Recent atmospheric changes and future projections along the Saudi Arabian Red Sea Coast

Recent and future climate diagrams (surface air temperature, surface relative humidity, surface wind, and mean sea level pressure) for the Saudi Arabian Red Sea Coast are analysed based on hourly observations (2016–2020) and hourly ERA5 data (1979–2020) with daily GFDL mini-ensemble means (2006–2100). Moreover, GFDL mini-ensemble means are calculated based on the results of three GFDL simulations (GFDL-CM3, GFDL-ESM2M, and GFDL-ESM2G). Observation data are employed to describe the short-term current weather variability. However, ERA5 data are considered to study the long-term current weather variability after bias removal via a comparison to observations. Finally, a bias correction statistical model was developed by matching the cumulative distribution functions (CDFs) of corrected ERA5 and mini-ensemble mean data over 15 years (2006–2020). The obtained local statistic were used to statically downscale GFDL mini-ensemble means to study the future uncertainty in the atmospheric parameters studied. There occurred significant spatial variability across the study area, especially regarding the surface air temperature and relative humidity, based on monthly analysis of both observation and ERA5 data. Moreover, the results indicated that the ERA5 data suitably describe Tabuk, Jeddah and Jizan weather conditions with a marked spatial variability. The best performance of ERA5 surface air temperature and relative humidity (surface wind speed and sea level pressure) data was detected in Tabuk (Jeddah). These data for the Saudi Arabian Red Sea coast, 1979–2020, exhibit significant positive trends of the surface air temperature and surface wind speed and significant negative trends of the relative humidity and sea level pressure. The GFDL mini-ensemble mean projection result, up to 2100, contains a significant bias in the studied weather parameters. This is partly attributed to the coarse GFDL resolution (2° × 2°). After bias removal, the statistically downscaled simulations based on the GFDL mini-ensemble mean indicate that the climate in the study area will experience significant changes with a large range of uncertainty according to the considered scenario and regional variations.

The air temperature is considered one of the most important climate factors, as the daily activities of people are influenced by even a minor change in air temperature 13 . According to Maia-Silva et al. 14 , relative humidity (RH; calculated from the surface air and dew point temperatures) rise is one of the main climate challenges, as the increase in surface air temperature is correlated with a decrease in relative humidity. This coupling effect can be disastrous, especially during heat wave events. This may increase mortality/morbidity rates. However, wind fields and the progress associated with wind technology may reduce CO 2 emissions 15 . In general, changes in mean sea level pressure (SLP) could impose a notable effect on climate because SLP controls atmospheric circulation. Therefore, this factor influences wind, moisture movement, precipitation, and temperature variabilities.
In 2000, IPCC 16 introduced the Special Report on Emissions Scenarios (SRES), covering a range of greenhouse gas (GHG) emissions (A1B, A2, and B1 scenarios) under the Coupled Model Intercomparison Project, phase three (CMIP3). Meehl et al. 17 reported that based on the success of SRES scenarios, the IPCC elevated the CMIP3 to the CMIP5 in late 2008 and described new future scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5). These new future RCP scenarios comprise various RCP combinations. RCP denotes the representative concentration pathway, and the numbers indicate the assumed radiative forcing by the end of the twenty-first century.
The current research examines the current and future climate conditions along the Saudi Arabian Red Sea Coast. First, observation data were employed to describe the current short-term weather variability from 2006 to 2020. Second, the current research qualifies the use of ERA5 data in capturing the surface air temperature (T2m), relative humidity (RH), surface wind field and mean sea level pressure. Third, 42 years (1979-2020) of ERA5 data after bias removal were analysed to describe the current long-term climate. Finally, the future climate in the study area was projected via statistical analyses through the above four CMIP5 scenarios. The materials and methods adopted are presented in "Results" section, the results and discussion are provided in "Discussion and Conclusions" section, and the conclusions are outlined in "Future work" section.

Results
Observed data. Monthly time series. Monthly average time series for T2m, RH, WS 10 , and SLP based on hourly observation data (2016-2020) for Tabuk, Jeddah and Jizan are shown in Fig. 2.
