A sixfold rise in concurrent day and night-time heatwaves in India under 2 °C warming

Heatwaves with severe impacts have increased and projected to become more frequent under warming climate in India. Concurrent day and nighttime heatwaves can exacerbate human discomfort causing high morbidity and mortality; however, their changes in the observed and projected climate remain unrecognized. Here using observations and model simulations from climate of 20th century plus (C20C+) detection and attribution (D&A) and coupled model intercomparison project 5 (CMIP5) projects, we show that 1 and 3-day concurrent hot day and hot night (CHDHN) events have significantly increased during the observed climate in India. Our results show that the anthropogenic emissions contribute considerably to the increase of 1 and 3-day CHDHN events in India. The frequency of 3-day CHDHN events is projected to increase 12-fold of the current level by the end of 21st century and 4-fold by the mid 21st century under the high emission pathway of RCP 8.5. The increase in 3-day CHDHN events can be limited to only 2-fold by the end of 21st century under low emission scenario of RCP 2.6. One and 3-day CHDHN events are projected to increase by 4, 6, and 8 folds of the current level in India under the 1.5, 2, and 3 °C warming worlds, respectively. Restricting global mean temperature below 1.5° from the pre-industrial level can substantially reduce the risk of 1 and 3-day CHDHN events and associated implications in India.


Results and Discussion
First, we estimate the average frequency of 3-day CHDHN events during the observed period of 1951-2016 ( Fig. 1). On an average two 3-day CHDHN events per year occurred over most of India during 1951-2016 suggesting that 3-day CHDHN are not common during the summer (April-June) season in the observed climate. To analyze if these events have increased during the recent period, our period of analysis was divided into two halves (as the period 1951-1983 (pre-1984), and 1984-2016 (post-1984)) each consisting of 33 years. Then, the average frequency of 1 and 3-day CHDHN events in these two periods was estimated (Figs 1 and S1), and the difference was examined (post-1984-pre-1984). We show that western, north-eastern, and southern part of India have experienced an increase of about three events (3-day CHDHN) per year in the post-1984 period. Furthermore, night-time extreme heat events are increasing more rapidly than daytime events (Fig. S2). The frequency of 3-day CHDHN events has declined over the Indo-Gangetic plain and part of eastern India in the post-1984 period (Fig. 1d). This decline in CHDHN events can be partially attributed to the influence of irrigation and atmospheric aerosols 21,22 . The indo-Gangetic plain is one of the most heavily irrigated regions in the world 23 . Irrigation influences the surface energy budget over the region by increasing latent heat flux and decreasing sensible heat flux 24 . The increase in latent heat flux enhances evaporative cooling which in turn results in reduced surface air temperature 25,26 . A decreasing trend in pan evaporation in India has been reported 27 during 1971-2010, which is mainly due to decline in short-wave radiation. However, both vegetation and evapotranspiration have substantially increased over the Indo-Gangetic Plain mainly due to intensive irrigation 21,28,29 . Therefore, increased evaporative cooling over the Indo-Gangetic Plain can be attributed to irrigation. Other than irrigation, atmospheric aerosols may also play a role in offsetting surface temperature by solar dimming over the Gangetic Plain, which has been reported in previous studies [30][31][32][33] . We confirmed this decline in the frequency of 3-day CHDHN over the Indo-Gangetic plain by estimating nonparametric (using Mann-Kendall test and Sen's slope method) trends in 3-day mean daily average temperature for the 1951-2016 period (Fig. S3). A significant (p-value < 0.05) decline in 3-day mean temperature over the Indo-Gangetic Plain indicates the potential role of irrigation and atmospheric aerosols. Overall, the majority of western and southern India has experienced a significant increase in the frequency of 1 and 3-day CHDHN events during the post-1984 period (Figs 1d and S1d). While irrigation and atmospheric aerosols may alter the changes in CHDHN events locally 21,33 , anthropogenic warming can lead to a wide-spread and significant increase. To evaluate the role of anthropogenic emissions on the occurrence of the CHDHN events in India, we estimated the ratio of the number of CHDHN events based on the Hist and HistNat (Hist/HistNat) scenarios during 1975-2013 using 50 simulations from the C20C+ D&A project (see methods for details). The ratio is higher than 1 in the majority of India indicating that anthropogenic emissions contribute to the increased frequency of CHDHN events. Furthermore, western and southern India show that 1 and 3-day CHDHN events have increased 1.5-2 times while in northeastern India, these events have more than doubled due to anthropogenic emissions. Since the C20C+ simulations do not consider the influence of irrigation, the differences in the observed and C20C+ simulations over the Indo-Gangetic Plain are expected. The increase in 1 and 3-day CHDHN events in western and southern India is consistent with the observations (Figs 1 and S1). Overall, a statistically significant (p-value < 0.05) increase in 1 and 3-day CHDHN events is observed due to anthropogenic emissions over India (Fig. 2c).
