Unprecedented Human‐Perceived Heat Stress in 2021 Summer Over Western North America: Increasing Intensity and Frequency in a Warming Climate

The unprecedented 2021 June‐July heatwave in Western North America resulted in record‐breaking human‐perceived heat stress across the region, measured by the humidex considering both air temperature and humidity. During extended summer (June‐September), both 95th percentiles of daily maximum humidex (HX95) and air temperature (TX95) have increased over the 1940–2022 period, with even faster intensification in the last two decades (2001–2022). HX95 has increased more than TX95 because of the positive monotonic nonlinear relationship between humidex and air temperature at a given level of relative humidity. The Canadian Earth System Model version 5 (CanESM5) projects a larger increase in human‐perceived heat stress than air temperature across the region under low to high emission scenarios (HX95 increases 4.40–7.04°C and TX95 increases 2.92–4.65°C between 1981–2010 and 2041–2060). Moreover, CanESM5 projects significant increases in the frequency of HX and TX conditions that exceed the levels reached in 2021 under intermediate and high emission scenarios.

• Unprecedented 2021 heatwave in Western North America resulted in record-breaking human-perceived heat stress, considering both air temperature and humidity • Extreme human-perceived heat stress increased at a faster rate than extreme air temperature, both showing rapid increases in recent decades • For events exceeding 2021 level, a larger future increase in extreme human-perceived heat stress is projected compared to air temperature

Supporting Information:
Supporting Information may be found in the online version of this article.
10.1029/2023GL105964 2 of 11 For example, Lytton in British Columbia set a new Canadian record of 49.6°C (ECCC, 2021) and Hanford in Washington state set a new state record of 48.98°C (Loikith & Kalashnikov, 2023).This extreme event had catastrophic impacts, such as deaths, marine life mortalities, crop losses, river flooding, wildfires, and landslides (White et al., 2023).Without reducing emissions below intermediate levels (SSP2-4.5),global climate models (GCMs) project that WNA will regularly face extreme summer temperatures like 2021 (Dong et al., 2023).
Despite the significant risks to human health associated with the combined effects of temperature and humidity (Kjellstrom et al., 2016;Scoccimarro et al., 2017), the assessment of human-perceived heat stress resulting from the 2021 summer heatwave event in the WNA region has been overlooked.Under hot and humid conditions, the body's ability to cool down through sweating is hindered by elevated atmospheric moisture, which can lead to adverse health effects and even mortality because of the combined impact of environmental and internal heat (Kjellstrom et al., 2016;Sherwood & Huber, 2010).Various temperature relevant indices have been proposed to measure human-perceived temperature, such as the Environment and Climate Change Canada (ECCC) humidex (Diaconescu et al., 2023), National Weather Service heat index (Rothfusz, 1990), Australian Bureau of Meteorology simplified wet-bulb globe temperature (Buzan et al., 2015), German Meteorological Service perceived temperature (Staiger et al., 2012), and US Army and Marine Corps wet-bulb globe temperature (Chavaillaz et al., 2019).Climate projections of the human-perceived temperature suggest a faster increase compared to air temperature throughout the 21st century (e.g., Coffel et al., 2018;Fischer & Knutti, 2013;W. Li et al., 2019;Matthews et al., 2017;Russo et al., 2017;Schwingshackl et al., 2021;Scoccimarro et al., 2017;Zhu et al., 2019).
Here, we evaluate the human-perceived heat stress, considering both air temperature and humidity, in the WNA region during the 2021 summer heatwave.We calculate ECCC humidex values and compare them with corresponding air temperatures.We first examine observed extreme events, defined as the 95th percentiles of daily maximum humidex and air temperature (HX95 and TX95) during each extended summer period (June-September), over the 1940-2022 period.We then explore climate model projections of changes in summer HX95 and TX95, as well as the occurrence probabilities of extremes surpassing the levels observed in 2021, across the WNA region in the 21st century.

