed to the various risks included in this study would be expected to be larger (smaller), and the benefits associated with limiting warming correspondingly larger (smaller) in absolute terms, although not necessarily in percentage terms (e.g. Supplementary Material of Arnell et al. 2018). For risk-related metrics which are associated with a threshold (such as the heat stress, water stress and flooding metrics explored here), levels of avoided exposure will depend on the threshold selected. Another important factor is the uncertainty between exposure/impact models. Such uncertainties can be large, as currently being explored by the ISIMIP model comparison exercise (Warszawski et al. 2014). Sensitivity studies exploring these issues would be of interest, but were beyond the scope of this study.
In general, there is good agreement with other studies, although our projections of climate change impacts on crop yields and exposure to heat stress may be relatively conservative. Projected risks to biodiversity which were beyond the scope of this study will also be important and will interact, via loss of ecosystem services, with the projected risks to human systems estimated here (Fischlin et al. 2007; Warren et al. 2013, 2018).