In Tabuk, the T2m data revealed that January 2020 (10.16 °C) is the coldest month, while July 2017 is the warmest month (34.68 °C), as shown in Fig. 2a. Similarly, the RH data revealed that November 2018 exhibits the maximum monthly RH mean value (50.25%), while May 2019 attains the lowest monthly RH mean value (14.9%), as shown in Fig. 2b. In addition, the WS 10 data revealed that June 2020 is the windiest month (4.6 m s -1 ), while November 2017 is the calmest month (2.29 m s -1 ), as shown in Fig. 2c. Generally, the SLP data revealed that January 2016 attains the maximum monthly SLP value (1019.33 hPa), while July 2020 exhibits the lowest SLP value (999.74 hPa), as shown in Fig. 2d.
In Jeddah, the T2m data indicate that January 2020 (22.3 °C) is the coldest month, while August 2020 (34.2 °C) is the warmest month. Similarly, the RH data confirm that October 2017 (68.0%) attains the maximum monthly RH mean value, while May 2018 (44.5%) exhibits the lowest mean value. Moreover, the WS 10 data reveal that April 2017 (4.8 m s -1 ) is the windiest month, while October 2016 (2.5 m s -1 ) is the calmest month. Generally, the SLP data reveal that January 2020 (1016.7) attains the maximum monthly SLP value, while August 2017 (1001.7 hPa) exhibits the lowest monthly SLP value.
In Jizan, the T2m data confirmed that the maximum (monthly) average value was 35.05 °C (26.1 °C). The maximum (minimum) value occurred in June 2020 (January 2019). Similarly, the RH data confirmed that December 2019 (74.1%) attained the maximum monthly RH mean value, while May 2019 (55%) attained the lowest monthly RH mean value. Moreover, the WS 10 data revealed that the highest (lowest) average wind speed value is 4.2 m s -1 (2.7 m s -1 ). The windiest (calmest) month is July 2020 (January 2020). Generally, the SLP data reveal that December 2017 (1015 hPa) attains the maximum monthly SLP value, while July 2020 (1002.0 hPa) achieves the lowest monthly SLP value.
Generally, T2m increased from Jizan to Tabuk. The highest temperatures were close among the three studied cities, while the lowest temperature markedly occurred in Tabuk (11 °C) rather than in Jeddah (24 °C) and Jizan (27 °C). In contrast, RH increased from Jizan to Tabuk, and there was a significant difference between the maximum and minimum RH values among the three studied cities. There was no significant difference among Tabuk, Jeddah and Jizan regarding the WS 10 regime. Regarding SLP, Jizan and Jeddah exhibited similar behaviour. However, SLP in Tabuk indicated markedly higher values than those in Jizan/Tabuk. Annual, monthly, and hourly cycles. In Jeddah and Jizan, the hottest (coldest) year was 2016 (2020). However, in Tabuk, the hottest (coldest) year was 2018 (2019). The highest/lowest annual average RH values in the study area changed from city to city, similar to WS 10 . However, the SLP annual pattern in the study area revealed similarities between Tabuk and Jizan, where the maximum (minimum) annual SLP value occurred in 2016 (2020). In Jeddah, the maximum (minimum) annual value occurred in 2019 (2017), as indicated in Table 1.
In terms of the annual cycle (climate monthly average), as described in Table 1 and Fig. 3, the maximum T2m value occurred in July for Tabuk and Jeddah and occurred in June over Jizan. In contrast, the minimum T2m value occurred in January in the three studied cities. There was no common pattern among the three studied cites regarding RH and WS 10 . In contrast, the SLP annual cycles were similar among Tabuk, Jeddah and Jizan, as the maximum (minimum) values occurred in January (July).
