The extreme heat wave of late July/early August 2021 in Greece under the context of the direct effect of anthropogenic greenhouse gases

Greece is characterized by a significant warming trend in recent decades, accompanied by increasing frequency, intensity, and duration of heat waves (HWs). A particularly devastating HW that affected the country was the late July/early August 2021 event (JA2021HW), which lasted for 9 days (July 28–August 5). Focusing on the hottest day of the event (August 3), the main characteristics of JA2021HW are presented in the current study, using model reanalysis data and up to 11‐year observations derived from the dense network of ground‐based weather stations operated by the Meteo Unit at the National Observatory of Athens (NOA). This analysis highlights the severity of JA2021HW, especially in the central and southernmost regions of Greece. Most importantly, the impact of the direct effect of anthropogenic greenhouse gases to the examined extreme event, in terms of intensity and probability of occurrence, is examined by employing a regional 31‐member ensemble (ENS) modeling approach based on Weather Research and Forecasting (WRF) model, which is operationally used by NOA/Meteo. Firstly, WRF is validated under 7‐day lead‐time ENS simulations with current‐state greenhouse gas (GHG) concentrations (GHG_2021 ENS), showing a robust model performance in replicating the JA2021HW's magnitude on August 3rd. Then, 7‐day lead‐time ENS simulations with the GHG concentrations reduced to the pre‐industrial (1854) levels (GHG_1854 ENS) are performed and compared to the GHG_2021 ENS experiment. The results reveal a contribution of the immediate anthropogenic warming due to the increased GHG concentrations to the JA2021HW intensity in West and South continental and insular Greece, which can be important in the framework of the human health impacts of extreme temperatures. For the event's occurrence probability, no robust evidence of any change could be derived. These statements are partially constrained by the fact that only the direct GHG effect on the timescale of a few days was examined.

climate change, extreme event attribution, Greece, heat waves, regional modeling, WRF

| INTRODUCTION
Situated in the eastern Mediterranean climate change hot spot (Zittis et al., 2022), Greece is frequently affected by heat waves (HWs), especially in July and August (Galanaki et al., 2023).Thus, the country is prone to the severe impact induced by such extreme weather events.However, prolonged HWs associated with exceptionally high temperatures set the stage for disastrous impacts.Unfortunately, this was the case during the late July/early August 2021 HW (JA2021HW) that affected Greece.The episode, defined based on a percentile-based approach (Giannaros et al., 2023), lasted for 9 days (July 28-August 5, 2021) and was characterized as extreme in terms of intensity and duration (Founda et al., 2022;Giannaros et al., 2023).Up to 11-year records of daily maximum temperatures (Tmax) were exceeded in more than 10 stations of the nationwide observational network operated by the Meteo Unit at the National Observatory of Athens (NOA; Lagouvardos et al., 2017) and 265 of them recorded temperatures higher than 40 C on August 3rd, 2021.The second-highest Tmax since 1900 was also recorded in the historical climatic station of NOA in Athens on the same day (Giannaros et al., 2023).In terms of heat-related parameters that affect the human thermal environment, JA2021HW ranks first among episodes identified since 1900, with respect to the highest nighttime temperature and greater accumulated heat and thermal discomfort (Founda et al., 2022).As a result, over 2300 excess deaths-excluding those related to the COVID-19 pandemic-compared to the 2016-2020 average were recorded during the episode, based on Eurostat data (Politico, 2021).Furthermore, the extreme heat contributed to the outbreak of five major wildfires in the south and central Greece, which collectively burnt $94,000 ha (Giannaros et al., 2022) and resulted in significant deterioration of air quality due to smoke (Masoom et al., 2023).
