Projected changes in daily precipitation, temperature and wet‐bulb temperature across Arizona using statistically downscaled CMIP6 climate models

To evaluate future changes in the climate system, the outputs from coarse‐resolution global climate models (GCMs) need to be downscaled to a finer scale, making them more directly applicable for impact assessment. Here we focus on examining the projected changes of three key climate variables (precipitation, air temperature, and wet bulb temperature) across Arizona (south‐western United States). We use daily outputs from GCMs from the sixth phase of the Coupled Model Intercomparison Project (CMIP6) and bias correct and downscale them to a 4‐km resolution. Through leave‐one‐out cross‐validation, we compare various bias correction methods and identify that the empirical quantile mapping approach performs the best regardless of the climate variable. Then, we analyse the projected bias‐corrected outputs for two future periods (Mid‐of‐Century: 2015–2048; End‐of‐Century: 2067–2100) with respect to the 1981–2014 period, under four shared socioeconomic pathway scenarios (SSP1‐2.6, SSP2‐4.5, SSP3‐7.0 and SSP5‐8.5). Our results show that Arizona's climate is projected to become overall warmer and wetter, more so towards the end of this century and for higher emission scenarios. Additionally, our findings project an increase in wet bulb temperature and cooling degree days, implying the ongoing warming climate's potential impacts on public health and the economy. These results provide a baseline understanding of climate change impacts in the state and highlight the projected changes in the future in response to the climate scenarios.

Heat extremes are one of the most dangerous climaterelated hazards, with fatal impacts in terms of public health (Jones et al., 2018).Previous studies found that heatwave-related deaths have increased by 0.21% worldwide over the last 20 years and high temperatures will become more frequent as climate change accelerates (Zhao et al., 2021).Moreover, many studies have concluded that more deaths are inevitable as the threats to human life from heat extremes increase (Haines et al., 2006;Kron et al., 2019;Mora et al., 2017).
Arizona is one of the most vulnerable regions to heat extremes, and the number of heat-related illnesses and deaths has been increasing during the last decade (Iverson et al., 2020).A more recent study indicates a general trend in Arizona's climate towards rising temperatures and decreasing total precipitation over the last five decades (Zhang et al., 2021), pointing to significantly warmer temperatures as a consequence of climate change (Garfin et al., 2013).However, precipitation patterns in the state are complex due to different atmospheric phenomena.For instance, there has been an observed increase in precipitation during the North American Monsoon (NAM) season (Ehsani et al., 2022), and an increase in heavy precipitation from atmospheric rivers (ARs) (Gershunov et al., 2019;Zhou & Kim, 2019).Future changes in precipitation and temperature are explored using climate projections from global climate models (GCMs) and regional climate models (RCMs).For example, Almazroui et al. (2021) used 31 GCMs from the sixth phase of the Coupled Model Intercomparison Project (CMIP6) and examined the projected changes in temperature and precipitation over the United States.They concluded that Arizona is projected to experience hotter temperatures overall, with northern Arizona expected to be wetter and southern Arizona expected to be drier.Georgescu et al. (2022) explored the future precipitation changes across the state using dynamically downscaled GCMs, finding an increase in mean and extreme precipitation for the winter season, with opposite trends in the other seasons.The GCMs and RCMs provide useful information to understand the overall trend in future climate for different climate scenarios.However, such projections are still limited in their utility for quantifying the climate change impacts at a more local scale due to their coarse resolution (Almazroui et al., 2021;Fowler, 2007;Tapiador et al., 2019).
To bridge the gap across scales and improve the skill of climate models, statistical bias correction and downscaling methods represent a viable path (Ahmed et al., 2013;Jakob Themeßl et al., 2011;Maraun et al., 2010).Statistical downscaling and bias correction are based on building a statistical relationship between observations and GCM outputs.This approach has the advantages of being simple and fast to implement, computationally inexpensive and adaptable to regions and variables of interest.There are various techniques available for statistical bias correction, ranging from simple linear scaling to more complex methods such as quantile mapping (Cannon et al., 2015;Maraun, 2013;Shrestha et al., 2017).Ghimire et al. (2019) employed eight statistical bias correction techniques, including linear scaling, parametric and nonparametric quantile mapping methods, for the daily precipitation of 13 GCMs from the fifth phase of the Coupled Model Intercomparison Project (CMIP5) in Myanmar, Southeastern Asia.They showed that these widely used bias correction techniques work well and reduce biases significantly.Golian and Murphy (2022) employed five statistical bias correction methods for precipitation forecasts by the European Center for Medium Range Forecasts (ECMWF) system for Ireland; they showed that the regression-based method (i.e., ordinary least squares regression [OLS]) performs well and the OLS-based ensemble mean shows the best performance for most of the precipitation distribution, but not for the extremes.Chen et al. (2022) applied a statistical bias correction method to directly adjust biases in the annual maximum wet bulb temperature.They derived the wet bulb temperature from climate data obtained from 18 CMIP6 models over China and found that future extreme heat-humidity events are projected to intensify throughout the 21st century.Previous studies indicate that while commonly used bias correction methods exist, no single method works best for all datasets.Accordingly, we need to conduct experiments with different techniques and identify appropriate methods that work best depending on the datasets and the purpose of the bias correction.
With the release of the CMIP6 outputs, the projected changes in climate variables have been extensively analysed using the bias-corrected CMIP6 outputs in different regions (Ayugi et al., 2022;Carvalho et al., 2021;Lovino et al., 2021;Masud et al., 2021;Mishra et al., 2020;Tian et al., 2021).Among the statistical bias correction techniques, quantile mapping has emerged as a popular choice.For example, Rao et al. (2022) employed empirical quantile mapping (eqm) to obtain bias-corrected precipitation and temperature for the Indus basin in India, concluding that daily precipitation and temperature extremes are expected to increase in magnitude.Similarly, Babaousmail et al. (2022) used quantile mapping to investigate the projected changes in extreme precipitation over the Mediterranean and Sahara regions, and emphasized that wet areas are becoming wetter and dry areas drier.Guo et al. (2023) performed downscaling and bias correction for temperature and precipitation from 12 CMIP6 models across China.They employed an improved quantile mapping known as the equidistant cumulative distribution function matching approach; they analysed the occurrence of extreme events using the 99% percentile of the joint precipitation and temperature distributions, concluding the risk of extremes in China is greater for both univariate and concurrent extreme events.
Therefore, while much progress has been made to improve the utility of GCM outputs on a more local scale, there is a need to compare the performance of different bias correction and downscaling techniques and to apply them to assess the projected changes in precipitation, temperature, and wet bulb temperature.This is particularly important in a state like Arizona that has been the subject of extreme climate conditions in recent years.The rest of this article is organized as follows.Section 2 presents the data and methods, and then we present our results and discussion in Section 3. Section 4 provides a summary of the main findings and concludes the study.

