Nighttime heat waves in the Euro-Mediterranean region: definition, characterisation, and seasonal prediction

The combined effect of above-normal nighttime temperatures with high humidity poses a high risk to human health by impeding the body’s recovery from daytime heat exposure. Seasonal predictions of nighttime heat waves (NHWs) can help to better anticipate these episodes and reduce their social and economic impacts. However, the ability of the seasonal forecast systems to predict NHWs has not been explored yet. This work investigates the potential of four seasonal forecasting systems and a multi-model (MM) ensemble to provide useful information on the frequency and magnitude of the NHWs in the Euro-Mediterranean region during the boreal summer season. The analysis employs a modified version of the heat wave magnitude index (HWMI) to evaluate the NHWs. Our results demonstrate for the first time that this index is an optimal choice for the seasonal prediction analysis as it is invariant to the mean biases and provides an integrated view of the NHWs for the entire season. In addition, the percentage of days in a season with temperatures exceeding the 90th percentile (NDQ90) has been used to assess the NHWs’ seasonal frequency. Different proxies for the assessment of NHWs have been considered: apparent temperature at night (ATn, computed from temperature and humidity at night), mean temperature at night, and daily minimum temperature. All these proxies are valid for the assessment of the NHWs, but ATn is more informative about the stress on human health since it includes the impact of humidity. This work has revealed that state-of-the-art seasonal forecast systems can represent the interannual variability of both HWMI and NDQ90 in Southern Europe, Eastern Europe, and the Middle East, but they show limitations in Northern Europe. The predictive capabilities of the seasonal forecasts in specific regions demonstrate the potential of these predictions for the effective management of the risks associated with summer NHWs.


Introduction
Temperature extremes threaten the global economy (Callahan and Mankin 2022) and their adverse impacts are harmful to human health and ecosystems, agriculture, and the energy sector (Materia et al 2020, Thomas et al 2020, García-Martínez and Bollasina 2021).Extreme heat can have various adverse effects, such as raised mortality rates (Perkins 2015, Lopez-Bueno et al 2021), boosting electricity demand while reducing power production (Zuo et al 2015), intensifying heat stress on vegetation (Stéfanon et al 2012), increasing the risk of wildfires (Parente et al 2018), exacerbating droughts, and reducing crop yields (e.g.Lesk et al 2016).
Several studies have shown that the frequency, intensity, and duration of extreme heat events are expected to increase in the future as a response to anthropogenic global warming (Meehl and Tebaldi 2004, Russo et al 2015, Wang et al 2020).
Furthermore, works are suggesting that heat extremes are fast extending to nighttime (Wang et al 2020).This can pose a significant problem since prolonged hot conditions during the day, coupled with little to no relief from nighttime cooling, can lead to heat-related deaths, particularly among individuals with cardiovascular diseases (Murage et al 2017, Royé et al 2021, He et al 2022, Majeed and Floras 2022).Furthermore, high temperatures combined with high humidity can increase mortality, affect working/exercise capacity, and alter cognitive performance (see Scoccimarro et al 2017, and references therein).Extreme temperatures at night can affect agricultural activities due to the influence of the night temperature in the growth and development of crops such as wheat or rice (e.g.Desai et al 2021, Giménez et al 2021).Beyond that, high temperatures at night can hamper fire control because the absence of night cooling hinders wildfire suppression efforts, which can result in more prolonged and larger fires (Balch et al 2022).The aforementioned impacts demonstrate the importance of anticipating the timing and intensity of extreme temperature events such as daytime heat waves (HWs) or nighttime heat waves (NHWs) on timescales from weeks to seasons (Domeisen et al 2022).
Seasonal climate predictions of extreme heat episodes can be extremely useful for early contingency planning such as volunteer training and the provision of early warnings.These extreme events typically persist for several days; however, providing climate information with day-to-day accuracy is not feasible on seasonal time scales (Vitart et al 2012).Despite this limitation, seasonal predictions have shown the potential to give insight into the average frequency of occurrence or the magnitude of HWs in a season.For example, Batté et al (2018) showed that dynamical seasonal prediction systems (hereafter, SPSs) can predict the interannual variability of spatially averaged Sahelian HWs, and Prodhomme et al (2022) demonstrated that seasonal predictions initialised in early summer can inform on the tendency of a season to be prone to the occurrence of HWs in Europe.In addition, both dynamical (Hamilton et al 2012) and statistical (Collazo et al 2022) prediction systems have shown limited, but still helpful skill in predicting the number of days exceeding the 90th climatological percentile over the extratropical regions.
One of the challenges for the prediction of the NHW is the identification of a suitable index that can be relevant at seasonal timescales.This index should contain information on specific NHWs properties such as duration, frequency, or intensity, as these aspects are crucial to guide specific decision-making processes.To combine information on these properties, several extreme indices have been defined (e.g.Alexander et al 2006, Donat et al 2013).The WMO expert team on climate Change detection and indices has proposed a set of extreme indices (http://etccdi.pacificclimate.org/list_27_indices.shtml)that can be used for specific applications and regions of interest (Perkins 2015).For example, a 'tropical night' is defined as a specific 24 hour period whose daily minimum temperature (hereafter, Tmin) exceeds 20 • C.This definition based on a fixed threshold might not be the most adequate for the seasonal prediction analysis, as these forecasts are affected by systematic errors.Furthermore, the 20 • C absolute value is not the optimal choice for all the regions, and this will hamper the comparison between different locations.A 'warm night' event defined as the day when the daily Tmin exceeds the 90th percentile (see http://etccdi.pacificclimate.org/list_27_indices.shtml) is more adequate in the seasonal prediction context, as it enables spatial comparisons by representing the same part of the temperature probability distribution at each location.The integrated version of this indicator for the full season (i.e. the percentage of days in a season exceeding the 90th percentile) can be a practical frequency indicator and it has been employed in this study.To complement this metric, we made use of the HWMI (Russo et al (2014)).HWMI allows us to describe the magnitude and duration of extreme temperature conditions with only one value per season (Prodhomme et al 2022).
Despite the seasonal predictions of daytime HWs and temperature extremes having been investigated (e.g.Hamilton et al 2012, Batté et al 2018, Prodhomme et al 2022), the ability of dynamical forecast systems to provide useful information on the occurrence of NHWs has not been explored yet.This work aims to fill this knowledge gap by assessing the potential of state-of-the-art SPSs to provide insight into the occurrence of NHWs over the Euro-Mediterranean domain during the summer season.For this analysis, the suitability of different temperature and relative humidity-based indicators for the characterisation of the NHWs have been considered.In section 2 the reanalysis and seasonal forecast datasets used in this work are described.This section also details all the methods employed for the analysis presented here.The main results are described and discussed in section 3, and finally, the summary and conclusions are included in section 4.   , and global coverage.These characteristics made the C3S systems the optimal choice for the seasonal prediction of the NHWs and the future integration of these predictions into operational climate services.The main specifications of the considered SPSs are listed in table 1.