Data Sets
The observation-based analysis is conducted using the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis version 5 (ERA5) data set (Hersbach et al., 2020).ERA5 is the latest atmospheric reanalysis data set of the ECMWF, which utilizes an integrated forecasting system with a horizontal resolution of 0.25°.It covers the period from 1940 to the present and integrates observations from diverse sources, including satellites, ground-based stations, and weather stations, in conjunction with an advanced numerical weather prediction modeling system.The combination of these data sources enables the ERA5 to provide a consistent and detailed representation of past weather conditions.Hourly air temperature and dew point temperature at 2-m are used to calculate the humidex.To ensure compatibility with CanESM5, we employ area-averaged ERA5 variables on a consistent 2.5° × 2.5° grid.
Historical climate states and future projections are from a bias-corrected 50-member ensemble of initial condition simulations from the Canadian Earth System Model version 5 (CanESM5) (Swart et al., 2019) under three different Coupled Model Intercomparison Project Phase 6 (CMIP6) emission scenarios: low (SSP1-2.6),intermediate (SSP2-4.5)and high (SSP5-8.5)scenarios.CanESM5 is the latest Earth System Model (ESM) developed by the Canadian Centre for Climate Modeling and Analysis of ECCC, which contributes to the CMIP6.We use near-surface (2-m) daily maximum air temperature and the daily minimum relative humidity outputs from the historical experiment (1850-2014) and the three emission scenarios (2015-2100).The members of the CanESM5 ensemble are subjected to identical historical and future external forcings, differing slightly only in the initial conditions in 1850.Consequently, these ensemble members exhibit different realizations of internal variability in response to the specified external forcing (e.g., Deser et al., 2012;Fyfe et al., 2017;Kay et al., 2015).The CanESM5 employs the spectral transform method for the atmosphere component, featuring a horizontal resolution of approximately 2.8° latitude × 2.8° longitude.To facilitate the assessment, variables of interest are also interpolated to a grid spacing of 2.5° × 2.5° using a first-order conservative remapping method (Schulzweida, 2019). 10.1029/2023GL105964 3 of 11 2.2.Humidex Index Masterton and Richardson (1979) introduced the humidex as an empirical heat stress index that relates outdoor thermal discomfort experienced by the general population with the air temperature (Ta) and the vapor pressure (e), based on relative humidity RH (%) or dew point temperature Td (°K).The humidex is expressed as a linear combination of Ta (°C) and e (hPa), using the following equations: (3) There is a great similarity between the humidex values computed using both RH and Td (Figure 1 of Diaconescu et al. (2023)).
Figure S1 in Supporting Information S1 demonstrates the interrelationships between Ta, RH, and the humidex.
According to the equations, these variables show a direct relationship, where an increase in Ta or RH results in a positive monotonic nonlinear increase of the humidex.For instance, when Ta is 30°C, the humidex ranges from 29.3 to 45.8°C for RH levels of 20%-90%.At a specific RH level, the humidex increases more than Ta, resulting in an expansion of the humidex range as Ta increases within the range of RH.The level of discomfort experienced during prolonged exposure and physical activity can be categorized into five levels: discomfort perceived by a few people (30°C > humidex ≥ 25°C), noticeable discomfort (35°C > humidex ≥ 30°C), evident discomfort (40°C > humidex ≥ 35°C), intense discomfort (45°C > humidex ≥ 40°C), and serious danger (humidex ≥ 45°C) (Alfano et al., 2010).
The daily maximum humidex (HX) of ERA5 is obtained by extracting hourly humidex values calculated from 2-m Ta and Td data for 1940-2022 period using Equations 1 and 3.As ERA5 does not provide hourly RH data, we estimate it using Ta and Td with the approximation of the Clausius-Clapeyron equation (Shaman & Kohn, 2009), using the following equations: where e is the vapor pressure for a given Td shown in Equation 3 and e s (hPa) is the saturation vapor pressure for a given Ta (°K).In CanESM5, we derive HX values for the historical (1850-2015) and future (2015-2100) periods using Equations 1 and 3 based on the daily maximum Ta (TX) and the daily minimum RH, which typically coincide within a given day (see Figure 2 of Diaconescu et al. ( 2023)).

Bias Correction
Bias correction methods applied to GCM outputs effectively reduce biases in historical simulations and generate future projections that closely align with observed data, playing a crucial role in climate change impact assessments and adaptation studies (e.g., Dieng et al., 2022;X. Li & Li, 2023;Maraun, 2016).This study applies the Quantile Delta Mapping (QDM) method (Cannon et al., 2015) to perform bias correction on the CanESM5 variables, specifically TX and HX.QDM is used to correct historical CanESM5 outputs to match the empirical distribution of ERA5 during the 1981-2010 baseline period.Outside of the baseline period, QDM applies the same bias correction adjustments, with the added constraint that projected changes in quantiles of the CanESM5 simulations are preserved.Internal variability of the CanESM5 ensemble is maintained by calculating the historical bias adjustment parameters using a single CanESM5 realization.To account for biases in the seasonal cycle, QDM is applied to pooled daily data within sliding 3-month windows centered on the month of interest.Figure S2 in Supporting Information S1 presents Taylor diagrams comparing CanESM5's performance between the original and QDM-adjusted TX95 and HX95 values over the study region. 10.1029/2023GL105964 4 of 11

