Northern Hemisphere atmospheric pattern enhancing Eastern Mediterranean Transient-type events during the past 1000 years

High resolution climate model simulations for the last millennium were used to elucidate the main winter Northern Hemisphere atmospheric pattern during enhanced Eastern Mediterranean Transient (EMT-type) events, a situation in which an additional overturning cell is detected in the Mediterranean at the Aegean Sea. The differential upward heat flux between the Aegean Basin and the Gulf of Lion was taken as a proxy of EMT-type events and correlated with winter mean geopotential height at 500 mb in the Northern Hemisphere (20oN 90oN and 100oW 80oE). Correlations revealed a pattern similar to the Eastern Atlantic / Western Russian (EA/WR) mode as the main driver of EMT-type events, with the past 1000 yr of EA/WRlike mode simulations being enhanced during insolation minima. Our model results are consistent with alkenone Sea Surface Temperature (SST) reconstructions that documented an increase in the west-east basin gradients during EMT-type events.

Mediterranean region entirely, and implements a spatial resolution of 45 km. As with ECHO-G, this model setup has been evaluated elsewhere (Gomez-Navarro et al., 2013;2015). Both models are consistently driven by reconstructions of three external forcings: greenhouse gas concentrations in the atmosphere, long-term variations in Total Solar Irradiance (TSI) and variations of Earth's orbital parameters. The results of coupling the RCM with the GCM are hereafter referred to as MM5-70 ECHO-G.
Upward heat fluxes calculated within MM5-ECHO-G are used in this study as a predictor of deep-water formation. In this regard, it is important to note that MM5-ECHO-G does not include a high-resolution regional ocean model. Instead, Sea Surface Temperature (SST) variations are directly taken from the driving GCM and imposed as an additional boundary condition to the RCM. Still, the latter calculates the heat fluxes between the atmosphere and the surface, including the 75 prescribed ocean SST, according to meteorological conditions. Therefore, the heat fluxes within the RCM simulation are consistently obtained according to the large-scale atmospheric circulation prescribed by the GCM, but improved according to the additional information provided by regional circulation features driven by the high-resolution orography and land mask of the RCM. Thus, monthly upward heat flux evaluation is needed to identify times of year when enhanced heat loss occurs.
MM5-ECHO-G simulations for the past 1000 years showed that on average, the September-February period (autumn-winter) 80 ( Fig. A1) accounted for most of the annual upward heat flux. Hence, the analysis is performed over this period, referred to as winter hereafter for convenience although it encompasses autumn months as well.

Detection of the synoptic circulation pattern related heat flux variability
In order to find the spatial pattern that most clearly influences deep-water formation in the Aegean Sea, we first define two boxes delimiting deep-water formation areas in the Mediterranean: (1) Gulf of Lions (GL) (41.5 0 N -43 0 N, 3.5 0 E -6.5 0 E) 85 and (2) Aegean Basin (AB) (35.7 0 N -37.5 0 N, 23.5 0 E -27 0 E) (Fig. 1). Winter upward heat flux difference between the AB and the GL is then calculated, so that positive values are associated with enhanced deep-water formation in the AB, and conversely with respect to the GL. As the magnitude of the upward heat flux in both regions can be very different and we seek relative variations, both series were standardized before estimating the difference. Thereby, we obtain the following annual series: 90 where denotes the gradient of heat flux, and and the heat fluxes averaged for the aforementioned boxes and months after standardization. To find the spatial structure of atmospheric dynamic that most strongly affects the gradient of upward heat flux, the series (eq:1) is correlated with the winter mean geopotential height at 500 mb (hereafter Z500) This pattern can be interpreted as a mode where associated variability is most strongly associated with the differences in 100 deep-water formation between the AB and the GL.
Mathematically, this correlation map can be treated as a vector, and can be used to find an associated index by projecting the original Z500 field onto it. For this, the pattern has to be normalized first: where ( ) represents the normalised vector and "⋅" is the scalar product. Now, the index that represents the "weight" of this pattern throughout the last millennium, but optimized for the explanation of deep-water formation in the AB, is simply obtained as the projection of Z500 onto the pattern: 110 The variance of ′ ℎ (nhp stands for Northern Hemisphere pattern) can be compared to the total variance of the original field of Z500, which results in 11% of the variance of the whole field. A possible drawback of the index defined by (4) is 115 that it is affected by changes in global temperature, as geopotential height is closely related to temperature through the hypsometric equation. This implies that this index responds simultaneously to changes in atmospheric circulation, but also in global temperature. In order to overcome this problem keeping while the signal of the atmospheric dynamics, the spatially averaged Z500 is removed to define a new index. This is: where 500′( , ) = 500( , ) − ⟨ 500( , )⟩ and "⟨⟩" denotes spatial average.
Lastly, to complement this analysis and gain insight on the physical relationship between this circulation pattern and the variables that modulate heat flux at the surface, we perform composite analysis based on the Inhp index. This analysis is 125 carried out filtering out situations according to the aforementioned index values. In particular, dates corresponding to values over the 90 th percentile are selected, and the corresponding fields of the variable target of the analysis are averaged. This is repeated for the dates of the lower 10 th percentile values, and finally both averages are subtracted, yielding a map of anomalies that represents the impact of the index variability on the given variable. The rationale for this approach is that https://doi.org/10.5194/cp-2021-24 Preprint. Discussion started: 24 March 2021 c Author(s) 2021. CC BY 4.0 License.
under the null hypothesis of no relation whatsoever between the variability of the index to select dates and the variables, a 130 composite is equivalent to a random selection of dates, which statistically cancels out after taking differences. And conversely, large deviations from zero, either positive or negative, are indicative of strong influence on the index on this variable.