The daily T2m cycle (Table 1; Fig. 4) amplitude in Tabuk (13.5 °C) is much higher than that in Jeddah (7.1 °C) and Jizan (5.3 °C). Similarly, the amplitude of the RH daily cycle in Tabuk (26.7%) is much higher than that in Jeddah (25.2%) and Jizan (17%). The amplitude of the WS 10 daily cycle reaches its maximum value (4.7 m s −1 ) in Jizan followed by ( www.nature.com/scientificreports/ Tabuk. The daily SLP cycle cannot be analysed in Jizan and Jeddah based on the available observed data due to a large number of missing observations. In Tabuk, the annual WS 10 cycle reveals a direct correlation (n = 12, R = 0.69) with T2m and inverse correlations (n = 12, R = − 0.88) with RH and (n = 12, R = − 0.76) SLP. Similar to Tabuk, the annual WS 10 cycle in Jizan indicates a significant direct correlation with T2m and significant inverse correlations with RH and SLP. In contrast, the annual WS 10 cycle in Jeddah reveals a non-significant correlation with T2m and SLP, while WS 10 achieves a significant inverse correlation with RH (n = 12, R = − 0.76). This indicates that the four parameters studied are co-dependent only in Tabuk and Jizan. In addition, in Tabuk, RH monthly maximum values occurred in December. After one month (in January), SLP reached its monthly maximum values. After 4 months (in May), WS 10 reached its monthly maximum values. After 2 months (in July), T2m attained its monthly maximum values,   Fig. 3. Similarly, in Jizan, RH monthly maximum values occurred in December. After one month (in January), SLP reached its monthly maximum values. After 6 months (in June), T2m attained its monthly maximum values. After one month (during July), WS 10 reached its monthly maximum values. In contrast, regarding the daily cycle in Tabuk, WS 10 attains a direct correlation (n = 24, R = 0.76) with T2m and an inverse correlation (n = 24, R = − 0.85) with RH. Moreover, T2m attains an inverse correlation (n = 24, R = − 0.97) with RH. Moreover, SLP exhibits a weak correlation with WS 10 , T2m, and RH, indicating that the daily cycle, which distinguishes WS 10 , T2m, and RH, is not significant for SLP (SLP is distinguished by a 3-h cycle),    Fig. 4. In Jizan and Jeddah, the daily RH, T2m and WS 10 cycles indicate a significant correlation (n = 24, R > 0.9 between any two parameters. This confirms the previous finding whereby T2m, RH and WS 10 are co-dependent in the study area. Due to the high percentage of missing SLP data, the hourly correlation between SLP and the studied parameters could not be determined. 10-m height wind direction (WD10). The prevailing annual wind direction originated from the northwest (12.2% of the time) followed by the south-southwest direction (10.2% of the time) in Tabuk, as shown in Fig. 5. However, the prevailing annual wind direction originated from the north (19.1% of the time), followed by both the west-northwest and north-northwest directions (14.7% of the time each). In Jizan, the wind rose diagram shows a different pattern than that in Jeddah and Tabuk, where the prevailing annual wind direction originated from the west (15.6% of the time) followed by both the south and south-southwest directions (10.3% of the time each).    Fig. 6, in November, December, January, and February, the prevailing monthly wind direction originated from the east for 15%, 16%, 12%, and 14%, respectively, of the time. In March, the prevailing monthly wind direction originated from the south-southwest direction (11% of the time). In April, June, July, and August, the prevailing monthly wind direction originated from the northwest direction at frequencies of 13%, 19%, 16%, and 19%, respectively, of the time. In May, the prevailing monthly wind direction originated from the northwest and south-southwest directions (11% of the time each) at equal frequencies. In August, the prevailing monthly wind direction originated from the west and south-southwest directions at equal frequencies (11% of the time each). In Jeddah (Fig. 7), from October until April, the prevailing monthly wind direction originated from the north. In the months from May to August, the prevailing monthly wind direction originated from the west-northwest.
In Jizan (Fig. 8), in January, February and March, the prevailing monthly wind direction originated from the south. In the months from April to September, the prevailing monthly wind direction originated from the west. From October to December, the prevailing monthly wind direction originated from the east.

ERA5. ERA5 verification.
To evaluate the feasibility of using ERA5 data in describing the selected atmospheric parameters in Tabuk, Jeddah and Jizan, a comparison was carried out between ERA5 data and observations covering the observation period from 2016 to 2020 (Table 2; Figs. 6,7,8).