Human-Perceived Heat Stress in 2021 Summer Heatwave Over Western North America
To evaluate the human-perceived heat stress during the 2021 heatwave in the WNA region, we analyze the anomalies of HX95 over North America using ERA5 data (Figure 1a).The findings indicate that the 2021 heatwave in the WNA region resulted in an unprecedented level of human-perceived heat stress, as measured by the humidex.Numerous scattered areas show significant deviations in HX95, with some exceeding 6°C, primarily during late June to early July 2021, when compared to the average HX95 for extended summer (June-September) during the 1951-1980 period.Based on these anomalies, we define an active center of action, encompassing most grids with HX95 deviations exceeding 5°C compared to the average of HX95 for 1951-1980.The WNA heatwave region includes parts of the western and central provinces and territories of Canada, as well as Washington and portions of the northwestern states in the US.This region largely overlaps with the area used in previous studies, which have mainly been defined based on extreme temperatures during the 2021 heatwave event (e.g., Dong et al., 2023;Heeter et al., 2023;Lin et al., 2022;Qian et al., 2022;Thompson et al., 2022;White et al., 2023).The HX95 anomaly in the summer of 2021 was approximately +5.40°C higher than the average of 1951-1980 (Figure 1b).
While less anomalous than those for HX95, the 2021 heatwave in the WNA region also resulted in a record-breaking air temperature extreme (Figure 1c), with above-normal TX95 anomalies (with respect to 1951-1980) observed across grid points in the region.The result here is consistent with previous studies that examined anomalies in mean air temperature in the region, albeit for slightly different timeframes, including Bartusek et al. (2022) for 25 June to 3 July, Lucarini et al. (2023) for 24 June to 8 July, and Lin et al. (2022) for 28 June to 4 July in 2021.
The 2021 TX95 over the WNA region was +4.85°C higher than the 1951-1980 average (Figure 1d).The spatial distributions of the 2021 HX95 and TX95 anomalies show similarities, with spatial correlation coefficients of 0.88 over North America and 0.68 over the WNA region.This suggests that the spatial pattern of HX95 anomalies can be largely explained by that of TX95 anomalies, accounting for approximately 78% of spatial variability over North America and 46% over the WNA region.
Figure 1e shows the timeseries of anomalies of HX95 and TX95 over the WNA region from 1940 to 2022.Both HX95 and TX95 have increased significantly over this period, with HX95 rising by 0.35°C/decade and TX95 by 0.31°C/decade.The recent trends from 2001 to 2022 are even more pronounced, with HX95 and TX95 increasing by 0.76 and 0.72°C/decade, respectively.HX95 exhibits a higher rate of increase compared to TX95 for both periods.This is due to the positive nonlinear relationship between humidex and air temperature at a specific relative humidity level, as shown in Figure S1 in Supporting Information S1.There is a strong correlation between the annual time series of HX95 and TX95, with a correlation coefficient of 0.97, indicating that 94% of the inter-annual variability in HX95 can be explained by that of TX95 when a simple linear regression is applied between the two variables.A relatively small amount of inter-annual variability (3%) can be explained by the daily minimum RHs for the entire period.The faster increase of human-perceived heat stress compared to air temperature, along with greater increases in both variables in recent decades, is consistent with findings from previous studies (e.g., Lee et al., 2021;J. Li et al., 2018;Russo et al., 2017).In addition, both HX95 and TX95 show the highest anomalies in 2021.However, the mean anomaly of daily minimum RH for days with HX > HX95 in 2021 was −8.9% relative to the average of 38.8% for 1951-1980, indicating that the extreme humidex values observed in 2021 were associated with drier than average conditions.The 2021 heatwave resulted from a combination of high atmospheric pressure and dry condition, as highlighted in recent studies (e.g., Bartusek et al., 2022;Lucarini et al., 2023;Thompson et al., 2022;Yu et al., 2023).Clear sky and low soil moisture created positive feedback, where more solar energy was converted into sensible heat instead of evaporating water from the soil, further warming the air near the ground (Jeong et al., 2022;Pfahl & Wernli, 2012).

