全球变化知识资源中心

The y axis represents the proportion of provinces across China experiencing a given ΔT_ρ value.

Disease- and province-specific baseline rates were propagated forward using the above projected temperature deviations for 2020 and 2030, disease-specific temperature response functions, and the WSH storylines in Table 1. The burden of WSH-attributable disease in all storylines was dominated by diarrhoeal diseases (Supplementary Table 4), similar to the results of our previous work⁶. The reference storyline 0.2, in which there is no climate change, accounted for projected urbanization, population growth and changes in population age distribution. Assuming a linear water and sanitation development path, this reference storyline projects a nationwide decrease in the burden of diarrhoeal diseases (in DALYs) of 51.3% between 2008 and 2020 and 63.7% between 2008 and 2030. On the same linear development path under RCP 2.6 (storyline 1.2), the decrease is 48.4% between 2008 and 2020, and 59.5% between 2008 and 2030. For RCP 8.5 (storyline 4.2), the decrease slows yet more, to 46.3% between 2008 and 2020, and 58.0% between 2008 and 2030. From 2008 to 2030 climate change is projected to slow the rate of decrease in diarrhoeal disease burden by 6.5% under RCP 2.6 and 8.9% under RCP 8.5.

Reference storyline results indicate that the total burden of malaria, dengue fever and Japanese encephalitis is expected to decrease 6.6% by 2020 and 20.9% by 2030, without climate change. Under RCP scenario 2.6 the burden decreases 7.5% by 2020 and 16.4% by 2030. Under RCP 8.5, the burden decreases only 6.5% by 2020 and 15.1% by 2030. Thus, between 2008 and 2030 climate change slows the rate of decrease in vector-borne disease burden by 21.2% under RCP 2.6 and by 27.8% under RCP 8.5. Notably, in the case of dengue fever, the confidence interval for α included the null value (zero).

At the national level, by 2020 the development delay imposed on China by climate change (Table 3) ranges from 1.2 months (RCP 4.5 and exponential development path; storyline 2.3) to 48 months (RCP 8.5 and maintenance development path; storyline 4.1). By 2030, the delay ranges from 8 months (RCP 6.0 and exponential path; storyline 3.3) to 62 months (RCP 8.5 and maintenance path; storyline 4.1). However, the delay in 2030 ranged from 9 months (under RCP 4.5) to 21 months (under RCP 8.5) when assuming a linear water and sanitation development path that makes minimal assumptions about the future trend in infrastructure improvement in China beyond the trend observed from 1990 to 2012 (ref. 27; Table 3). On the same development path, the delay reaches 15 months under RCP 2.6, which is the climate change scenario consistent with the lowest greenhouse gas emissions²⁸. The full range of development delays imposed by climate change, and the sensitivity of these values to changes in development paths, RCPs and uncertainty in α, can be seen in Fig. 3 and Supplementary Fig. 9; geographical variation in delays across China is shown in Supplementary Fig. 10.

This analysis represents the first assessment of the impact of climate change on WSH-attributable diseases across China. The country’s demographic and epidemiologic transitions, as well as changes to water and sanitation infrastructure, are expected to rapidly decrease the burden of infectious disease²⁹. Our findings indicate that climate change will blunt this progress, even while controlling for demographic changes, infrastructure development and urbanization (Fig. 3 and Supplementary Fig. 10).

The development delay imposed by climate change in 2030 amounts to 15 months under RCP 2.6 and 21 months under RCP 8.5, assuming a linear water and sanitation development path. In other words, in the presence of climate change, China loses on the order of 1–2 years of progress in reducing the burden of diarrhoeal and other WSH-attributable diseases; an additional 15–21 months of continued infrastructure investment, urbanization and demographic shifts will be required to achieve the decreased disease burden projected in the absence of climate change. Under a storyline assuming high emissions (RCP 8.5) and a maintenance development path, this development delay increases to over 5 years by 2030. Under the most aggressive (exponential) water and sanitation development path, the development delay by 2030 is 13 months under RCP 2.6 and 18 months under RCP 8.5.