Calculating western/eastern Sea Surface Temperature proxy
In order to validate model simulations a western/eastern alkenone-based SST gradient was calculated. Western (W) (Moreno 135 et al., 2012;Nieto-Moreno et al., 2013;Sicre et al., 2016) and eastern (E) (Versteegh et al., 2007;Grauel et al., 2013;Gogou et al., 2016) marine SST proxies were first standardized (Table S1) and average values of a period before, during and after solar minimum events (Crowley et al., 2000) of both basins were calculated in order to evaluate the evolution of W-E gradients around solar minima ( Table 1). The length of the period chosen to calculate average SST values was equal to the duration of solar minimum. 140

Identification of the Northern Hemisphere atmospheric pattern most closely related to EMT-type events
The correlation coefficient between the AB-GL gradient and Z500' in the Northern Hemisphere ( for relationships between such modes and surface heat flux release in the Mediterranean. The results of this analysis revealed that the EA/WR mode most likely plays a major role in the deep-water formation in the AB. In our study, we applied a different strategy by undertaking a bottom-up approach, where the phenomenon to explain, i.e. changes in the locations of deep-water formation in the Mediterranean, is used to find a pattern based on physical processes. This type of approach enables more flexibility, as it allows the associated index to be optimized to explain the fraction of the atmospheric 155 variability that most directly affects the given phenomenon, hence maximizing the signal sought. Therefore, the fact that the pattern obtained through a completely different approach resembles the EA/WR structure reinforces the findings of Josey et al. (2011) and extends them over the longer temporal frame of the past 1000 years. Our results demonstrate that the index representing the "weight" of this correlation pattern through the last millennium, calculated in equations (4) and (5)  can be used as a proxy of EA/WR-like variability. This variability is associated with changes in the deep water formation 160 zones and, in particular, to the occurrence of EMT-type events.

Heat loss in Mediterranean Sea during EMT-type events
To gain insight on how the EA/WR variability mode is related to changes in heat exchange, we have obtained composites of various variables defined according to the Inhp index. To calculate the net heat exchange between sea and atmosphere, four components should be taken into consideration: (1) sensible heat flux, (2) latent heat flux, (3) longwave flux and (4)  165 shortwave flux. Winter net heat exchange is dominated by latent heat flux and to a lesser extent by sensible heat flux (Josey, 2003). These two components are driven by the product of the wind speed and the sea-air humidity and the sea-air temperature gradient (Josey et al., 1999). Therefore, to unravel the driving mechanisms of sea surface heat loss associated with the EA/WR-like mode, it is necessary to consider anomalous wind speed and air temperature fields (the atmospheric humidity field tends to follow air temperature and it is neglected) (Josey et al., 2011). 170 The composites of winter 2-m air temperature (i.e. near surface air temperature), upward heat flux and 10-m wind speed, obtained using the Inhp index, are shown in Fig. 3. The intensification of the spatial pattern described in the former section is associated with an increased western flux in the eastern Mediterranean, which favors the intensification of cold winds from the continental regions that, in turn, increase the upward heat flux in this region promoting deep-water formation. 175 Conversely, the pattern tends to reduce this zonal flow over the western Mediterranean, which therefore reduces the heat flux exchange there. These changes are summarized in the heat anomaly pattern of the top panel of Fig. 3, which is associated with an increased gradient between the AB and GL. This can also be appreciated in the near surface temperature pattern, with the warm (cold) anomaly in the western (eastern) Mediterranean driven by reduced (enhanced) zonal flow, and that agrees with the anomalies of heat exchange aforementioned. This pattern is due to the anomalous high-pressure system 180 centered over the North Sea that results in cold northwesterly airflow over the eastern Mediterranean and Black Sea, and a warmer southeasterly airflow in the western Mediterranean, generating a dipole in the heat exchange (Josey et al., 2011).
Usually in the Mediterranean Sea, the Levantine basin is characterized by higher temperatures, and high differences in the Evaporation-Precipitation balance facilitates LIW formation (Millot, 1999). Considering near surface air temperature varying in parallel with SST the predominance of this mode of variability results in reducing or compensating the average 185 temperature gradient in the Mediterranean.

