Contrasting Signals of the Westerly Index and North Atlantic Oscillation over the Drought Sensitivity of Tree-Ring Chronologies from the Mediterranean Basin

: Extreme drought events are becoming increasingly frequent and extended, particularly in Mediterranean drought-prone regions. In this sense, atmospheric oscillations patterns, such as those represented by the North Atlantic Oscillation (NAO) index and the Westerly Index (WI) have been widely proven as reliable proxies of drought trends. Here, we used the Standardized Precipitation–Evapotranspiration Index (SPEI), as a reliable indicator of drought, to investigate the drought sensitivity of tree-ring width data (TRW) from several long-lived tree species ( Abies borisii-regis , Abies cilicica , Abies pinsapo , Cedrus atlantica , Cedrus libanii , Pinus nigra , Pinus heldreichii ). NAO and WI relations with TRW were also investigated in order to identify potential non-stationary responses among those drought proxies. Our temporal and spatial analyses support contrasting Mediterranean dipole patterns regarding the drought sensitivity of tree growth for each tree species. The spatial assessment of NAO and WI relationships regarding SPEI and TRW showed on average stronger correlations westward with non-stationary correlations between annual WI index and TRW in all species. The results indicate that the drought variability and the inferred drought-sensitive trees species (e.g., C. atlantica ) are related to the NAO and the WI, showing that TRW is a feasible proxy to long-term reconstructions of Westerly Index (WI) variability in the Western Mediterranean region. Spatial variability of drought severity suggests a complex association between NAO and WI, likely modulated by an east–west Mediterranean climate dipole.


Introduction
Climate change-related drought events are becoming increasingly frequent and extended [1]. The influence of climate change on drought variability is particularly significant in drought-prone regions, such as the Mediterranean Basin, a biodiversity hotspot where warming and increasing droughts frequency are expected over the 21st century [2]. The Mediterranean mountains have a complex climatology due to the situation of transition between arid climates in the south to temperate climates in northern forests [3][4][5][6]. Thus, the predicted changes in drought trends are linked to shifting atmospheric circulation patterns with contrasting impacts at the regional scale, such as the North Atlantic Oscillation (NAO), one of the main atmospheric circulation patterns affecting the surface climate of continental Europe and North Africa [3][4][5][6].

Dendrochronological Dataset and Chronology Building
In each site, dominant or codominant living trees from undisturbed sites were sampled (Table  1) following dendrochronological methods [26]. To remove the age-related trend and non-climatic signals (e.g., past management or competition) and to preserve growth variation on inter-annual to multi-decadal timescales, we detrended raw tree-ring width series using a double detrending procedure; first, fitting a negative exponential function followed by 120-year-long splines, which maximized the high-frequency climatic information and minimized the non-climatic variance related to ontogenetic and local disturbance trends [39]. Detrending allowed transforming individual tree-ring series to dimensionless growth indices (hereafter TRWi) by dividing observed ring values by fitted values. The resulting series were pre-whitened by autoregressive models to remove temporal autocorrelation. We performed a variance stabilization technique for every single series for minimizing the effects of changing sample size throughout time. Mean site chronologies were then calculated using a bi-weight robust mean [40]. To assess the statistical quality of tree-ring width series, several dendrochronological parameters were calculated considering the common period 1901-2010 [26]: AC, first-order autocorrelation for raw tree-ring series (a measure of the Colored squares represent the studied species in green (Abies pinsapo, A. borissi-regis, A. cilicica), blue (Pinus nigra), and red (Cedrus atlantica, C. libani and P. heldreichii) populations, respectively. Climatic diagrams for each study site (mean temperature, red lines; precipitation, blue areas) are shown.