Next, we analyze the temporal changes in 1 and 3-day CHDHN events in the projected future climate using data from eight CMIP5-GCMs for the four RCPs. For each 21-year moving window centered on each year from 2005 to 2090, the ratio (Future/Current) of the frequency of 1 and 3-day CHDHN events to the current world was estimated (21-year window centered on 2016 based on RCP8.5; see methods). Therefore, CHDHN ratio higher than one suggests an increase in the frequency of 1 and 3-day CHDHN events under the future climate. Under the RCP4.5 scenario, all India averaged frequency of 3-day CHDHN events (relative to the current world) is likely to become 4-fold by the end-21 st century (Figs 3e and S4e). If the global mean temperature continues to rise rapidly, India is projected to witness 4-5 fold rise in 3-day CHDHN events by mid-21 st century under RCP 8.5. More remarkably, India is projected to experience a 12-fold increase CHDHN events by the end of the 21 st century under RCP 8.5 (Figs 3e and S4e). This rise in 3-day CHDHN events under the projected warming will not be localized instead it is projected to cover a majority of India (Figs 3d and S4d). The benefit of climate change mitigation is well reflected as under the low-emission scenario (RCP 2.6), the increase in 3-day CHDHN events can be limited to only 2-fold by the end of the 21 st century in comparison to the 12-fold under RCP 8.5. The increase in both 1 and 3-day CHDHN events is consistent, and the projected future climate and is statistically significant at 5% level in the reference period (Figs 3f and S4f). To quantify the reliability of muli-model ensemble (MME) mean 34 , we estimated the ratio of MME mean and standard deviation (intermodel variation) for each RCPs for the projected future climate. Projections based on CMIP5 models showed the ratio (MME/std) greater than one indicating high reliability of CHDHN projections (Fig. S5b,d). We find that despite the intermodel variation among different RCPs (Fig. S5), our projections of 1 and 3-day CHDHN are robust ( Fig. S5).
To compare the impact of low and high emission pathways (RCP 2.6 and RCP 8.5), we estimated changes in the frequency of 3-day CHDHN events in the mid (2030-2050) and far (2070-2090) periods to the current world (Figs 3a-d and S4a-d). This comparison between the two emission scenarios makes the impact of climate change more prominently visible and has been used in the previous study related to surface temperature over India 35 . The low emission scenario (RCP 2.6) is unlikely to lead to a substantial increase in 1 and 3-day CHDHN events in the mid and far periods of the 21 st century (Figs 3 and S4), which highlights the importance of the climate change mitigation. However, if the global mean temperature follows the high-emission pathway of RCP8.5, the frequency . In addition to that, the increase in the frequency of 1 and 3-day CHDHN events in the mid and far periods are found to be statistically significant at 5% level based on the RCP 8.5 scenarios (Figs 3f and S4f). Under the two intermediate scenarios of RCP 4.5, and 6.0, the frequency of 1 and 3 day CHDHN events is projected to increase by 4-6 fold during the far-period (2070-2090) (Fig. S6), which is higher than the low emission scenario of RCP 2.6. Our results again highlight the benefits of climate change mitigation as the projected increase in the low-emission scenario (RCP 2.6) is much lesser than the other emission scenarios (RCP 4.5, 6.0, and 8.5). Finally, we estimated the changes in the frequency of 1 and 3-day CHDHN events under the 1.5, 2 and 3 °C warming worlds. The Paris agreement aims to limit the global mean temperature below 2 °C and more ambitiously below 1.5 °C from the pre-industrial level by the end of 21 st