The formation and evolution of the above HW episode constitute a textbook example of synoptic atmospheric conditions that favor summer heat extremes in the eastern Mediterranean: Stagnant ridges of high pressure associated with atmospheric blocking over the greater area of Greece resulting in (i) persistent subsidence of warm air, (ii) extensive advection of hot air masses originating from North Africa, through the northward displacement of the sub-tropical jet stream, and (iii) enhanced adiabatic heating due to the increased clear-sky solar radiation (Giannaros et al., 2019;Lhotka & Kyselý, 2022).This large-scale weather pattern is similar to those that led to the deadly HWs of 1987 (Prezerakos, 1989) and 2007 (Demirtaş, 2018;Kotroni et al., 2011;Retalis et al., 2010).Examining the role of the immediate anthropogenic warming on such extremes allows for a better understanding of the direct climate change effects and raising awareness about them.Numerical weather prediction (NWP) models are valuable tools in this direction, providing robustness in terms of reliability, computational demands, and operational timeliness (Hope et al., 2015;Leach et al., 2021;Meredith et al., 2015;Pall et al., 2017;Wang et al., 2018).For instance, Leach et al. (2021) showed that the European winter heat wave of February 2019 was intensified over the British Isles by 0.31 [0.24, 0.37] K due to the direct effect of CO2, using 9-day lead ensemble simulations of the operational forecasting model of the European Centre for Medium-Range Weather Forecast.
Here, we employ the mesoscale Weather Research and Forecasting (WRF) model within the operational weather forecasting framework applied by the Meteo Unit at NOA, in order to investigate how the increased levels of greenhouse gases due to the human activities impacted the extreme late July/early August 2021 HW in Greece.First, a description of the event is provided using observational and numerical model reanalysis data.Then, the predictability of the episode is examined using the NOA/Meteo operational WRF system.Finally, an ensemble analog of the examined event is defined and simulated under a hypothetical preindustrial climate setting in terms of greenhouse gas (GHG) concentrations, and it is compared to JA2021HW concerning the heat wave probability of occurrence and amplitude in terms of absolute maximum temperature on the hottest day of the event (August 3, 2021).Note that we do not examine the total long-term climate change effects, including the sea surface temperature (SST) warming, but rather the immediate impact of anthropogenic greenhouse gases, as expressed by the additional heat trapped in the atmosphere in the course of JA2021HW.

| DATA AND METHODS
Near-surface air temperature observations, derived from the NOA/Meteo network of ground-based weather stations (Kotroni et al., 2021;Lagouvardos et al., 2017), and the ERA5 model reanalysis data (Hersbach et al., 2020) are used to (i) climatologically characterize the JA2021HW and (ii) validate the NOA/Meteo operational WRF model (version 4.4;Skamarock et al., 2019) in terms of reproducing the August 3, 2021, Tmax in the study area (Greece; Figure 1a).The configuration of the WRF model includes a high-resolution simulation domain (2 Â 2 km) over Greece and neighboring countries (Giannaros et al., 2023).
For the ensemble analog of the examined episode, two series of 31-member ensemble (ENS) simulations with the WRF model are performed using the same GEFS (Global Ensemble Forecasting System) surface and upper-air analysis and forecast data (Zhou et al., 2022) for atmospheric initial and boundary conditions (ICBC).The same data are also used in both experiments for the SST initialization at the lower model boundary, which are based on highresolution multi-satellite observations provided by the Copernicus Climate Change Service (C3S, 2019).For the first simulation series (GHG_2021) current-state greenhouse gas (GHG) concentrations were set for the WRF simulations, while the GHG atmospheric concentrations were reduced to pre-industrial (1854) levels for the second ensemble series (GHG_1854; Table 1).
All simulations are performed using a 12-h spin-up period and a 7-day lead time (i.e., initialized on 27 July 2021, 1200 UTC), as the impact of GHG forcing on the simulated temperatures is greater compared to smaller lead times, according to sensitivity tests and previous studies (Leach et al., 2021).