Dendrochronological Dataset and Chronology Building
In each site, dominant or codominant living trees from undisturbed sites were sampled (Table 1) following dendrochronological methods [26]. To remove the age-related trend and non-climatic signals (e.g., past management or competition) and to preserve growth variation on inter-annual to multi-decadal timescales, we detrended raw tree-ring width series using a double detrending procedure; first, fitting a negative exponential function followed by 120-year-long splines, which maximized the high-frequency climatic information and minimized the non-climatic variance related to ontogenetic and local disturbance trends [39]. Detrending allowed transforming individual tree-ring series to dimensionless growth indices (hereafter TRWi) by dividing observed ring values by fitted values. The resulting series were pre-whitened by autoregressive models to remove temporal autocorrelation. We performed a variance stabilization technique for every single series for minimizing the effects of changing sample size throughout time. Mean site chronologies were then calculated using a bi-weight robust mean [40]. To assess the statistical quality of tree-ring width series, several dendrochronological parameters were calculated considering the common period 1901-2010 [26]: AC, first-order autocorrelation for raw tree-ring series (a measure of the year-to-year growth similarity); MS, mean sensitivity (a measure of the year-to-year variability in width or other variables over consecutive tree rings); rbt, mean between-trees Atmosphere 2020, 11, 644 4 of 22 correlation of indexed values; SNR, signal-to-noise ratio; PC1, the percentage of variance accounted for by the first principal component which accounts for the common variability in growth among all trees at each site; and EPS, expressed population signal (a measure of the statistical quality of the mean site chronology as compared with an infinite perfectly replicated chronology). The chronology segment with EPS > 0.85 was considered as the common interval that was statistically reliable [41].

Climate Data, Drought Index, and Large-Scale Atmospheric Oscillations
Climate data over the 20th century in each of the nine study sites were obtained from a high-resolution gridded data from CRU TS 3.0 (University of East Anglia Climate Research Unit (CRU)) [42] with 0.5 • of resolution. To assess the climate sensitivity of the individual site chronologies with respect to monthly mean temperature and monthly precipitation, the common period of 1930-2010 was considered. To evaluate the effect of drought intensity and duration on tree growth, we used the Standardized Precipitation-Evapotranspiration Index (SPEI), which is calculated using precipitation and temperature data from the same homogeneous and spatially dense dataset of local observatories as the CRU climate data [21] by using the SPEI library in the R environment [43]. The SPEI monthly values were calculated for the studied sites considering the period of 1901-2010 and from 1-to 24-month-long scales, since this is a multi-scalar index to observe the cumulative effects of drought. The SPEI accounts for the negative effect of warmer temperatures on water availability by statistically modelling cumulative water balances. For the SPEI, NAO, and WI comparison, we used 3-month (hereafter SPEI3) scales, since [22,23] showed the higher correlations among atmospheric circulation indices and drought for short-term scales in the Mediterranean.
We assessed spatial and temporal effects of the North Atlantic Oscillation (NAO) and the Westerly Index (WI) on tree radial growth of each site, given their wide spatial scale of influence [44]. The NAO is a north-south dipole that determines the position of the Icelandic low pressure and the Azores high-pressure systems and, therefore, the climatic patterns in Southwestern Europe, which allows extending its record beyond the 19th century [28]. The NAO monthly data were downloaded from the Climate Research Unit webpage (https://crudata.uea.ac.uk/cru/data/pci.htm; [42]). The Westerly Index (WI) quantifies the Atlantic air masses that enter Europe by measuring the frequency of days with westerly winds over the English Channel for each month since 1685. The WI provides information, better in winter and summer than in transition seasons, about temperature and precipitation anomalies over large areas of Europe. Usually, its climatic signal is similar to the one provided by NAO, but its signal is better than that of NAO during summer and very similar during the rest of the year. It is important to notice that even when the climatic signals of NAO and WI tend to be similar, it has been described that both indices were disconnected during some decades of the 20th century and during some longer periods in the previous centuries. It is also important to highlight that the WI has been a completely instrumental index during its whole duration (more than 350 years), and this cannot be said for any equivalent index (NAO reconstructions are mostly based in proxy data for its reconstructions before the 19th century) (see [20] for further information). Thus, the WI seems to be a better index than NAO to calibrate and analyze the climatic signal of climatic records with lengths of several centuries.