Generally, the ERA5 data closely matched the observations in the studied cities. With ERA5 data, T2m and WS 10 were underestimated, while RH and SLP were overestimated, as indicated in Table 2. Moreover, the applied statistical tests (t and f tests) indicated that the observed and ERA5 T2m, RH, WS 10 , and SLP values originated www.nature.com/scientificreports/ from two equal distributions based on the mean and variance at a 99% significance level. In addition, based on Figs. 6, 7 and 8, it is clear that the observed wind direction agrees with the ERA5-based wind direction. In general, with the ERA5 data, the simulation efficiency in Jeddah was higher than that in Tabuk, with the lowest simulation efficiency in Jizan.

ERA5 bias correction.
To remove the bias in the ERA5 data, CDF bias correction was applied to match the CDF of the observations to that of the ERA5 data from 2016 to 2020 (Fig. 9). This strategy conserves the nature of the data while maintaining the same correlation values and adjusts the bias to zero. This strategy was applied to the long-term ERA5 data to obtain corrected ERA5 data (C_ERA5) for Tabuk, Jeddah and Jizan. Figure 9 confirms the current findings that the ERA5 data more reasonably simulate the current atmospheric parameters in Jeddah and Tabuk than is achieved in Jizan.
C_ERA5 statistical analyses. Annual mean and trend analyses. The C_ERA5 data (1979-2020) revealed that the annual average T2m, RH and SLP values across the study area exhibits significant spatial variation (Table 3), where T2m and RH increased meridionally from north to south, while SLP decreased meridionally from north to south. The annual long-term WS 10 average value was very close between Tabuk and Jizan. However, Jeddah was much windier. The C_ERA5 data (1979-2020) confirmed a significant spatial long-term trend in T2m, RH and SLP. The longterm WS 10 trend in Tabuk and Jeddah, however, was less notable than that in Jizan. The study area experiences a positive monotonic warming trend in the three studied cites. Similarly, WS 10 attained a positive monotonic trend in the study area. In contrast, RH and SLP attained negative monotonic trends in the study area (Table 3). www.nature.com/scientificreports/ In detail, the warming trend in Tabuk is more intense than that in Jizan, indicating that the T2 m difference across the study area may decrease in the future. Similarly, RH attains a much more significant negative trend in Jizan than that in Tabuk, indicating that the spatial RH variation in the study area may decrease in the future. In addition, the negative long-term SLP trend reaches its highest value in Jeddah, followed by Jizan and Tabuk, indicating that the SLP difference between Tabuk and Jeddah may decrease in the future, while the difference between Jeddah and Jizan may increase in the future. In Tabuk, the maximum RH value (= 93.9%) was recorded on 30 November 1989 at 02:00 GMT. In contrast, the minimum RH value (= 0.1%) in Tabuk was recorded on 24 April 1994 at 16:00 GMT.
In Tabuk  Jeddah's highest SLP value, 1024.9 hPa, was recorded on 26 February 1992 at 08:00 and 07:00 GMT. In contrast, Jeddah's lowest SLP value reached 995.7 hPa and was recorded on 13 August 1998 at 0:00 GMT, virtually on par with the value of 995.8 hPa on 12 August 1998 at 0:00 GMT.
Jizan In Jizan, the highest temperature of 47.2 °C was recorded on 11 July 1995 at 11:00 and 12:00 GMT, while the lowest temperature of 0.1 °C in Jizan was recorded on 7 February 1993 at 04:00 and 03:00 GMT.
In detail, the first PC, which is responsible for 62.71% and 45% of the parameter variance in Tabuk and Jeddah, respectively, indicates a strong correlation with three of the examined variables. This first PC decreases with decreasing T2m but increases with increasing RH and SLP. This indicates that T2m, RH, and SLP are highly Table 3. Long-term annual mean and trend analyses of the corrected ERA5 weather variables in Tabuk, Jeddah and Jizan from 1979 to 2020. The nonparametric Mann-Kendall test is used to detect monotonic trends in the corrected ERA5 data to examine whether the C_ERA5 data follow a significant (monotonic) trend.  10 . In Jeddah and Jizan, the second PC is responsible for 39% and 28%, respectively, of the parameter variance. The second PC attains a strong correlation with only two of the examined variables, RH and WS 10 , indicating the strongest co-dependence between RH and WS 10 .