CanESM5 Simulations and Projections
Over the historical period, the ranges of internal variability in the bias corrected HX95 and TX95 series from CanESM5 are consistent with inter-annual variability in both the HX95 and TX95 series from ERA5 (Figures 2a  and 2b).Moreover, the 2021 HX95 and TX95 values from ERA5 fall within the range of internal variability simulated by CanESM5, indicating that the CanESM5 historical simulations adequately capture the occurrence of the unprecedented extreme events in 2021.However, CanESM5 exhibits stronger increasing trends for both HX95 and TX95 when compared to ERA5 (Figures 1e, 2a, and 2b) over the most recent decades.For instance, the The projected increases in HX95 are higher than those of TX95 in the 21st century, suggesting that human-perceived heat stress is rising at a faster rate than air temperature extremes in the WNA region (Figures 2a  and 2b).In CanESM5, global mean surface temperature is projected to increase 2.22°C for SSP1-2.6,2.70°C for SSP2-4.5, and 3.28°C for SSP5-8.5 by 2041-2060 relative to the pre-industrial period .Noting the high sensitivity of CanESM5, 2041-2060 corresponds to global warming levels of 1.7°C, 2.0°C, and 2.4°C under each of these scenarios (Lee et al., 2021).Relative to the 1981-2010 reference period, HX95 increases by 4.4°C for SSP1-2.6,5.46°C for SSP2-4.5, and 7.04°C for SSP5-8.5 in the middle of the century, while TX95 displays increases of 2.92°C for SSP1-2.6,3.64°C for SSP2-4.5, and 4.65°C for SSP5-8.5 over the same period.The 5%-95% ranges of HX95 are approximately 25%-30% broader than those of TX95 during both the historical and future periods.This indicates that HX95, which accounts for both air temperature and humidity, exhibits a greater degree of internal variability compared to TX95, which is solely based on temperature.According to CanESM5 projections, the average HX95 in the WNA region is expected to surpass the "evident discomfort" threshold (>35°C) by 2092 under SSP2-4.5 and by 2056 under SSP5-8.5.In addition, since approximately 2010, clear signals of change for both HX95 and TX95 have emerged, outweighing the noise across all SSP scenarios.This is evident from the signal-to-noise ratio, calculated by dividing the mean changes relative to the reference period by the standard deviation of the 50-member ensemble (Figures 2c and 2d).Meanwhile, HX95 consistently demonstrates higher signal-to-noise ratios than TX95 for the same SSP scenario and future years, indicating HX95 is more responsive to changes in external forcing than TX95.
CanESM5 projections indicate an increased frequency of HX and TX values under the intermediate and high emission scenarios (SSP2-4.5 and SSP5-8.5),while showing a similar frequency under the low emission scenario (SSP1-2.6)(Figures 3a and 3b).Changes in frequency are quantified by the risk ratio, which is the ratio in the occurrence probabilities of HXs or TXs surpassing the 2021 HX95 or TX95 levels in a future year relative to the probabilities of 2021 (i.e., 5% for each).For the same scenario and future year, HX > 2021 HX95 will occur more often than TX > 2021 TX95.In CanESM5, global mean surface temperature is projected to increase 2.43°C for SSP1-2.6,3.80°C for SSP2-4.5, and 6.56°C for SSP5-8.5 at the end of the century (2081-2100), relative to the pre-industrial period.Relative to the 1981-2021 reference period, risk ratios [5th and 95th percentiles] of HX values at the end of the century are projected to increase to 2.11 [0.61, 1.23] for 3.11 [2.01,4.44]for and 9.95 [8.41,11.39]for SSP5-8.5, with statistically significant changes observed under SSP2-4.5 and SSP5-8.5 (i.e., quantified in terms of internal variability at a 90% confidence level, as indicated by their 5th percentiles being above one).While the risk ratios [5th and 95th percentiles] of TXs are smaller, they are projected to increase to 0.75 [0.36, 1.40] for SSP1-2.6,1.83 [1.05, 2.94] for and 6.96 [5.30,8.60]for SSP5-8.5 during the 2081-2100 period, with statistically significant changes observed under SSP2-4.5 and SSP5-8.5 as well.In addition, it is projected that the WNA region will, on average, experience approximately 6.7 (12.7), 10.0 (18.7), and 16.5 (60.0) days of HXs and about 4.5 (4.5), 6.3 (11.0), and 10.2 (41.9) days of TXs exceeding the 2021 HX95 and TX95 levels by 2050 (2090) for the low, intermediate, and high emission scenarios, respectively.Risk ratios for greater extremes (i.e., HXs and TXs exceeding 2021 daily maximum humidex and air temperature (HXmax and TXmax) show even faster and larger increases (Figures 3c and 3d).The projected change of TX aligns with findings from previous studies that have reported a more frequent occurrence of air temperature extremes surpassing the 2021 level in the WNA region across various future emission scenarios, utilizing different statistical measures including 5-day temperature anomalies (Dong et al., 2023), maximum daily mean temperature during June-August (Bartusek et al., 2022), and temperature anomalies for June-August (Heeter et al., 2023).However, CanESM5 generally has difficulty simulating extreme dry humidex values (i.e., those that coincide with drier than average conditions, as in the 2021 event) in the current climate (Figure 3e).
Based on the increase in air temperature, CanESM5 projects an average of 1.45 days that are higher than the 2021 HXmax and smaller than the 2021 RH at HXmax for SSP2-4.5 and 8.76 days for SSP5-8.5 per summer by 2090.
Changes in extreme HX are primarily driven by corresponding changes in TX, with RH associated with extreme HX generally showing minor increasing changes of less than 3% on average for both mid-century and end-of-century periods, except for the SSP5-8.5 for the end-of-century period (Figures S3a and S3c in Supporting Information S1).Figures S3b and S3d in Supporting Information S1 further illustrate the internal variability of CanESM5 in simulating the relationship between extreme HX values and their corresponding TX and RH values for the center years of the mid-century and end-of-century periods (2050 and 2090); values corresponding to the 2021 event in ERA5 are shown for reference.The results illustrate that the anomalies of HX > HX95 and TX are dependent, with linear correlation coefficients ranging from 0.75 to 0.86 for a given emission scenario in both the mid-century and end-of-century periods, whereas there is usually no statistically significant correlation between the anomalies of extreme HX and the corresponding RH.The Figures S3b and S3d in Supporting Information S1 show that the 2021 extreme HXs surpassing HX95 had a mean anomaly of daily minimum RHs of −6.6% relative to the average of the reference period, and extreme HXs surpassing HX95, under a dry condition like the 2021 level, might not be common in the mid-century and end-of-century periods as demonstrated by the internal variability of the future years.This finding could be partly explained by the conclusion of Jeong et al. (2022) that the relationship between atmospheric blocking, which was a key contributor to the 2021 event, and heatwaves is expected to remain consistent between the reference and future periods over the North America.Atmospheric blocking results in prolonged periods of above-normal high-pressure, clear-sky, and dry conditions (Bartusek et al., 2022;Lucarini et al., 2023;Thompson et al., 2022).