| EVENT DESCRIPTION AND MODEL VALIDATION
Figure 1a shows the locations of the 45 NOA/Meteo surface weather stations, which have operated since 2010 Greece with borders identification of six major regions (yellow: North Greece; blue: North Aegean; black: Central Greece; white: West Greece; red: South Greece; orange: South Aegean).Black dots indicate the locations of the 45 used surface weather stations, while the red square denotes the location of the Makrakomi observational site.(b) ERA5 850 hPa temperature (filled contours, at 1 C intervals) and 500 hPa geopotential height (black contours, at 25 gpdm) anomalies on 3 August, 2021, with respect to the reference period 1991-2020.The blue rectangle encloses the area shown in Figure 1a.and were used in the current analysis.It also denotes the location of the Makrakomi observational site, where the highest Tmax (46.3 C) was recorded on August 3rd, 2021.Overall, the nine highest temperatures recorded that day were greater than 45 C (Founda et al., 2022), highlighting the severity of the JA2021HW.The extreme character of the episode is also indicated by the significant anomalies in the upper-level synoptic conditions based on ERA5 model reanalysis data and using the 1991-2020 period as a reference (Figure 1b).The 850 hPa temperature together with the 500 hPa geopotential height anomalies are shown to illustrate the prevailing synoptic weather pattern during the JA2021HW.A deep trough in western Mediterranean on August 3rd, 2021, resulted in a strong south-westerly flow of warm and dry air over eastern Mediterranean.The 850 hPa temperature anomalies, in particular, exceeded 8 C over Greece.Also, the high anomalies of 500 hPa geopotential heights over Greece indicate adiabatic warming (compression) in subsiding air (Figure 1b).These anomalies correspond to 4-5 and 2 σ (standard deviations) for 850 hPa temperature and 500 hPa geopotential height, respectively.
Figure 2  August 3rd, 2021 exceeded notably all the respective Tmax observations in June-July-August (JJA) from 2010 to 2020 over Central and South continental Greece (Figure 2c,e), as well as over the South Aegean (Figure 2f).The North and West Greece (Figure 2a,d) and the North Aegean (Figure 2b) also experienced increased regionally averaged Tmax on the examined JA2021HW day compared to the whole observational Tmax time series.Figure 2 also presents the WRF-simulated Tmax on August 3rd, 2021 under the GHG_2021 ENS experiment.The model was capable of replicating the high regionally averaged Tmax observations on the examined heat wave day (Figure 2).The differences between the model ensemble and observed mean in particular, were lower than 2 C over all six regions, and close to 0 C, especially for North Greece, North Aegean, and West Greece (Figure 2a,b,d).Moreover, the WRF modeled synoptic circulation is in close agreement with the ERA5 reanalysis synoptic setting on August 3rd, 2021, showing stationary high-pressure ridges toward the eastern Mediterranean and Balkan Peninsula at 500 hPa and strong adiabatic and advective warming in the underlying atmospheric levels (850 hPa; Figure 3a,b).It is important to note that the model dynamics were not affected by the applied changes in the GHG concentrations, as indicatively illustrated in Figure 3b,c.For all ensemble members in particular, the modeled circulation when the GHG concentrations were set equal to those corresponding to the F I G U R E 3 Synoptic conditions (500 hPa geopotential height in gpdm and 850 hPa temperature in C) on August 3rd, 2021, at 1200 UTC: (a) ERA5 reanalysis data; (b,c) 10 km WRF-simulated data for the GHG_2021 and GHG_1854 ENS experiment, respectively, based on the 17th member of the ensemble, which is the best performing member in terms of predicting the nearest to the ERA5 synoptic pattern based on qualitative evaluation.Gray shade areas in (b,c) correspond to mountain regions, where the computation of the upper-level air temperature is not possible, as the 850 hPa isobaric surface is below the WRF model ground level.