Data and methods
The individual ensemble members from each SPS have been combined by pooling all the ensemble members together to construct a large ensemble with 120 ensemble members.This MM based product can increase the robustness of the analysis as it has been shown that the combination of all the ensemble members from different systems tends to compensate for modelling errors and uncertainties from other sources (DelSole et al 2014).
This work employs retrospective forecasts (i.e.hindcasts) available for the C3S systems in the 1993-2016 period.The predictions initialised on the 1st of May have been used and all the analyses have been performed for the summer season defined from the 15th of May until the 31st August (15MJJA), similarly to Prodhomme et al (2022).The inclusion of the last two weeks of May allows to provide information on NHWs that can be detected in that month, reducing at the same time the effect of the initial conditions on the forecast quality assessment.

NHWs indices: HWMI and NDQ90
This work employs the HWMI defined by Russo et al (2014) and the percentage of days in a season with a certain temperature indicator exceeding the daily 90th climatological percentile (NDQ90).The HWMI is a dimensionless index that provides the maximum magnitude of the HW in a season, and in the present work, the index will be adapted to quantify the maximum magnitude of the NHWs.The NDQ90 is an indicator of the expected frequency of extreme heat conditions in a particular season and it has been used to assess the seasonal mean frequency of the NHWs.For the sake of brevity, a detailed description of how these indices have been computed has been included in the supplementary material (section S1).
The HWMI and the NDQ90 indices used in previous works (Russo et al 2014, Prodhomme et al 2022) for the HWs' analysis are usually derived from the daily maximum temperature (hereafter, Tmax).However, the main goal of this work is to investigate the NHWs, therefore the proxy variables considered for this analysis are slightly different.The daily Tmax has been replaced by the daily Tmin, the average temperature at night (between 23-06 local time, hereafter, Tn), and the apparent temperature at night (ATn).The ATn computation is based on Russo et al (2017) and it employs the relative humidity and temperature.These three proxy variables have been selected for the NHWs' analysis because the daily Tmin has been already employed for the characterization of NHWs (Yeh et al 2021), the Tn allows the assessment of long-lasting extreme temperature at night which can have devastating impacts, and the ATn has been considered because it combines both temperature and humidity.The role of humidity in extreme heat conditions can have a striking impact on human health for high levels of ambient humidity or, conversely, its effect can be ameliorated by a dry ambient (Batté et al 2018).