Statistical downscaling for future projection. In this section, the results of the GFDL mini-ensemble mean simulations under the different RCP scenarios are investigated for Tas, RH, WS 10 , and SLP. Table 5. In addition, the GFDL mini-ensemble mean underestimates (overestimates) RH in Jeddah and Jizan (Tabuk). Moreover, the GFDL mini-ensemble mean underestimates (overestimates) WS 10 in Tabuk and Jeddah (Jizan). In general, the GFDL mini-ensemble mean overestimates SLP in the three studied cities.

GFDL model bias correction over the control period, 2006-2020. The GFDL mini-ensemble mean underestimates (overestimates) Tas in Tabuk and Jizan (Jeddah), as indicated in
To overcome the above underestimation/overestimation of the GFDL mini-ensemble mean that is incomparable to the C_ERA5 data, a simple statistical model was developed by matching the CDF of the C_ERA5 data to that of the GFDL mini-ensemble mean data over the control period (2006-2020) under the different RCP scenarios, as shown in Fig. 11. Figure 11 shows the bias correction for only RCP2.6. However, the bias correction for RCP4.5, RCP6, and RCP8.5 is not shown due to the similarity with the RCP2.6 effect over the control period.     Generally, the future warming uncertainty is 2.14 °C, where 86% (14%) of the uncertainty is associated with scenario design (regional variation). In regard to the future RH, the uncertainty is 2.3%, and 25% of the uncertainty is associated with scenario design, while 75% is associated with regional variation. In regard to the future WS 10 , the uncertainty is 0.23 m s −1 , where 21% of the uncertainty is associated with scenario design, and 79% is associated with regional variation. Regarding the future SLP, the uncertainty is 0.56 hPa, and 93% of the uncertainty is associated with scenario design, while only 7% is associated with regional variation.

Discussion and conclusions
As mentioned above, a limited number of scientific studies has been conducted on the climate in the study area. The current paper presented an overview to bridge the present gap in climate knowledge of the Saudi Arabian Red Sea coast, especially regarding future projections.
Based on hourly observed time series for T2m, RH, WS 10 and SLP, the present research studied the short-term atmospheric changes in Tabuk, Jeddah and Jizan from 2016 to 2020. The observed time series was employed to remove the bias in ERA5 data and calculate C_ERRA5 data for the three studied cities from 1979 to 2020 via cumulative distribution functions (CDFs). The long-term C_ERA5 data were analysed statistically with the following techniques: trend analysis, historical days, probability of occurrence, and PCA. The C_ERRA5 data were employed to downscale GFDL mini-ensemble mean simulations and ensure relevance to Tabuk, Jeddah and Jizan with the CDF strategy. Finally, the S_D_ GFDL mini-ensemble mean was considered to analyse the uncertainty in the future Tabuk climate.
The results based on the observed data indicated that the air temperature increased meridionally from north (Tabuk) to south (Jizan), partly due to the amount of net absorbed solar energy. The warmest month was July in Tabuk and Jeddah, with average values of 33.3 °C and 33.95 °C, respectively. However, in Jizan, the hottest month occurred one month earlier (June), with an average value of 34.7 °C. The coldest month was January in Tabuk, Jeddah and Jizan, with average values of 11.3 °C, 23.87 °C and 26.9 °C, respectively. The lowest surface air temperature that was 12.5 °C (15.6 °C) lower than that in Jeddah (Jizan) indicated a much more intensive cooling process during cold months in Tabuk stemming from its mountainous nature. Similarly, the RH increased meridionally from Tabuk to Jizan. In contrast, there were no consistent patterns of meridional changes in the study area regarding WS 10 and SLP. In addition, the prevailing annual wind direction was NW (12.2% of the time) www.nature.com/scientificreports/ in Tabuk, N (19.11% of the time) in Jeddah, and W (15.6% of the time) in Jizan. Moreover, the prevailing wind direction indicated significant monthly variability in the study area. Generally, the monthly T2m/RH variability was lower in Jeddah and Jizan than that in Tabuk, partially reflecting the importance of studying the dynamic mechanism that shapes the temperature variability near the Earth's surface and upper layer. Generally, the monthly T2m/SLP patterns were very close among the 3 studied cities. However, the monthly RH pattern in Tabuk was much closer to that in Jizan than to that in Jeddah. In the same manner, the monthly RH pattern in Tabuk was much closer to that in Jizan than to that in Jeddah.