pre-industrial era (GHG_1854 ENS) shows a pattern correlation of greater than 0.99 with the modeled circulation with the current-state greenhouse gases (GHG_2021 ENS).Thus, the differences between the two experiments that are examined in the next section can be attributed to the direct effect of GHGs (i.e., diabatic heating of the surface and adjacent air due to the increased downward long-wave radiation).denoting an overall warming over Greece.This is associated with the additional heat trapped in the atmosphere in the course of the heat wave due to the direct effect of anthropogenic greenhouse gases, as illustrated in Figure 4b.The daily mean surface clear-sky long-wave radiation downwards on August 3rd, 2021, specifically is greater by up to 3.1 W/m 2 under the ensemble with enhanced levels of GHGs in the atmosphere (Figure 4b).This extra amount of thermal radiation is impacted by indirect cloud effects, which may increase or decrease the greenhouse radiative forcing under clear-sky conditions.However, this impact is relatively small compared to the direct contribution of GHGs in diabatic heating (not shown).

| IMPACT OF GHG FORCING ON JA2021HW'S AMPLITUDE
The positive differences in Tmax on August 3rd, 2021, are stronger in the South continental and insular Greece, and in the western parts of the country (> +0.2 C).In these areas, the spatial coverage of statistically significant positive differences at the 90% confidence interval (CIbased on non-parametric Wilcoxon signed-rank test; Wilks, 1995) is also greater (Figure 4a).The highest regionally averaged increase is found particularly in West Greece, being equal to 0.10 C (green lines in Figure 4c) and ranging from 0.02 to 0.13 C at the [0.05, 0.95] % CIs (dark blue areas in Figure 4c), which were computed using a 100,000-member bootstrap.The overall increase in the JA2021HW's amplitude due to the immediate impact of the anthropogenic greenhouse gases is also evident in the South continental (0.08 [0.03, 0.11] C) and insular (0.07 [0.03, 0.09] C) Greece, as well as in the rest of the examined regions.However, a greater uncertainty characterizes this outcome in the latter regions, as the lower CIs are below zero, especially in North continental (0.03 [À0.03, 0.07] C) and insular (0.05 [À0.05, 0.07] C) Greece (Figure 4c).
The ensemble spread also indicates negative changes in all regions (light blue boxes and whiskers in Figure 4c).This is related to the model's internal variability, which can be high at the medium-range forecast lead time applied (7 days).A greater ensemble size in numerical experiments conducted could help reduce this variability.However, it should be noted that even with the current ensemble size (31 members), the negative differences in the ensemble median on August 3, 2021, Tmax are mostly statistically insignificant.Thus, confidence may be placed in the results showing that the increased GHG concentrations in the second set of ensemble simulations are associated with warming, as expected.Concerning the probabilistic impact assessment, a risk-based approach (Leach et al., 2021;Stott et al., 2016) was applied (not shown), but there was no robust evidence of any change in the event's occurrence probability.

| DISCUSSION AND CONCLUDING REMARKS
Focusing on the hottest day of the JA2021HW, the analysis of the characteristics of the episode in the present study revealed the exceptionally hot conditions that prevailed at the surface (highest Tmax equal to 46.3 C in Makrakomi) and upper-troposphere (4-5 σ anomalies in 850 hPa temperature over Greece).Examining the regionally averaged observed Tmax on August 3rd, 2021, under a recent-past context (2010-2020) highlighted further the extreme magnitude and spatial extent of the event, with the central and southernmost areas of Greece being affected the most.These findings are in line with the outcomes of Lhotka and Kyselý (2022) and complement the study of Founda et al. (2022), which showed that JA2021HW ranks first in terms of duration, highest nighttime temperatures and accumulated heat when compared to heat waves identified since 1900.