| Study area and dataset
Arizona is in the southwestern United States and has two major physiographic regions: the Colorado Plateau and the Basin and Range Province.The southern boundary of the Colorado Plateau is the Mogollon Rim, and a narrow transition zone extends in a northwest-southeast direction (Hendricks & Plescia, 1991).Köppen-Geiger climate types classify Arizona as belonging to arid and semi-arid climates, with most of the southwestern part of the state characterized by warm desert climate, while the northeastern part by a semi-arid one (Beck et al., 2018;Peel et al., 2007).Here we use PRISM datasets (accessible from the Prism Climate Group webpage at https://prism.oregonstate.edu/)from 1981 to 2014 as reference.The daily precipitation and air temperature are directly obtained from the PRISM datasets, and we calculate the wet bulb temperature (T w ) using Equations ( 1) and (2) (Lawrence, 2005;Stull, 2011): where T is the air temperature, atan is the arctangent function, RH% is the relative humidity, and T d is the dew point temperature that can be obtained from the PRISM datasets.
We collect the three climate variables from the CMIP6 outputs.The daily precipitation and air temperature are downloaded from the World Climate Research Programme CMIP6 website (https://esgf-node.llnl.gov/search/cmip6/), and the daily wet bulb temperature is calculated using Equation (1).Table 1 presents the list of the CMIP6 models used in this study for precipitation, air temperature and relative humidity, which shows that we use 22 GCMs for precipitation, 21 GCMs for air temperature, and 20 GCMs for wet bulb temperature.For all the GCMs, we use the 'r1i1p1f1' member, and they all have four shared socioeconomic pathway scenarios (SSP1-2.6,SSP2-4.5,SSP3-7.0 and SSP5-8.5),which provide projections up to the year 2100.Here we consider two future periods (Mid-of-Century: 2015-2048;End-of-Century: 2067-2100) with respect to the historical period .