Forecast quality metrics
To explore the ability of the C3S SPSs to effectively predict the NHWs indices a deterministic and a probabilistic skill score have been used: Pearson's correlation coefficient and Brier probability skill score (BSS).These metrics have been computed for the seasonal values in the 1993-2016 period (i.e. the sample size is 24 years).
The Pearson correlation coefficient (Wilks 2011) has been used to measure the linear association between the ensemble mean of the seasonal forecasts and the reference data set (Jolliffe and Stephenson 2011).A t-test has been applied to evaluate if the correlation values are significant at the 95% confidence level.
The BSS is the skill score based on the Brier score which is the sum of the square differences of the probabilities of forecast and observations for binary events (Jolliffe andStephenson 2011, Wilks 2011).In this case, the binary event has been defined as the forecast probability of one specific variable in the upper quartile (i.e.exceeding the 75th percentile) of the climate distribution, as the BSS has been used to assess if the distribution of the ensemble members is representative of the likelihood of occurrence of an extreme NHW.The 75th percentile threshold has been selected, as this will allow us to assess how good are the SPSs at representing the six seasons (out of the 24 summers considered in our analysis) with the highest NHWs conditions.The sampling uncertainty of the BSS values has been assessed according to the Diebold-Mariano test (Diebold and Mariano 1995), which allows us to identify if the differences between the seasonal forecasts and a climatological forecast are statistically significant at the confidence level of 95%.