EA/WR-like pattern variability during the past 1000 years and its influence on Mediterranean climate
Solar activity and last millennium EA/WR-like pattern variability Inhp are shown in Fig. 4a-b. After applying eq(5) to I'nhp, the global temperature signal, and thus the possible thermodynamic effect of solar forcing on the index, was removed. The residual signal is solely attributed to variations in the atmospheric circulation. When comparing the Inhp variability with solar 190 forcing (Crowley, 2000), a good correspondence is revealed for the analyzed interval. In particular, the Lomb periodogram https://doi.org/10.5194/cp-2021-24 Preprint. Discussion started: 24 March 2021 c Author(s) 2021. CC BY 4.0 License. (Fig. 4f) reveals significant peaks of both signals with a ~125 yr periodicity (frequency = 0.008 yr -1 ). After applying a Gaussian filter to both signals (Fig. 4e), frequency=0.008  0.001 yr -1 (i.e. 110-140 yr periodicity range), a strong relationship arises (r=-0.83, p<0.001). Interestingly, a similar variability has been previously documented (Baumgartner et al., 1992;Patterson et al., 2004Patterson et al., , 2005Cortina and Herguera, 2014, among others), attributed to solar activity expressed as 195 changes in the 14 C content of the atmosphere(Δ 14 C) (Neftel et al., 1981;Sonett, 1984;Stuiver and Braziunas, 1993). Our analysis suggests that solar activity minima with approximately 125 yr periodicity is related to Inhp enhancement and the ensuing expression of EA/WR-like atmospheric patterns. The latter is related to generation of EMT-type events through the physical relationship described above. Our results are in line with previous interpretations of circulation perturbation in the Mediterranean by Incarbona et al. (2016) who related solar irradiance lows with enhancement of EMT-type events, but we 200 restrict this relationship to a 125 yr cycle. The length of the simulation (1000 yr) could preclude detection of longer periodicities, and low resolution of the solar forcing proxy from year 1000 to 1700 (Crowley et al., 2000) could prevent evaluation of low periodicities such as the 88-yr Gleissberg cycle (Gleissberg and Schove, 1958). In fact, a 90-80-yr periodicity is present in the Inhp index (Fig. 4f), and could be responsible for the last EMT event.

205
Model simulations were also compared with oceanic proxy reconstructions during three singular periods of solar minima: (1) Maunder (1645-1715 yr), (2) Dalton (1790-1830 yr) and (3) Gleissberg (1900Gleissberg ( -1920 (Table 1) (Fig. 4c, 4d). Since EMTtype events co-occurred with freshening events in the Sicily channel , anomalous low  18 O seawater values in this region (Fig. 4d) should be contemporaneous with enhanced Inhp associated with EA/WR-like mode. This correspondence is precise during the Gleissberg and Dalton minima, and the 10-year lag observed between the freshening 210 event and the end of Maunder minimum (i.e. 1715 AD) is within its own chronological uncertainty (±25 years; (Incarbona et al., 2016)). On the other hand, the near surface temperature composite map revealed a reduced or compensated average temperature gradient between western and eastern Mediterranean basins during enhanced Inhp (EA/WR-like) pattern (Fig. 3).
The W-E gradient derived from SST proxy reconstructions (Fig. 4c), that is independent from model simulations, agrees with these results, showing higher values (i.e. increased difference between western and eastern basin SSTs) during solar 215 minima and an enhanced Inhp (EA/WR-like pattern).
The fact that the EA/WR-like mode dominated periods with increased differential upward heat flux between AB and GL, increased W-E temperature gradient and hence the occurrence of EMT-type events, does not exclude the influence of other important modes of atmospheric variability, such as positive phases of the North Atlantic Oscillation (NAO) (Incarbona et 220 al., 2016). The EA/WR-like pattern explains about 11% of atmospheric variability in the simulation, whereas studies based on Principal Component Analysis suggest that NAO accounts for about 40% of total variance, demonstrating the strong influence of this mode on North Atlantic atmospheric circulation. However, our model simulation results discard a direct influence of positive NAO during periods with an increased upward heat flux gradient between AB and GL, restricting its impact most likely to atmospheric preconditioning.

Conclusions
The MM5-ECHO-G simulation can be used to characterize the global EA/WR-like atmospheric mode in the Mediterranean region, which favors continental cold winds to penetrate into the AB, and blocks their influence in the GL. The model results predict an increase in the winter upward heat flux gradient between the AB and GL, enhanced Mediterranean deep-water formation in the eastern basin, with its impact on the circulation of the entire basin. At present, these oceanographic 230 conditions have been related to the EMT event, which demonstrates the suitability of this model configuration to study the variability of EMT-like events in the past. Our results show that during the past 1000 yr, a dominant EA/WR-like mode and EMT-type events, were contemporaneous with solar minima, likely related with cycles of approximately 125 and 80-90 years.
Moreover, the simulation results are in line with alkenone-based SST proxies that document an increase of the W-E gradient during these periods as consequence of winter-time northerly air outbreaks over the AB.