century. Therefore, changes in the frequency of 1 and 3-day CHDHN events under these temperature targets were estimated to determine the potential benefits of climate change mitigation. Additionally, we considered the 3 °C warming world for our analysis. Raftery et al. 36 argued that the global mean temperature is most likely to overshoot the 1.5 and 2 °C limits and reach the 3 °C (ranging between 2 °C to 4.9 °C) limit by the end of 21 st century. The ratio of the number of 1 and 3-day CHDHN events under the 1.5, 2, and 3 °C warming worlds to the number of 1 and 3-day CHDHN events was estimated under the current climate (Fig. 4). One and 3-day CHDHN events are projected to increase by 4, 6, and 8-folds (of the current level) in India under the 1.5, 2, and 3 °C warming worlds, respectively (Fig. 4). Moreover, all India averaged frequency of 1 and 3-day CHDHN events (Fig. 4g) is likely to increase by 4, 5.5, and 6.5 fold increase under the 1.5, 2, and 3 °C temperature targets respectively. The projected increase in 1 and 3-day CHDHN events under the different temperature targets is statistically significant at 5% level for both mean and distribution (Fig. 4).
The CHDHN events are likely to rise substantially in India under the warming climate. Change in mean and distribution of summer temperature under the future climate can result in an increased frequency of CHDHN  north Atlantic, which results in a cyclonic anomaly over the west of North Africa 9 . Ratnam et al. 9 reported that anomalous cooling over the Pacific causes the occurrence of heat waves over coastal eastern India. Heatwaves in India are influenced by not only the sea surface temperature (SST) anomalies over the Pacific but also by the SST conditions in the Tropical Indian Ocean 18 . Under the warming climate, the frequency of El Nino is projected to rise 40,41 , which can result in more temperature extremes over India 18 .
Other than the large-scale climate features (e.g., ENSO), the frequency of CHDHN events in India under the current and projected future climate can be affected by local conditions (land use/land cover, irrigation, and aerosols) related to land surface and atmosphere. For instance, intensive irrigation and aerosols over the Indo-Gangetic Plain can reduce surface and air temperature 21,22 . Presence of aerosols can influence solar radiation 30 , which in turn can result in changes in the diurnal temperature range (DTR). Our analysis show that CMIP5 models capture the observed DTR variability during the summer season reasonably well (Fig. S7a,b) with low intermodel variation in the majority of India (Fig. S7c). Additionally, other than local and large-scale factors, a combination of high temperature and humidity can cause heat stress 6 , however, our aim was to estimate the changes in CHDHN events based on air temperature in the future climate. Our estimates of reliability (Fig. S7d) confirm the robust increase in CHDHN events in India under the warming climate 42 . Understanding the role of local factors and large-scale teleconnections that can provide the physical explanation of temperature extremes over India is essential and can be attempted in future studies.

Conclusions
We provided the first-ever assessment of concurrent heatwaves in India with the following primary conclusions: 1. The frequency of 1 and 3-day CHDHN events has increased in large part of western and southern India during the post-1984 period. Night-time heat events have increased more rapidly than the day-time heat events during the recent few decades in India. However, Indo-Gangetic Plain and eastern parts have experienced a decline in the frequency of 1 and 3-day CHDHN events, which can be partially associated with the local cooling due to irrigation and atmospheric aerosols.