Sensitivity of the NHWs' indices to different proxy variables
To illustrate how the choice of one specific temperature indicator affects the characterisation of NHWs, three summers with extreme temperature conditions have been selected: 2003, 2010, and 2015 (figure 1).The daytime heat conditions for these three summers have been widely investigated (see Russo et al 2015, and references therein) due to their numerous socio-economic consequences over the Euro-Mediterranean domain.Nevertheless, the specific nighttime conditions have not been explored so far.For the assessment of the NHWs through HWMI, the ATn has been used (HWMI_ATn) as this variable allows the investigation of the influence of the humidity on the NHWs.However, HWMI has also been calculated using Tmin and Tn, as these variables have already been used in the literature to assess nighttime heat conditions.To explore the differences between the HWs and NHWs the daily Tmax has been considered.This comparison is interesting, as most of the works available in the literature are based on the evaluation of the HWs by using Tmax.This sensitivity analysis has not only been performed for the HWMI (figure 1) but the corresponding results for the NDQ90 index have been included in the supplementary material (figure S2).
The 2003 heat wave caused several thousands of deaths in Europe, and the physical drivers, impacts, and relation with global warming have been widely discussed in the literature (e.g.Thommen 2005, Fouillet et al 2006).To explore the summer of 2003 from the nighttime perspective, the HWMI_ATn is shown (figure 1(a)).This extreme episode was particularly intense in Portugal, Northern Spain, France, Switzerland, and southwestern Germany, where the HWMI_ATn values are higher than 3 for the summer season.However, if other temperature-based indicators are considered, such as the daily Tmin (figure 1(d)) or the mean temperature at night (figure 1(g)), these regions show less intense NHWs, as suggested by the positive values of the differences.Differences with HWMI_ATn are related to the role of humidity, which might amplify the effect of the high temperatures at night in specific regions such as Northern Africa, and specific locations over the Iberian Peninsula and France.However, this impact was limited, as the HWMI_ATn values only reached values as large as 1.5 in Northern Africa, while in Western Europe (where the heat conditions were particularly harmful in the summer of 2003) they exceeded 3. The influence of humidity on the 2003 NHW is particularly evident in southern Portugal and Sardinia, where the HWMI_ATn is stronger than the HWMI_Tmin and HWMI_Tn.The comparison of the NHW conditions in 2003 with the daytime counterpart (figure 1(j)) shows more intense heat conditions at night than during the day in Portugal and Northern Spain.The opposite effect is shown in the Pyrenees, parts of Switzerland, and Northern Italy.These differences may be explained by the persistent omega blocking pattern (Liu et al 2020), responsible for a mean southwesterly flow over Western Europe.Therefore, the advection of humid oceanic air inhibited daytime extreme peaks in West Iberia but limited the heat dispersion at night through the increased humidity.Contrarily, scorching dry air from inland Africa was advected to the Pyrenees, western Alps, and Northern Italy, causing the strongest Tmax anomalies than for the ATn.
In the summer of 2010, record-breaking temperatures were registered in a vast domain over Russia causing several socio-economic impacts such as several thousands of deaths (Barriopedro et al 2011).The drivers of this extreme event were a persistent anticyclonic pattern combined with soil moisturetemperature feedback mechanisms (Barriopedro et al 2011, Hauser et al 2016).In this case, HWMI_ATn and HWMI_Tn show little differences (figure 1(h)) in a few grid points, indicating that the role of humidity was low for this specific summer.This is in agreement with results obtained for the analysis of the humidity influence in the daytime HWs Russo et al (2017), and confirmed by HWM_Tmax, that are much stronger than NHWs calculated with HWMI_ATn (figure 1(k)).Nevertheless, the HHWMI_ATn (figure 1(b)) shows values above 5 over a large domain, which suggests the extreme heat conditions during the day persist at night.Miralles et al (2014) attributed this to the persistence of the heat generated during the day which was preserved in the atmospheric layer located several hundred meters above the surface, and which is available to re-enter the atmospheric boundary layer during the next diurnal cycle.Another possible explanation is the concurrence of enormous wildfires, whose radiative effect directly impacted temperatures over Russia both during day and nighttime (Péré et al 2014).However, the intensity of the NHWs is lower in Russia if the HWMI_Tmin instead of the ATn (figure 1(e)).Possibly, in a high-latitude continental climate, the temperature during the very few dark hours was allowed to drop enough to mark the difference with the average nighttime temperature.
The HWMI_ATn in 2015 (figure 1(c)) shows lower values than the corresponding index for 2003 and 2010, but still, HWMI_ATn values around 3 can be found over eastern Spain, Italy, the Czech Republic, and also at the border between Libya and Egypt.The extreme temperature conditions in the summer of 2015 affected Western Europe (Ardilouze et al 2017), and this extreme episode was related to the specific atmospheric circulation and the forcing of the greenhouse gases and to a smaller effect of soil moisture (Wehrli et al 2019).In some specific locations over eastern Spain, and central Italy, the differences between the HWMI_ATn and for both HWMI_Tmin and HWMI_Tn are evident (figures 1(f) and (i)), suggesting a heat amplification effect due to the humidity at night.This effect is also noticeable in Northern Africa (particularly in a large portion of North Africa), where positive differences can be identified in several locations.By contrast, over Northern Scandinavia and some regions of Eastern Europe, the magnitude of the daytime HWs exceeded the corresponding index for the NHWs (figure 1(l)).This shows the importance of highlighting the extreme heat conditions from both the daytime and nighttime perspectives and the need for different indicators that can be used to explore the impacts of the NHWs in specific regions.