The contribution of the direct effect of anthropogenic greenhouse gases to the above exceptional characteristics of JA2021HW was investigated in the current study using a regional NWP modeling methodology (Meredith et al., 2015;Pall et al., 2017;Wang et al., 2018) and following the approach of Leach et al. (2021) that allows for both analyzing the absolute impact (storyline approach; Trenberth et al., 2015) and probabilistically assessing the impact (risk-based approach; Stott et al., 2016).Given the same atmospheric circulation, two high-resolution (2 km) ensemble experiments were performed using the WRF model in particular, in order to examine how the higher GHG concentrations due to human activities altered the magnitude and probability of occurrence of the event, as expressed by the maximum temperature on the hottest day (August 3, 2021), within the 7-day forecast period.Αn overall increase of the JA2021HW intensity was indicated by the positive ensemble median differences on August 3, 2021, Tmax that are primarily evident between the conducted experiments.The results are more robust in South continental and insular Greece and in West Greece, based on the statistical significance of the computed differences, reaching up to 0.10 [0.02, 0.13] C in the latter region.These regionally averaged increases in the heat wave's amplitude are comparable with those provided by Leach et al. (2021) and they are associated with the direct effect of GHGs, as indicated by the increased amounts of surface clear-sky long-wave radiation downwards on August 3rd, 2021, under the simulations with the elevated current-state GHG levels.Less confidence characterizes the temperature increases in Central and North Greece and in North Aegean.In these areas decreases in intensity were indicated, but they were mainly statistically insignificant and they are associated with the model's internal variability.
Concerning the likelihood of the event, no reliable conclusions were possible to be derived based on the probabilistic assessment results.
Overall, the above outcomes may indicate a small impact of the immediate anthropogenic warming induced by the elevated GHG concentrations to the late July/early August 2021 HW in Greece.However, mortality risk ratios due to excess heat rise sharply, especially for vulnerable people, when daily maximum temperatures exceed 40 C (Díaz et al., 2002;McMichael et al., 2006;Tsoutsoubi et al., 2021).For instance, Díaz et al. (2002) showed that an increase of 1 C above 41 C is related to an increase of 51% in baseline mortality for elderly (>65 years).Thus, under extreme heat conditions, as in the studied heat wave, even a decimal increase in Tmax may cost hundreds or even thousands of lives.
It is important to note that the presented outcomes are limited by the fact that only the immediate direct GHG effect on the timescale of a few days was considered in the analysis.This analysis could be enriched in the future by examining more thoroughly the mechanisms and feedback associated with the GHG-induced adiabatic heating (Stuecker et al., 2018).Moreover, future efforts on characterizing the anthropogenic contributions to JA2021HW could focus on removing the long-term trends associated with the human-induced climate change from the forcing ICBC (Pall et al., 2017;Wang et al., 2018) and on including anthropogenic climate changes related to SST (Meredith et al., 2015) and atmospheric circulation (Faranda et al., 2022).The SST warming, in particular, reflects the indirect effect of increased GHG levels in the long term and it has a greater impact on air temperatures (Hope et al., 2015).Thus, its consideration allows for capturing a greater and more significant portion of the total anthropogenic climate change.This is important in the framework of the associated impacts on human health mentioned above.Under the same context, a future study could also focus on the effects of anthropogenic greenhouse gases on the night-time temperatures, which are known to have a more damaging impact on the human body (Giannaros et al., 2023).
highlights further the abnormal and widespread nature of the examined HW.The analysis is performed based on ground-based observations in six distinct regions in Greece, as denoted by different color borders in Figure 1a.The regionally averaged observed Tmax on F I G U R E 2 Regionally averaged Tmax over (a) North Greece, (b) North Aegean, (c) Central Greece, (d) West Greece, (e) South Greece, and (f) South Aegean.Black lines with dots correspond to the 2010-2020 JJA ground-based observations.Red solid lines and blue dotted lines correspond to the ground-based observations and modeled data (GHG_2021 ENS experiment), respectively, on August 3rd, 2021.T A B L E 1 Summary of the conducted 31-member ensemble experiments with respect to the GHG concentrations.by volume.b Parts per billion by volume.

Figure
Figure4ashows the geographical pattern of the ensemble median differences in Tmax on August 3, 2021, between the conducted experiments with varying GHG forcing in the study area.Positive differences are primarily evident,