The ERA5 data described the real atmospheric characteristics of the Saudi Arabian Red Sea Coast in a reasonable way, as CDF bias correction against real observations was insignificant in 66% of the study cases. This high performance indicates significant spatial variation, reaching a maximum performance in Jeddah and a lowest performance in Jizan. In detail, the highest match for the prevailing wind direction between the observed and ERA5 data was detected in Jeddah (Shift = 0), while in Tabuk, the shift was 22.5°, and finally, in Jizan, the shift was 45°. In contrast, ERA5 describes the present observation with a higher accuracy in Tabuk regarding T2m and Rh. However, a higher ERA5-based accuracy regarding WS 10 and SLP was detected in Jeddah. This variability between the observed and ERA5 data initiates a future discussion regarding the reliability of ERA5 data versus land use and topography data in the study area.
In Tabuk, the present calculated long-term T2m annual mean (22.4 ± 9.3 °C) is approximately 1 °C higher than that modelled by 13 . This difference may be related to the different time spans. However, the current calculation of the long-term WS 10 annual mean (3.3 ± 2.2 m s −1 ) is close to that modelled by 14 . In contrast, the C_ERA5 annual wind speed is approximately 50% lower than that previously calculated by 15 . This significant difference mainly occurs because 15 ) the calculations were based on climate data, while the current calculations were based on hourly data.
The annual mean (from 1979 to 2020) of the C_ERA5 data in the study area confirms the occurrence of an 8.3 °C spatial T2m variation ranging from 22.4 to 30.7 °C, a 37.2% spatial RH variation ranging from 29.1 to 66.3%, a 0.4 m s −1 spatial WS 10 variation ranging from 3.2 to 3.7 m s −1 , and a 1.4 hPa spatial variation ranging from 1008.5 to 1009.9 hPa.
In addition, the C_ERA5 data revealed that T2m and WS 10 attained a significant positive trend. However, RH and SLP exhibited a significant negative trend. The significance of the calculated trends was tested with the Mann-Kendall test. In detail, the Saudi Arabian Red Sea coast exhibits a significant spatial warming trend (from 1979 to 2020) for T2m (0.33-0.49 °C decade −1 ), RH (− 0.34 to − 0.45% decade −1 ), WS 10 (0.01-0.02 m s −1 decade −1 ) and SLP (− 0.04 to − 0.11 hPa decade −1 ). However, the Saudi Arabian Red Sea Coast exhibits a significant spatial variation in the T2m trend, and RH and SLP may lead to a decrease in the spatial variation in the annual average values. The T2m and RH values in the south remain higher than those in the north, and the SLP value in the south remains lower than that in the north.
A historical warming day in the study area was recorded in Jeddah (55.7 °C) on 5 June 2013 at 9:00 GMT, while a historical cold day was recorded in Tabuk  In general, the probability occurrence patterns indicate a significant spatial variation in each studied parameter. In addition, the C_ERA5 long-term records (1979-2020) confirm that T2m, RH and SLP in the study area are rarely below 3.7 °C, 1%, and 997 hPa, respectively, or above 41.1 °C, 85.9% and 1022.5 hPa, respectively. Moreover, WS 10 rarely exceeds 8.1 m s −1 .
At least 75% of the variance in the studied parameters can be explained by only two principal components in the study area, much more markedly in Tabuk (85%). In Tabuk and Jeddah, T2m, RH and SLP are closely co-dependent. However, in Jizan, only T2m and SLP are closely co-dependent.
The comparison between the GFDL mini-ensemble mean and corrected ERA5 data over the control period (2006-2020) confirms that the GFDL mini-ensemble mean in general underestimates Tas, RH and WS 10 with average values of − 2.4 °C, − 8.4%, and − 0.2 m s −1 , respectively. Moreover, the GFDL mini-ensemble mean generally overestimates SLP with an average value of 4.5 hPa. Thus, the GFDL mini-ensemble mean is applied to project the future climate in the study area after bias correction with the CDF technique.