| Statistical downscaling and bias correction
To perform statistical downscaling and bias correction of CMIP6 outputs, we employ the following ten bias correction methods: detrended quantile matching (dqm), eqm, generalized quantile mapping (gpqm), local intensity of scaling (loci), mean variance adjustment (mva), parametric quantile mapping (pqm), power transformation (ptr), quantile delta mapping (qdm), scaling and variance.Note that 'ptr' and 'loci' are only available for precipitation, and 'variance' is only available for temperature; therefore, we use nine methods for precipitation and eight for temperatures.These statistical bias correction methods are designed to adjust the statistical properties of the model output to match with those in observations.Methods based on quantile mapping (i.e., eqm, eqm, gpqm, pqm and qdm) aim to match the quantiles of the data to correct biases.Quantile mapping uses cumulative distribution functions (CDFs) from both observed and model data to transfer the model's CDF to match that of the observed data using Equation (3) (Piani et al., 2010;Themeßl et al., 2012): where x and y are the observed and bias corrected data, respectively; F model is the CDFs of the model data, and F − 1 obs is the inverse of the CDFs of the observed data.While quantile mapping methods share this common concept, there are differences in details.For example, the eqm is a non-parametric method using empirical CDFs, whereas the gpqm and the pqm are the parametric methods employing probability distributions (e.g., generalized Pareto, gamma, and Gaussian) for the CDFs.The dqm initially removes a long-term linear trend before applying standard quantile mapping to detrended data (Cannon et al., 2015); the qdm is similar to the dqm but it preserves the modelprojected relative changes and apply quantile mapping to correct the remaining biases (Ansari et al., 2022).Other methods (i.e., loci, mva, ptr, scaling, and variance) employ linear transformation or scaling factors to match the statistical properties (i.e., mean, variance or standard deviation) for bias correction (Cannon et al., 2015;Fang et al., 2015;Teutschbein & Seibert, 2012).Specifically applicable to precipitation, the loci use the wet-day frequency to obtain a precipitation threshold, and a scaling factor is derived from the wet-day intensities of both model and observed data (Schmidli et al., 2006).The ptr uses the exponential form for non-linear correction to adjust the mean and variance of model data (Leander & Buishand, 2007).Several methods operate relying on simple linear scaling and transformation (scaling uses mean only, while the mva and the variance use mean and standard deviation).We utilize functions implemented in the downscaleR R package developed by the Santander Meteorology Group (Bedia et al., 2020).
In performing our analyses, we start by matching the spatial resolution of the GCM outputs to PRISM (4-km) and then prepare the calibration and validation sets.The calibration set is generated by using a time window of ±15 days for a given day in the historical period.For instance, for the ith day out of the total of 365 calendar days, we collect the candidate days within a time window of 31 days (from i − 15th day to i + 15th day) that represents the day, similar to Michalek et al. (2023) (see Figure S1a).This process is repeated for each historical year, allowing us to collect the candidate days for the ith day from all historical years.Once the calibration sets are prepared, we perform leave-one-out cross-validation (LooCV) for each calendar day to select the bestperforming bias correction method.Within each calibration set, we withhold the data from one year (i.e., test) and train the different bias correction methods on all the remaining years (i.e., training); we repeat this process for all the available years.We provide a schematic of the LooCV in Figure S1b.We also apply the Kolmogorov-Smirnov (KS) test to compare the performance of bias correction methods based on testing whether there are statistically significant differences between the distributions from PRISM and the bias-corrected CMIP6 outputs.
We set the significance level to 5% and apply the false discovery rate (FDR) to conduct multiple comparisons by the adjusted p-values from all grids in Arizona.By doing this, we select the best-performing bias correction method as the one with the highest percentage of pixels in Arizona with FDR-adjusted p-values above 0.05.Finally, we employ the best-performing method and get the bias-corrected CMIP6 outputs for both historical and future periods in the validation set.