Summary and Discussion
By utilizing the ERA5 reanalysis data set, we showed that the 2021 heatwave in the WNA region resulted in unprecedented records for both air temperature and human-perceived heat stress, as indicated by the anomalies of the 95th percentiles of daily maximum humidex (HX95) and air temperature (TX95) during the extended summer period (June-September 2021), surpassing the 1951-1980 average by +5.40°C and +4.85°C, respectively.The analysis also revealed significant increasing trends in both HX95 and TX95, with the trend being more prominent for HX95 than for TX95, over a long historical period  and recent decades .
The HX95 anomalies exhibited a strong correlation with TX95 anomalies, indicating that air temperature anomalies can largely account for the variations in human-perceived heat stress, explaining approximately 94% of the inter-annual variability over 1940-2022 and 46% of the spatial variability observed during the 2021 event in the WNA region.However, the human-perceived heat stress experienced in 2021 was associated with drier conditions than the average.This can be attributed to a combination of factors, including prolonged high atmospheric pressure (atmospheric blocking), clear skies and low soil moisture, as highlighted in recent studies.
In terms of future projections from CanESM5, all emission scenarios (SSP1-2.6,SSP2-4.5, and SSP5-8.5)indicate notable increases in HX95 over the WNA region, ranging from 4.4°C to 7.04°C, during the mid-century (2041-2060) period compared to the reference (1981-2010) period.In contrast, TX95 exhibited smaller increases, ranging from 2.92 to 4.65°C for the same period and scenarios.This can be attributed to the positive monotonic nonlinear relationship between humidex and air temperature at a given relative humidity, resulting in higher increasing rates of human-perceived heat stress compared to air temperature.The internal variability of HX95 is also larger than that of TX95, suggesting a more pronounced variability in human-perceived heat stress due to the combined effects of air temperature and relative humidity.Furthermore, CanESM5 projections indicate a substantial increase in the frequency of HXs and TXs surpassing the 2021 level HX95 and TX95 in the WNA region, with estimates of a threefold increase for HXs under SSP2-4.5 and a tenfold increase under SSP5-8.5, as well as a twofold increase for TXs under SSP2-4.5 and a sevenfold increase under SSP5-8.5 during the long-term future (2081-2100) period.The projections also indicate that changes in extreme HXs will primarily be driven by corresponding changes in TXs in the future, while daily minimum relative humidity associated with extreme HXs is expected to pay a relatively minor role.However, the probability of extreme HXs occurring in dry conditions like the 2021 event is expected to remain consistently low compared to historical levels in the future, potentially influenced by the persistent interaction of atmospheric blocking and high temperatures between the reference and future periods in the WNA region (Jeong et al., 2022).
Future projections of CanESM5 indicate a faster increase in HX95 compared to TX95, emphasizing the rising impact of temperature extremes on human-perceived heat stress.These findings highlight the importance of considering both temperature and humidity when assessing heat stress and call for further attention to the potential health implications and adaptation strategies in the face of increasing heat stress.The high climate sensitivity of CanESM5 among CMIP6 GCMs (Meehl et al., 2020) results in a faster increase in HX95 and TX95 from 1940 to 2022 compared to ERA5, suggesting the possibility of higher rates of warming in CanESM5 compared to other GCMs.Moreover, CMIP6 GCMs generally project slight decreases in relative humidity, ranging from 0% to −4%, over the region until 2081-2100 (Lee et al., 2021), which differs from CanESM5, particularly for the SSP1-2.6 and SSP2-4.5 scenarios.Hence, further study of heat stress projections in the CMIP6 multi-model ensemble that accounts for uncertainties due to model structure, resolution, parameterizations, climate sensitivity, and other factors is warranted.