Seasonal prediction of temperature-based variables
The current ability of the C3S SPSs to simulate the interannual variability of the ATn is high in most of the European regions (figure 2).The summer skill in Southern Europe has been attributed to the external forcing associated with the global warming trends (Patterson et al 2022).By contrast, in the British Isles, southern Scandinavia, and the Atlantic coasts the correlation values are low and often not significant.This is related to the limitations of the SPSs in reproducing some teleconnections and the low potential for predictability in the summer season for those regions (Patterson et al 2022).
Most SPSs show a similar regional distribution of high and low correlations, however, some differences can be identified.Positive and significant values over the UK are only shown by MF-7, while mostly no skill is shown for the other individual SPSs.The CMCC-35 prediction system shows the highest correlation values over Spain, France, and Italy.ECMWF-5 presents the maximum correlation values over Eastern Europe and eastern Mediterranean regions.The MM does not provide the highest correlation values than the best individual seasonal forecast system in every single grid point, however, it offers a coherent distribution of the higher correlation values over a very large domain.This illustrates the MM potential for the generation of actionable seasonal forecasts of the ATn that can be used for decision-making in vulnerable sectors.
To assess how the seasonal predictions reproduce the interannual variability of the different proxies (i.e.ATn, Tmin, and Tn), the differences in correlation have been computed.The results for each SPS are included in the supplementary material (figure S3).To simplify the discussion, figure 3 shows the correlation differences between the MM-based seasonal predictions of the different temperature indicators.Figure 3 displays the coherence in the sign of the correlation differences with the individual SPSs.The correlation differences are generally low but they are systematically obtained in specific regions for all the SPSs considered.Positive correlation differences between ATn and Tmin are observed over the Iberian Peninsula for all systems (figure 3(a)).By contrast, negative differences are identified in several Northern African regions.The correlation differences between ATn and Tn are generally low, except in Northern Africa, but consistent positive differences can be identified in the British Isles and Central Europe, which suggests slightly higher seasonal predictive capacity for the ATn than for the Tn in those specific regions.Most of these differences are not statistically significant, which suggests similar levels of seasonal forecast quality for the three temperature proxies considered for the characterisation of the NWHs.Hence, ATn can be a good proxy for the seasonal assessment of the NHWs as it includes information on the humidity conditions, and the SPSs can reproduce its interannual variability.
Figure 3(c) shows the correlation differences between ATn and daily Tmax.SPSs generally provide higher correlations in Northern Europe (i.e.Scandinavia and the UK) and some Mediterranean countries (e.g.Italy, Bosnia, Serbia, Greece, etc) for the ATn than for the Tmax, which suggests higher potential skill in the NHWs than for the daily HWs in those regions.Contrarily, in the Iberian Peninsula, Northern Africa, and Eastern Europe, Tmax shows a higher correlation than for the ATn.Despite the high and systematic correlation differences, these results should be interpreted with caution as only a few grid points show statistically significant differences (figure S3) which might be related to the small sample available for this analysis (24 years).The limited sample size and particularly short hindcast lengths can affect seasonal forecast quality evaluations (Manzanas et al 2022).

Seasonal prediction of NHWs indicators: HWMI and NDQ90
To investigate the ability of the SPSs to provide useful predictions of the HWMI and NDQ90 indices, our domain has been divided into seven different regions (figure 4(a)) plus the entire Euro-Mediterranean domain (i.e.averaging all the grid points in the region, figure 4(a) black box).These regions are similar to those used by Prodhomme et al (2022) to analyse the performance of the seasonal predictions of the daytime heatwaves.The regional analysis allows us to provide more insight into the occurrence of temperature extreme events like NHWs, which have an impact on multiple nearby grid points simultaneously.However, the results obtained at the grid point level for the MM predictions have been shown in figures 4(d) and (e).Figure 4  This indicates the average frequency of the NHWs over the season is more predictable than other NWHs' properties such as the duration or the intensity.This might be related to the persistence of the extreme temperature anomalies which is difficult to capture for some of the SPSs.However, SPSs can still reproduce the frequency of extreme NHWs.
In general, the correlations are higher for the full domain than for the single regions, except for the Middle East.This increased skill is expected, and it is associated with the aggregation of many grid points (Gong et al 2003, Diro et al 2012).The values.As discussed in the previous section, limited skill is found in the seasonal temperatures in Northern and Western Europe for both NHWs indices.Overall, the MM provides predictions with positive and significant correlations for both HWMI and NDQ90 in several regions (figures 4(d) and (e)).This evidences the potential of the seasonal predictions to provide useful information on the NHWs that can be used for the decision-making processes in different socio-economic applications.
The seasonal prediction skill for the two NHWs' indicators considered has been also assessed from the probabilistic point of view by using the Brier skill score (BSS, figure S6).The probabilistic assessment provided similar results to those obtained for the correlations, with higher BSS values for the NDQ90 than for the HWMI and a similar spatial distribution of the skill with the highest BSS values for specific Mediterranean regions and in the Middle East.

