Human and natural drivers of the recent contrasting trends 1 between daytime and nighttime hot extremes

The frequency and duration of extreme heat events, including heat waves (HWs, daytime 12 hot extremes) and tropical night (TNs), are increasing significantly as the climate warms, 13 adversely affecting human health, agriculture, and energy consumption. Although many detection and attribution studies have examined extreme heat events, the underlying mechanisms associated with the recent increase in HWs and TNs remain unclear. In this study, we analyze the controlling factors behind the distinct increases in HW and TN 17 events over the Northern Hemisphere during boreal summer (June to August). We found that the occurrence of HW events has been increasing gradually since 1980, mostly due 19 to anthropogenic forcing. However, the occurrence of TN events increased abruptly 20 during the late 1990s and has changed little since then. We demonstrate that this sudden increase in TN events is closely associated with low frequency variability in sea surface temperature, including the Pacific Decadal Oscillation, indicating its natural origin. We 23 further found that CMIP5 climate models fail to capture the observed non-linear

occurrence is dominant on the interannual timescales without a linear trend, whereas a lowfrequency variability of TN occurrence is prominent ( Supplementary Fig. 2). The differences 75 in the temporal properties of HWs and TNs could indicate that the causes of their recent 76 increases are not the same. 77 The significant increasing trend in HW occurrence for the entire period suggests that 78 a primary influence is anthropogenic forcing, as indicated in previous studies 2,4 . We discuss 79 this later in our analysis of the CMIP5 climate models. Most previous studies showed that 80 anthropogenic forcing is the primary contributor to the gradual increase in HWs at the regional 81 and global scales 23,24 . They argued that the number of global HW days increases as the global 82 mean surface temperature increases 25,26 . Furthermore, they predict that future HW events in 83 most regions in the NH will become more intense, more frequent, and longer lasting in the 84 second half of the 21 st century 5 . Although a regression analysis does not imply causality, 85 regressed sea surface temperature (SST) anomalies against the frequency of HWs supports the 86 notion that the increasing trend in HW events is associated with an increase in global SST, in 87 addition to an increase in land surface temperature ( Supplementary Fig. 3). Furthermore, the 88 variability in HW events (without a linear trend) is closely associated with an El Nino-like SST 89 structure 27,28 (Supplementary. Fig. 4). This suggests that the increasing trend in HWs in the NH 90 during JJA is mostly due to anthropogenic global warming and the remaining interannual 91 variability in HW occurrence is affected by natural variability, including ENSO.

92
In contrast, the significant increasing trend in TN occurrence across the entire period 93 is characterized by a regime shift-like increase in the late 1990s (Figs. 1e-h and Supplementary 94 Fig. 1). In other words, the increase in TNs after the late 1990s cannot be explained solely by 95 anthropogenic forcing. 96 and JJA present a regime shift increase in the occurrence of HWs in the late 1990s 125 ( Supplementary Fig. 6), mostly due to a large increase in HW occurrences after the late 1990s 126 (Supplementary Table 3) that can be explained by stronger anthropogenic forcing during recent 127 decades. It is noteworthy that the increasing trend in HW events after the late 1990s is also 128 larger than that before the late 1990s in the observations (Supplementary Table 2). Thus, this 129 supports that anthropogenic forcing significantly contributes to the increase in observed HWs  Table 3). We argue that increases in both HW and TN 138 events after the late 1990s could be caused by anthropogenic forcing in CMIP5 climate models.

139
Furthermore, we find coherence in the HW and TN occurrences simulated in the CMIP5 140 climate models regardless of a linear trend, which is in contrast to the observations. Whereas  Table 4). Furthermore, the simultaneous correlation coefficient between 144 the ensemble mean for the occurrence of HW and TN events in JJA is 0.99 and 0.88 with and 145 without linear trends, respectively, which are statistically significant above the 99% confidence  We conclude that the contributing factors associated with variability in the occurrence of HWs 164 and TNs are similar in the CMIP5 climate models, which is not seen in the observation. While 165 the increasing trend in HWs in the NH during JJA is mostly due to anthropogenic global 166 warming, the significant increasing trend in TN occurrence across the entire period is 167 characterized by a regime shift-like increase in the late 1990s. Therefore, the increase in TNs 168 in the observation cannot be explained solely by anthropogenic forcing. In particular, we 169 emphasize that the CMIP5 climate models fail to simulate the characteristics of TN occurrence 170 found in the observations, which could be caused by the models' failure to correctly simulate 171 decadal SST variability in the tropics 29 . This implies that the uncertainty in future projections 172 based on the CMIP5 climate models can be large for extreme heat events, particularly TNs, which should be carefully considered.

176
Observational datasets In this study, we used two observational gridded datasets, the Climate     unit is number of events/year. Statistical significance at the 90%, 95%, and 99% level is 359 denoted by *, **, and ***, respectively.

403
The unit is ℃. The black dots indicate areas significant at the 95% confidence level.

417
HWs without a linear trend. SST anomalies were calculated using climatology across the 418 entire period and averaging the individual models. The unit is ℃. The black dots denote areas 419 significant at the 95% confidence level.