| Assessment of projected changes
We focus on examining the projected changes at the annual and seasonal scales for mean and extreme values.We consider the four calendar seasons: winter (i.e., December-January-February: DJF), spring (March-April-May: MAM), summer (June-July-August: JJA) and fall (September-October-November: SON).For annual and seasonal means, we calculate the absolute amount of change in the future compared to the historical data.In the case of extremes, we obtain extreme thresholds using the 99th percentile of the daily climate variables from all observations (i.e., for all years and seasons); then we count the number of days above or below the thresholds (Zhang et al., 2011).In addition, we employ Cooling Degree Days (CDD) as an indicator to capture the potential social impact of high temperatures (Meehl et al., 2000;Ramon et al., 2020;Spinoni et al., 2018).CDD is defined as the cumulative sum of the difference between the average daily temperature and a base temperature of 65 F (18.3 C), with an increase in CDD being a proxy for an increase in energy consumption.Note that we calculate the CDD based on Fahrenheit to account for the original definition.For the analysis of wet bulb temperature, we define the threshold of 25 C for extreme wet bulb temperature based on historical records throughout Arizona, consistent with previous research (Raymond et al., 2017).

| Evaluation of the different biascorrection and downscaling techniques
We compare the performance of the LooCV results to select the best-performing bias correction method in Arizona. Figure 1 shows the percentage of pixels in Arizona with FDR-adjusted p-values above 0.05 from the LooCV results with respect to the KS test, for the seasonal mean (left column) and seasonal maximum (right column) and the three climate variables.The percentages displayed in Figure 1 are averaged across all CMIP6 models and pixels in Arizona.The results for individual CMIP6 models are provided in Figures S2 and S3.As shown in Figure 1, the performance of the different bias correction techniques varies by season, technique and climate quantity.Focusing first on precipitation, only eqm and pqm perform well in terms of bias correction across Arizona for both seasonal mean and maximum quantities, with slightly lower but still acceptable percentages in the summer.For temperatures, techniques such as gpqm and mva perform well compared to eqm and pqm, especially for seasonal maxima.However, only eqm keeps the same level of performance across all seasonal quantities, while this is not the case for other approaches.The results for all seasons and individual CMIP6 models are shown in Figures S2  and S3, highlighting the consistently good performance of eqm across individual CMIP6 models and seasons for the three climate variables.In summary, our results show that 'eqm' consistently outperforms the other methods for the three climate variables in terms of not only seasonal mean but also seasonal maxima across all CMIP6 models and seasons, highlighting that while this nonparametric technique is known for its simplicity, it performs well in reducing bias at a daily scale.Therefore, we select the 'eqm' as the method we use in the rest of this article for the three climate variables.