The statistical downscaling results indicate that the study area will experience significant warming trends and significant negative RH/SLP trends. The future WS 10 exhibits a wide uncertainty range from a positive to a negative trend. Moreover, the emission assumptions (according to the RCP used) combined with regional variation impose a significant effect on the uncertainty in all the studied parameters. The uncertainty related to the simulation performed is negligible, as the research considers only GFDL mini-ensemble means.
To support future management in Tabuk, statistical downscaling was applied to improve the GFDL miniensemble mean results, which exhibits a coarse resolution (2° × 2°), to ensure applicability in Tabuk. The S_D_ GFDL mini-ensemble mean simulations over the current century revealed a significant increase in Tas and WS 10 with decreasing RH and SLP.

Future work
The current research provides the first scientific analyses based on the current and future climate conditions in the Saudi Arabian Red Sea Coast. The main result of the current paper confirmed that the study area will face dramatic climatic changes, and the responses in terms of Tas, RH, WS 10 , and SLP are now better identified. This information can comprise a powerful database needed to improve the future vision of Tabuk. In future studies, the authors will expand the current work to study new stations distributed throughout the KSA to obtain a full picture of the future uncertainties in the KSA. Moreover, the use of a regional climate model as a scientific tool for dynamical downscaling merits our attention in future work.
Finally, the most important conclusion of the current research addresses the project statistics of the studied climatic parameters up to 2100 through the different RCP scenarios. The projected climate uncertainties are likely to yield many negative socioeconomic impacts, especially in agriculture, water demand, and health. Thus, scientists in the fields of agriculture, medicine, and climate change must work with specialists in the field of water resources and decision-makers to improve climate policies in the KSA to identify innovative ways to turn the expected future climate challenges into social and economic opportunities.

Methods
This paper obtained observation data to describe the short-term atmospheric variability in T2m, relative humidity (RH), and surface wind field components (eastward wind (U10), northward wind (V10) and SLP) in Tabuk, Jeddah and Jizan. Moreover, the long-term characteristics of the studied atmospheric parameters were analysed based on the ERA5 database after bias removal. Regarding the projection of future scenarios, statistical downscaling of the GFDL mini-ensemble mean was performed to study the future uncertainty in the study area.
ERA5 data, which are distributed by the Copernicus Climate Change Service (C3S) and produced by the European Center for Medium Range Weather Forecasts (ECMWF), were intended to mitigate earlier issues (e.g., ERA Interim and ERA40 data) and improve the study of atmospheric parameters 18,19 with a finer grid resolution of 0.25° × 0.25° and hourly data. These data, after a comparison to observations, were considered to analyse the long-term trends of the studied atmospheric variables. These data were further used for statistical downscaling. Methods. Analysis of the recent and future variabilities in T2m, relative humidity (RH), 10-m height wind speed (WS 10 ), wind direction (WD 10 ), and SLP requires three steps. The first step is to analyse the current shortterm atmospheric system from 2016 to 2020 using observation data. The second step concerns the ERA5 database (based on observations, remote sensing, and modelled information), where these data are employed after validation against observation data to analyse the current long-term climate system from 1979 to 2020. The third step is to conduct statistical downscaling to project the future climate from 2020 to 2100.
Generally, the relative humidity (RH) is calculated according to 23 based on T2m and d2m, as expressed in Eq. (1).
Observation. Statistical analysis (observed as short-term means at hourly, daily, and monthly scales) of T2m, RH, WS 10 , and SLP was performed to investigate the temporal variation. In addition, the wind rose method is applied to similarly describe the temporal variation in WD 10 .
The SLP data pertaining to Tabuk, Jeddah and Jizan contained notable missing data fractions of 19.41%, 19.37% and 19.67%, respectively. Thus, the monthly mean was calculated only if the hourly data covered at least 33% of the month. Missing data for T2m, RH, and WS 10 were very rare among the three stations.