Figure 1 .
Figure 1.(a) Anomalies of HX95 during the summer (June-September) of 2021 over North America compared to the 1951-1980 period, calculated from ERA5.The cyan contour represents the study region (Western North America [WNA]), encompassing most grids with the HX95 anomalies exceeding 5°C, observed mainly during late June to early July 2021.(b) Density distribution (bars and black curve) and cumulative probability (blue curve) of the HX95 anomalies over the 1940-2022 period relative to the 1951-1980 period for the WNA region.The dashed cyan line corresponds to the year 2021.(c, d) Are similar to (a, b) but based on TX95 data.(e) Anomalies of summer HX95 (circles) and TX95 (squares) across the WNA region, relative to the 1951-1980 climatology from ERA5.The solid lines and dashed lines indicate the linear trends and their 95% confidence intervals for HX95 (cyan) and TX95 (red), calculated by simple linear regression and t-test for two periods: 1940-2022 and 2001-2022.Mean anomaly of daily minimum RHs is presented for HXs exceeding the HX95 threshold, with different colors specified in the legend, relative to the 1951-1980 period.

Figure 2 .
Figure 2. Projected changes in HX95 and TX95 over the Western North America region under different scenarios.(a, b) Mean changes and 5%-95% internal variability range from the 50-member ensemble of CanESM5 for SSP1-2.6,SSP2-4.5, and SSP5-8.5 scenarios.Cyan solid lines show ERA5 values for 1940-2022.(c, d) Signal to noise ratios calculated by dividing the mean change of each year from reference (1981-2010) period by the standard deviation of the 50-member ensemble for HX95 and TX95.Vertical dashed lines mark the center years of 20-year periods when global mean surface temperature changes from the 50-member ensemble reach a certain level relative to pre-industrial (1850-1900) period.

Figure 3 .
Figure 3. (a, b) Risk ratios calculated as the ratios of the probabilities that summer HXs and TXs exceed 2021 HX95 and TX95 to the probabilities of them in 2021 (i.e., 5%), respectively.(c, d) Risk ratios of summer HXs and TXs exceed 2021 HXmax and TXmax, respectively.(e) Conditional risk ratios in dry conditions calculated as the ratio of the probabilities that summer HXs exceed the 2021 HXmax and the corresponding RHs lower than the 2021 RH of HXmax.Mean changes and 5%-95% internal variability range from the 50-member ensemble of CanESM5 for SSP1-2.6,SSP2-4.5, and SSP5-8.5 scenarios are shown.Cyan solid lines show ERA5 values, which are one in 2021.Vertical dashed lines mark the center years of 20-year periods when global mean surface temperature changes from the 50-member ensemble reach a certain level relative to pre-industrial period.