| Projected changes in annual quantities
Figure 2 shows the CMIP6 ensemble for annual total precipitation, annual mean air temperature and annual mean wet bulb temperature from the eqm-corrected climate variables.Based on the historical runs, the transition zone including the Mogollon Rim exhibits the largest precipitation values, with precipitation over 600 mm per year; other parts of the state, especially in the southwest, are much drier, with precipitation below 200 mm per year.For both air and wet bulb temperatures, the southwest is warmer than the northeastern part of the state, and the general difference between those two regions is about 5 C or more for both air and wet bulb temperatures.For the four future scenarios, annual total precipitation shows more precipitation around the transition zone as emission levels increase, but there are no large differences between the near and far futures.On the other hand, annual air and wet bulb temperatures exhibit higher values in the future than in the historical runs as we move from SSP126 to SSP585, and this is particularly true as we move towards the end of the 21st century.Here we examine the projected changes in each climate variable, and assess their changes in the middle and end of the 21st century under four SSP scenarios.To quantify the variability of CMIP6 models, we calculate the coefficient of variation (CV; computed as the ratio between the standard deviation and the mean) for the annual total precipitation, annual mean air temperature and annual mean wet bulb temperature (Figure S4).For the historical period, the CV values throughout the three F I G U R E 1 Percentage of pixels in Arizona with FDRadjusted p-values above 0.05 from the LooCV results, for the seasonal mean (left column) and seasonal maximum (right column) on the three climate variables (pr: precipitation; tas: air temperature; wbt: wet bulb temperature).Note that the percentage displayed are averaged of all CMIP6 models and pixels in Arizona.The first row ('original') is the result from raw CMIP6 outputs, and the rest of the rows represents the results for different bias correction techniques.The higher the percentage, the better the overall agreement (based on the Kolmogorov-Smirnov test) between the observations and the GCM outputs across Arizona.In each panel, 'DJF' refers to the months of December-January-February, 'MAM' to March-April-May, 'JJA' to June-July-August and 'SON' to September-October-November.climate variables are generally less than 5%, with slightly larger values for precipitation.Compared to the historical period, the CV for the projections tends to increase as we move towards the end of the century and for increasing emission scenarios, especially for precipitation.These results indicate that the standard deviation increases at a faster rate compared to the mean, pointing to a more rapid increase in variability than the magnitude of these annual quantities.
Figure 3 illustrates the differences in total precipitation between historical runs and the two future projections at the annual and seasonal scales under the four SSP scenarios.The annual total precipitation and the fractional contribution of seasonal precipitation to the annual totals are provided in Figure S5.Overall, Arizona is projected to become wetter, especially in the northern half of the state, as the annual total precipitation increases with increasing emission levels.However, this annual increase is not uniform across seasons, with spring projected to become drier while the other seasons (i.e., DJF, JJA and SON) wetter.This difference among seasons is exacerbated by increasing emissions, especially towards the end of the 21st century.Our findings are consistent with existing studies on seasonal differences in total precipitation in Arizona, particularly in relation to atmospheric circulation.Two major phenomena affecting Arizona's precipitation patterns are: (1) the NAM, which is a seasonal change in wind that brings moisture from the Pacific Ocean and the Gulf of California into the U.S. Southwest, especially in the summer (Adams & Comrie, 1997;Jana et al., 2018;Prein et al., 2019); and (2) the ARs (narrow bands in the atmosphere that transport moist air, leading to intense rainfall events) in the fall and winter (Demaria et al., 2017;Lavers & Villarini, 2015;Rutz & Steenburgh, 2012).Furthermore, increasing precipitation in winter aligns with Singh et al. (2018) who found that northeast Arizona is projected to experience an increase in extreme precipitation events due to the seasonal pattern of landfalling ARs in Arizona with the orographic effect.
The differences in mean air temperature between the historical runs and the two future projections under four SSP scenarios are shown in Figure 4.The air temperature is projected to increase by 2 C in the near future regardless of SSP scenario; however, the picture changes as we move towards the end of the 21st century, where the magnitude of these increases is larger and dependent on the scenarios, in agreement with previous studies (Almazroui et al., 2021;Balling Jr et al., 1990;Coppola et al., 2021).The annual and seasonal mean air and wet bulb temperatures for the historical runs and two future projections under four SSP scenarios are respectively presented in Figures S6 and S7.The mean air temperatures increase with increasing emissions regardless of seasons, indicating that the impact of climate change on temperature is dominant all year round.This uniform increase is different from what was discussed for precipitation, for which the difference between dry and wet seasons becomes more apparent.Similar patterns are observed in the differences in mean wet bulb temperature, as presented in Figure S8.Interestingly, we find that the wet bulb temperature exhibits a larger increase during the fall in the far future at the highest SSP scenario compared to the other seasons, suggesting that heat stress impacts may be felt even beyond the summer season.This result can be attributed to the interplay of complex climate dynamics, combining the effects of the NAM in September and landfalling ARs.The increased evaporation resulting from the NAM from June to September has the potential to further elevate air temperatures, and it can lead to overall warmer and wetter conditions in the fall.
Furthermore, we calculate CDD for historical and two future periods under four SSP scenarios to evaluate F I G U R E 3 Difference in total precipitation between historical runs and two future projections (Mid-of-Century and End-of-Century) under four SSP scenarios (SSP126, 245, 370 and 585).For each time slice, the first column represents the results for annual scale (legend on the left), while columns 2-5 represent the seasonal (DJF: December-January-February; MAM: March-April-May; JJA: June-July-August; SON: September-October-November) results (legend on the right).
the expected energy demand based on air temperature (Figure S9).We present the difference in CDD between the historical period and two time slices in Figure 5, highlighting the projected increasing power demand for cooling purposes, especially for southwest Arizona.As we move towards the end of the 21st century, a significant increase in CDD is projected across most seasons due to increasing emissions, with the exception of the winter.