Observed short-term hourly means for the parameters studied were considered to compute average values of T2m at hourly intervals throughout the observation period. Similarly, observed short-term monthly means were computed by averaging all the hourly data on a monthly basis during the observation period. www.nature.com/scientificreports/ ERA5. Direct comparisons of the hourly observed data and ERA5 data were conducted to assess the efficiency of ERA5 data in Tabuk, Jeddah and Jizan. Moreover, f and t tests at a 95% confidence level were executed to examine whether the ERA5 and observation data exhibited similar means and variances (from the same population). Moreover, the ERA5 data were subjected to CDF bias correction between the ERA5 and observation data from 2016 to 2020. The present strategy for bias correction is to match the CDF of the observations to that of the ERA5 data. Many researchers have applied this strategy. Anagnostou et al. 24 implemented the CDF strategy to statistically adjust satellite microwaves for monthly rainfall estimates. Furthermore, Wood et al. 25 used the CDF technique for long-range hydrologic forecasting. Reichle and Koster 26 employed this strategy to match the CDF between satellite retrievals and soil moisture estimates.
Linear trend analysis of the ERA5 data after correction (hereafter, C_ERA5) from 1979 to 2020 was conducted to characterize the current long-term climate in the three studied cities. Moreover, the nonparametric Mann-Kendall test 27,28 was used to detect monotonic trends in the C_ERA5 data to examine whether these data followed a significant (monotonic) trend. The Mann-Kendall test is a nonparametric test that is suitable for all distributions except data subject to serial correlation. Serial correlation denotes the relationship between observations of the same variable with different lag periods. If the serial correlation of a given variable is zero, each observation is independent. Conversely, if the serial correlation approaches one, the observations are serially correlated, and the Mann-Kendall test cannot be applied to detect monotonic trends 29 . Thus, the trend-free prewhitening approach was applied to eliminate serial correlation in the surface air temperature and wind speed time series 30 .
In addition, historical days containing peak events were analysed across all parameters considered, based on C_ERA5 hourly data, to determine the occurrence time of maximum and minimum values.
Moreover, the probability of occurrence of hourly T2m values for each degree Celsius (°C) covering the full T2m range was calculated. Similarly, the probability of occurrence of hourly RH values was also calculated for each 1% increment and each 0.5-m s −1 increment in the hourly wind speed. In addition, the probability of occurrence of hourly SLP values was calculated for each 1-hPa increment.
Finally, principal component analysis (PCA) was applied to the four atmospheric parameters considered (T2m, RH, WS 10 , and SLP) from 1979 to 2020 to produce a linear combination of the original values to initiate ordination of these parameters 31 . PCA is an unsupervised mathematical method to reduce the original variables into smaller-dimension new variables referred to as principal components (PCs) that still contain the original variables. Jolliffe 32 described the PCA method in great detail. Generally, PCA aims to reduce a data matrix with a certain number of columns without much loss of information through (1) calculating the mean values of each column, (2) calculating the anomalies in each column, (3) calculating the covariance matrix of the obtained anomaly matrix, and (4) calculating lists of eigenvalues and eigenvectors of the covariance matrix. The eigenvectors represent the components/directions of the calculated reduced matrix, whereas the eigenvalues represent the magnitudes of the directions. The first PC contains the maximum possible information (data with the most variance), followed by the second PC containing the maximum remaining information, etc.
Statistical downscaling for future projection. The results of the three GFDL realizations were averaged to calculate the GFDL mini-ensemble mean from 2006 to 2100. Next, the cumulative distribution function (CDF) of the C_ERA5 data was matched with the GFDL mini-ensemble mean over the control period (2006-2020) to establish a simple statistical model for bias removal. This statistical model was applied to statistically downscale the studied atmospheric parameters over the long term (2006-2100) under the different future RCP scenarios. The statistically downscaled GFDL mini-ensemble mean simulation results (hereafter, S_D_ GFDL mini-ensemble mean) were used to calculate the future atmospheric uncertainty with a suitable accuracy and validity in Tabuk, Jeddah and Jizan. Future uncertainties in the studied parameters under the four RCP scenarios were calculated based on the 30-year running average.