Conclusions
This work investigates the ability of the state-of-theart SPSs to provide useful information on the NHWs over the Euro-Mediterranean region at seasonal time scales.This information is not currently available in the literature, but it can be crucial to ameliorate the impacts of the NHWs on human health, agriculture, energy, or water management activities on time.
This study is the first to illustrate how specific proxy variables and indices can be used to describe the NHWs at seasonal time scales.In addition, the seasonal prediction skill of four of the C3S European systems and their MM combination has been assessed.The apparent nighttime temperature (ATn) was used as a proxy variable for the assessment of the NHWs, and it has been compared with other possible indicators such as the daily minimum temperature and the mean temperature at night.This work reveals ATn is a very informative indicator because it includes humidity and therefore the concept of heat stress during the night, which impacts human health and agricultural activities.In addition, the different SPSs can reproduce the interannual variability of the ATn in most of the Euro-Mediterranean regions considered, except in Northern Europe.Therefore, the use of ATn can be more valuable for decision-making than other proxies that do not include information on humidity.
The evaluation of the seasonal predictive capacity for the NHWs has been performed by using the HWMI and the NDQ90.This work has shown for the first time the suitability of these indices for the seasonal analysis of NHWs as they provide one value per season and they are based on relative thresholds which reduce the impact of mean biases.Based on these two specific indices, it has been shown that seasonal predictions can provide useful deterministic and probabilistic information on the NHWs during the summer season in some Euro-Mediterranean regions, particularly over the Mediterranean and the Middle East.However, these SPSs provide better results for the NDQ90 than for the HWMI, which indicates these predictions are better at capturing the frequency of the NHWs than the magnitude of these episodes.
The seasonal prediction skill of HWMI and NDQ90 in specific regions such as Southern and Eastern Europe, or the Middle East shows that these indices can be integrated into specific climate services intended to reduce the impact of the NHWs in vulnerable sectors such as public health or agriculture.However, the ability to predict temperature extremes in Northern Europe is still limited and it also affects the quality of the indices employed for the assessment of the NHWs.The improvement of the seasonal predictions in these regions requires a better understanding of the physical mechanisms involved in the occurrence of the NHWs.For example, the influence of nighttime land-surface processes on the frequency and magnitude of extreme temperatures should be further investigated.These mechanisms are different from those affecting daytime, and likely involve other players such as boundary layer stability and long-wave radiation (Materia et al 2022).Artificial intelligence and machine learning algorithms can be exploited for the detection of new predictability sources, processbased understanding (Barriopedro et al 2023), and for the combination of statistical and dynamic models to enhance the current seasonal forecast quality of extreme temperature events.

Figure 1 .
Figure 1.ERA5 Heat wave magnitude index (HWMI) in the boreal summer (15MJJA) of 2003, 2010, and 2015.The HWMI has been computed with the apparent temperature at night (ATn, first row) and the differences with the corresponding index based on daily minimum temperature (Tmin, second row), mean temperature at night (Tn, third row), and daily maximum temperature (Tmax, fourth row).

Figure 2 .
Figure 2. Correlation between the ensemble mean of the C3S seasonal of the apparent temperature at night (ATn) and the corresponding variable from ERA5 in the 15MJJA season for the 1993-2016 period.The seasonal forecasts were issued on the 1st of May.White colour has been used to mask regions with non-statistically significant correlations at the 95% confidence level.

Figure 3 .
Figure 3. Correlation differences between the correlation maps of the multi-model seasonal predictions for the ATn and the corresponding correlation maps but for (a) Tmin, (b) Tn, and (c) Tmax for the 1993-2016 period in the 15MJJA season.These correlation maps were computed with ERA5 as an observational reference.Hatched areas indicate where the four individual prediction systems agree in the positive (green lines) or negative (purple lines) correlation differences.The seasonal forecasts were issued on the 1st of May.
illustrates the results for ATn-based indices, but the corresponding results for the other proxy variables have been included in the supplementary material (figures S3-S5).The individual SPSs and the multimodel can simulate the interannual variability of both HWMI_ATn (figure 4(b)) and NDQ90_ATn (figure 4(c)), as shown by the positive and significant correlations over the Euro-Mediterranean domain.All SPSs show similar correlations for both indices, with NDQ90_ATn (figures 4(c) and (e)) systematically overperforming HWMI_ATn (figures 4(b) and (d)).

Figure 4 .
Figure 4. (a) Domains of the regions considered in the analysis: Northern Europe (NE), Western Europe (WE), Mediterranean (MED), Northern Africa (NAF), Central Europe (CE), Eastern Europe (EE), Middle East (ME), full European domain (ALL).Ensemble mean correlation values for the seasonal predictions of the (b) HWMI and (c) NDQ90 based on the ATn for the individual C3S seasonal prediction systems and their multimodel combination (rows) for each region (columns).Significant correlations at the 95% confidence level are marked with an asterisk ( * ).Ensemble mean correlation values for the seasonal predictions of the (d) HWMI and (e) NDQ90 based on the ATn for the multimodel combination at the grid point level.The seasonal forecasts were issued on the 1st of May and the observational reference is ERA5.The analysis corresponds to the 15MJJA season in the 1993-2016 period.White colour has been used to mask regions with non-statistically significant correlations at the 95% confidence level.

Table 1 .
C3S seasonal prediction systems employed in this work.