| Projected changes in extremes
To investigate the projected changes in extremes, we examine the annual maxima of each climate variable (i.e., annual maximum precipitation, annual maximum air temperature and annual maximum wet bulb temperature) from the eqm-corrected outputs (Figure 6).The overall trend in extremes follows a similar pattern to what was observed when aggregated to the annual scale in Figure 2. Historically, extreme precipitation tends to be concentrated around the transition zones along the Mogollon Rim.In the future, extreme precipitation tends to be more intense across all emission levels, especially on the upper side of the Mogollon Rim.Southwest Arizona exhibits the highest temperatures, which are projected to increase during the 21st century and for increasing emissions: under the highest emission scenario (i.e., SSP585), maximum temperatures are projected to reach up to 45 C for air temperature and 32 C for wet bulb temperatures.We calculate the CV value for extreme climates (Figure S10).Across Arizona, the values of the CV for the extremes during the historical period are comparable to what we found for the annual quantities (compare Figures S4 and S10).However, the picture is different in terms of projections, with the largest differences being for precipitation, while temperature and wet bulb temperature exhibit conditions comparable to the historical past.Therefore, these results indicate that the projected changes in the variability and magnitude of temperature-based extremes are of the same level in the future compared to the historical period; this is not the case for precipitation extremes, for which we project a larger increase in variability compared to the mean during the 21st century.
We have also performed more detailed analyses of extreme precipitation characteristics, including the occurrence of dry days (i.e., the number of days with less than 0.01 mm of daily precipitation) (Figure S11).The percentage of dry days in historical runs highlights the prevalence of dry days in southwest Arizona, with large seasonal and regional variations across the state.In the spring, for instance, southwest Arizona has more than 90% of dry days, and the areas with >90% of dry days expand for increasing emission levels and towards the end of the century.Figure 7 shows the difference in the percentage of dry days between the historical period and the two time slices in the 21st century.In the middle of the century, the dry days are projected to change by ±2% across all seasons.At the end of the century, the dry days are projected to significantly increase, more so for increasing emissions, in the spring along Mogollon Rim, while the dry days in the summer are expected to decrease significantly in northern Arizona.
We analyse the percentage of extreme hot days that are above the 99th percentile of the daily air temperature distribution based on the historical runs to assess the projected change in extreme air temperature.Figure 8 shows the difference in the percentage of extreme hot days between historical runs and two future projections.On an annual scale, there are limited increases in extreme hot days in the near future, while they are projected to rapidly rise further into the future under increasing emission scenarios.On a seasonal scale, summer accounts for the majority of the increase, showing that the percentage of extreme hot days increases up to around 20% even in the near future under the highest emission level, suggesting that the extreme hot days are projected to become more frequent.Additionally, the fall season also shows a 10+% increase in extreme hot days under the highest emission levels in southern Arizona.
In addition to air temperature, we also examine the combined effects of heat and humidity by counting the number of days with the wet bulb temperature above 25 C.The difference between historical runs and future projections is presented in Figure 9. Similar to the patterns observed for extreme hot days, the majority of the increase in the number of days with extreme wet bulb temperature is expected during the summer, with a slight increase projected in the fall under the highest emission scenario in the far future.It is noteworthy that these increases are spreading from western Arizona towards the interior regions.Considering the heat extremes in Arizona, it is important to pay attention to the threshold of wet bulb temperature that could become a public health issue.