Data and Methods
We used the daily maximum (Tmax) and minimum (Tmin) air temperature for the period 1951-2016 from the India Meteorological Department (IMD) to estimate CHDHN events in the observed climate. The gridded temperature data are developed using Shepard's distance weighted interpolation 44 using data from 395 observational stations located across India, which are available at 0.5° spatial resolution. We have regridded the temperature data to 1° spatial resolution using bi-linear interpolation to make it consistent with the data from Coupled Model Intercomparison Project Phase 5 (CMIP5) 45  For the projected future climate, we obtained daily Tmax and Tmin data from eight GCMs that participated in the CMIP5 45 . We selected the GCMs based on their availability for the historical scenario and the four representative concentration pathways (RCPs: RCP2.6, RCP4.5, RCP6.0, and RCP8.5) for the r1i1p1 realization available from the period 1950 onwards. The four RCPs correspond to the four different levels of emission scenarios for the future climate with RCP8.5 representing the highest and RCP2.6 representing the lowest emission scenarios. The data were regridded to 1° spatial resolution using bi-linear interpolation to make it consistent with the other (IMD and C20C+) datasets. The regridded temperature was then compared against temperature at GCMs' native resolution, and we found that the interpolated data were consistent spatially and temporally.
The selected GCMs show both negative and positive bias in Tmax and Tmin against the observed data (Fig. S1). Warm bias in CMIP5 GCMs is centered mainly in the central and western India (Fig. S8) well represented in the models 47 and may result in an over or underestimation of GCMs simulated evapotranspiration 48 . To understand uncertainty CHDHN events, we estimated multi-model mean and standard deviation (uncertainty) for diurnal temperature range (DTR) during the summer season (AMJ) for the historic period of 1971-2000 (used as reference period to estimate CHDHN events). Consistent with the warm bias over the western and central India, we notice an intermodel uncertainty of 1-3 °C in DTR inCMIP5-GCMs (Fig. S7c). A percentile-based approach was used to estimate the CHDHN events based on the summer (April to June) daily Tmax and Tmin during the reference period of 1971-2000. Hot days (nights) for each grid location are identified if Tmax (Tmin) exceeds the 95 th percentile threshold for the reference period . Then CHDHN events were identified if a hot day and hot night occur for the same day for a given period (Fig. S9). Lin et al. 49 and Karl and Night 17 reported that extreme temperature events occurring continuously for 3 or more days pose a significant threat to human health. Therefore, we analyze the CHDHN events following the same methodology (as for daily events) but using the 3-day moving mean of daily Tmax and Tmin. Thus, our analysis is based on two sets of CHDHN events, one based on the daily Tmax and Tmin data  and the other is based on the 3-day Tmax and Tmin .
Extreme climatic events attributed to increased greenhouse gas emission have significantly increased notably since the mid-1970s 50 . Therefore, we focus on the period 1975 onwards to study the role of anthropogenic emission on CHDHN events. To do so, we estimated the ratio of the number of CHDHN events in the Hist to the number of CHDHN events in the HistNat scenario. Since the Hist scenario incorporates the anthropogenic forcings unlike the HistNat scenario which represents the climate without human influence, a ratio higher than one indicates the presence of the anthropogenic contribution to the CHDHN events.
The changes in the frequency of the CHDHN events under the projected future climate were estimated based on the 8 CMIP5-GCMs for the four RCPs to the current world. We define the current world as the period of 21 years centered on 2016 based on the highest emission scenario (RCP8.5) to avoid the overestimation in our estimates as described in King et al. 51 . This is also justified as the observed increase in the global mean temperature is higher than the lower emission scenarios 36  Finally, the changes in CHDHN events were analysed under the warming limits of 1.5, 2, and 3 °C rise in global mean temperature to the pre-industrial period (1861-1900). We followed the same procedure as used in King, et al. 51 and selected the model years corresponding to each warming limit (1.5, 2, and 3 °C) and refer them as 1.5, 2, and 3 warming world. First, we estimated temperature anomaly of the decadal average of global mean temperature using the baseline period (pre-industrial:1861-1890) under the historical scenario. We selected the 1.5 °C world for each model as all years within the decades with temperature 1.3-1.7 °C warmer than the corresponding model baseline for all the RCPs. For the 2 °C and 3 °C warming worlds, the same method was applied, only the temperature range was changed to 1.8-2.2 °C, and 2.8-3.2 °C, respectively. Then, the CHDHN ratio was estimated for each grid as the ratio of the number of CHDHN events in the 1.5, 2, and 3 °C worlds, respectively, to that in the current world.

Data Availability
All the data used in this study will be made available on request to the corresponding author.