After RCP 8.5, RCP 2.6 produces the next longest development delay, despite being the scenario consistent with the lowest concentration of greenhouse gases in 2100. The apparent paradox is the result of the radiative forcing pattern experienced under RCP 2.6 (ref. 28), which peaks at 3.1 W m⁻² in the mid-twenty-first century. Thus, our time frame captures a period during which the projected Td (°C) for China is greater under RCP 2.6 than under RCPs 4.5 or 6.0 (see Supplementary Figs 2–4), before the projected decline in radiative forcing occurs²⁸. This reinforces the need, when carrying out impact studies, to carefully consider the time frame over which impacts are being estimated in the context of the time course of warming projected by the scenarios used.

There were two potential sources of uncertainty in our analysis. The first was the rate at which improved water and sanitation access will increase. The three development paths presented represent increasingly ambitious policy options for China’s future investment in water and sanitation infrastructure. The linear path was consistent with China’s recent development as described by the WHO (World Health Organization) and UNICEF (United Nations International Children’s Emergency Fund) Joint Monitoring Program²⁷, and thus was considered the midline estimate for our analysis. The most conservative estimate, the maintenance-level path, assumed that access to improved water and sanitation was held constant at 2008 levels. This may seem unrealistic for a rapidly developing country, yet China’s unprecedented rural-to-urban migration and its population growth³⁰ will necessitate tremendous investments to maintain 2008 levels of access, particularly in urban areas. However, even if China were to exponentially increase the proportion of the population with access to improved water and sanitation as in the most liberal path, we project 8–18 months (depending on RCP) of development delay due to climate change by 2030.

Of interest, for diarrhoeal diseases the highest pooled estimates of α that we derived from the literature were not associated with low water and sanitation access scenarios. This may be due to the fact that α estimates across water and sanitation access scenarios are based on very few studies. However, the use of an aggregate α, which pooled a larger number of studies, instead of stratified values had little impact on our estimates of diarrhoeal disease burden (changing estimates by − 0.04 to 2.85%). It is also possible that settings with higher baseline exposures because of poorer infrastructure may have lower α estimates because the physiologic response is blunted (that is, a hormetic effect). Still, for reasons unrelated to scenario-specific α estimates, our analysis illustrates the limits of adaptation in provinces where access to improved water and sanitation is very high. Populations in these settings have reached their capacity to adapt through improving water and sanitation infrastructure, and our analysis projects large delays attributable to climate change in such provinces (Supplementary Fig. 9a). These findings raise the question of how adaptation afforded through other means (for example, improving health systems) might be used to overcome the burden of disease attributable to climate change in regions that are able to achieve complete access to improved water and sanitation infrastructure, and, more generally, how high- and low-resource populations may both be impacted by climate stressors.

The second source of potential uncertainty was the HadGEM2-ES model output we used to generate our projections. HadGEM2-ES is a complex Earth system model with numerous uncertainties associated with, for instance, the direct and indirect effect of aerosols, biogeochemical processes and carbon cycling. However, the model has been thoroughly evaluated against a variety of observed data and found to realistically represent near-surface temperature, even in the presence of its complex Earth system model components³¹. In this study, we further evaluated the model’s temperature outputs over the baseline period (2006–2010), and found that HadGEM2-ES performed comparable to other CMIP5 models when temperatures are compared to observations (Supplementary Results). Our projected temperature change across China from 2008 to 2030 was consistent with the Intergovernmental Panel on Climate Change Fifth Assessment Report projections over the same time period²⁵ (Supplementary Results). We acknowledge the HadGEM2-ES model as a source of uncertainty even still, but note that a full analysis of this source of uncertainty was beyond the scope of this study.

There are other potential sources of bias, but where possible we sought to direct the influence of our analytical assumptions towards the null. For instance, we assumed that any urban population had access to improved water and sanitation, despite several studies suggesting urban centres in China suffer from overwhelmed and inadequate water treatment facilities and inconsistent provision of piped water^{8, 21}. Furthermore, we limited our consideration to a small subset of WSH-related sequelae associated with climate change, and we did not include the WSH-related impacts of changes in precipitation levels, shifts in variability of temperature, sea level rise, salt-water intrusion, subsidence, flooding events, destruction of existing infrastructure, increased urban heat island effect, or drought on the incidence of WSH-attributable disease^{26, 32}. Thus, the baseline burden of diarrhoeal diseases presented in our previous analysis is probably an underestimate, as are the results of the analysis presented here.

Our analysis required the assumption that response functions representing the relationship between temperature and WSH-attributable diseases derived elsewhere apply to the population being studied, although we adjusted these functions to match provincial-level infrastructu