| SUMMARY AND CONCLUSIONS
We examined the projected changes in daily precipitation, air temperature and wet bulb temperature from the statistically downscaled and bias-corrected CMIP6 models for Arizona, considering two future periods and four SSP scenarios.Consequently, our results provided more localized information on the future climate conditions in Arizona at a daily and 4-km resolution.Among the statistical bias correction methods considered here, the 'eqm' method consistently outperformed the others across different conditions throughout the state.This performance was observed in both the seasonal mean and seasonal maximum of the three climate variables (i.e., precipitation, air temperature and wet bulb temperature) obtained from the CMIP6 model ensemble.These results highlight the reliability and robustness of the 'eqm' in adjusting biases in the projected climate dataset.
Our findings on the future changes of climate variables suggest that Arizona is projected to experience increased future precipitation in most seasons.However, it is noteworthy that the spring season is projected to become drier.In contrast to precipitation, Arizona is becoming highly vulnerable to heat stress from increasing emissions, especially in the far future, for the southwest region and the summer.Taken together, southwest Arizona is projected to become a hotspot due to the combined effects of rising temperature extremes and humidity.
While a wet bulb temperature above 35 C is widely known to be fatal (Coffel et al., 2017), the maximum wet bulb temperature observed in Arizona is much lower.Therefore, further investigation is needed to understand the implications of wet bulb temperature and their implications on human health because high fatality levels were recorded in the 2003 European and 2010 Russian heatwaves when the wet bulb temperature did not exceed 28 C (Raymond et al., 2020).An alternative approach to assessing heat stress is the wet bulb globe temperature (WBGT) index.The WBGT index is used to assess labour capacity in the presence of heat stress (Budd, 2008;Dunne et al., 2013;Kong & Huber, 2022) such as 'red flag' and 'black flag' days, which determine suspending all physical training and strenuous exercises to protect workers from heat-related illnesses and injuries.However, the CMIP6 outputs are limited to get the necessary variables to calculate the WBGT index, and this is beyond the scope of this study.
In conclusion, Arizona's climate is expected to experience a general increase in temperature and precipitation, particularly towards the end of twenty-first century under higher emission scenarios.Building on our findings in this study, we have provided insights from both scientific and engineering perspectives in terms of projected changes in energy demand and the associated public health and socio-economic impacts (Akadiri et al., 2020;Alola et al., 2019;Deroubaix et al., 2021).These outcomes contribute to a holistic understanding of the impacts of climate change in Arizona by providing a better quantification of future climate conditions to prepare for and mitigate the expected risks.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

F
I G U R E 2 Annual climates (i.e., annual total precipitation, annual mean air temperature and annual mean wet bulb temperature) from the bias-corrected climate variables using 'eqm'.The first row shows the annual climate from the historical runs, and the second through fifth rows show the annual climate for two future projections (Mid-Of-Century on the left and End-Of-Century on the right) under four SSP scenarios (SSP126, 245, 370 and 585).

F
I G U R E 4 Difference in mean air temperature between historical runs and two future projections (Mid-of-Century and End-of-Century) under four SSP scenarios (SSP126, 245, 370 and 585).For each time slice, the first column represents the results for annual scale, while columns 2-5 represent the seasonal (DJF: December-January-February; MAM: March-April-May; JJA: June-July-August; SON: September-October-November) results.

F
I G U R E 5 Difference in CDD between historical runs and two future projections (Mid-of-Century and End-of-Century) under four SSP scenarios (SSP126, 245, 370 and 585).For each time slice, the first column represents the results for annual scale (legend on the left), while columns 2-5 represent the seasonal (DJF: December-January-February; MAM: March-April-May; JJA: June-July-August; SON: September-October-November) results (legend on the right).

F
I G U R E 6 Extreme climates (i.e., annual maximum precipitation, annual maximum air temperature and annual maximum wet bulb temperature) from the bias-corrected climate variables using 'eqm'.The first row shows the annual climate from the historical runs, and the second through fifth rows show the annual climate for two future projections (Mid-Of-Century on the left and End-Of-Century on the right) under four SSP scenarios (SSP126, 245, 370 and 585).

F
I G U R E 7 Difference in the percentage of dry days between historical runs and two future projections (Mid-of-Century and End-of-Century) under four SSP scenarios (SSP126, 245, 370 and 585).For each time slice, the first column represents the results for annual scale (legend on the left), while columns 2-5 represent the seasonal (DJF: December-January-February; MAM: March-April-May; JJA: June-July-August; SON: September-October-November) results (legend on the right).

F
I G U R E 8 Difference in the percentage of extreme hot days (99th percentile) between historical runs and two future projections (Midof-Century and End-of-Century) under four SSP scenarios (SSP126, 245, 370 and 585).For each time slice, the first column represents the results for annual scale (legend on the left), while columns 2-5 represent the seasonal (DJF: December-January-February; MAM: March-April-May; JJA: June-July-August; SON: September-October-November) results (legend on the right).
models used in this study.
Note: 'O' refers to a model that is available, while 'NA' to one that is not.In terms of variables, 'pr' refers to precipitation, 'tas' to air temperature and 'hurs' to relative humidity.