Statistical Analysis
To compare the SPEI, NAO, and WI signals in tree-ring growth, we used Pearson correlations and spatial patterns of the relationships between indices using the KNMI Climate Explorer (http://climexp.knmi.nl/, [45]). Temporal trends of SPEI3 for each site and regional NAO and WI indices were estimated using Mann-Kendall coefficients, respectively. To quantify the influence of monthly climate data (mean temperature and precipitation) and SPEI3, NAO, and WI indices on growth, bootstrapped correlations were carried out for the common period of 1901-2010. These correlations were calculated from previous August to current December, i.e., prior to and during the year of tree-ring formation, based on previous analyses of the study species' growth responses to climate [11]. To assess whether these relationships were stable through time, we calculated 30-year-long moving correlations, taking into consideration only those climate variables strongly and significantly (p < 0.05) correlated with TRWi. We also calculated Pearson correlations between TRWi and SPEI obtained for 1to 24-month long scales [46].
Finally, we built a transfer function model using ordinary least square regressions based on significant (p < 0.01) sites' signals related to mean temperature, total precipitation, SPEI3, NAO, and WI indices. The transfer function model used TRWi as the predictor and the climate variables as the predictands. The skill of TRWi for reconstructing past climate was tested by a split-sample procedure and verified the stability of the model [47]. This procedure divided the study period into two equal subperiods, which were used as calibration and verification periods, respectively. The models were evaluated using regression statistics such as R 2 , the reduction of error statistic (RE), the coefficient of efficiency (CE), and the Durbin-Watson test (DW). The value of RE theoretically ranges from minus infinity to +1.0, and an RE value greater than 0 is considered positive skill of the model [26]. The calculated transfer function was then applied to the site chronology to obtain the time series of the reconstructed abovementioned climate variables and indices. Bootstrapped and moving correlations, as well as the transfer function model, were calculated using the software treeclim [48]. All stages of chronology building and data analyses were performed using the package dplR [49] in the R environment [50].

Atmospheric Circulation Indices and Drought
Spatial correlations between NAO, WI, and SPEI3 were observed with large differences across the Mediterranean Basin ( Figure 2). The winter drought variability was well explained by NAO and WI, controlling the precipitation patterns. Annual NAO correlated positively with annual WI (r = 0.48), winter WI (r = 0.79), spring WI (r = 0.50), summer WI (r = 0.34), and autumn WI (r = 0.56) ( Table 2).
The relationships between annual SPEI3 and annual WI reported disparate results along the Mediterranean Basin ( Table 2). The highest correlations were observed in the western region for the A.  (Table 2). C. atlantica and P. nigra sites in the Western Mediterranean also showed negative correlations between annual SPEI3 and winter WI (C. atlantica and P. nigra) and annual and autumn values for P. nigra. However, no significant correlations were observed for the Central Mediterranean region, while in the eastern area, only A. cilicica showed a significant negative correlation between annual SPEI3 and annual and summer WI index ( Table 2).
Spatial correlations between NAO, WI, and SPEI3 were observed with large differences across the Mediterranean Basin ( Figure 2). The winter drought variability was well explained by NAO and WI, controlling the precipitation patterns. Annual NAO correlated positively with annual WI (r = 0.48), winter WI (r = 0.79), spring WI (r = 0.50), summer WI (r = 0.34), and autumn WI (r = 0.56) (   Table 2. Correlation matrix of the seasonal (wi, winter; sp, spring; su, summer; au, autumn) and annual WI (dependent variable) versus annual NAO and annual SPEI3 for each of the nine different sites for 1946-2010. Correlation values in bold letters are significant at the p < 0.05 level.

Chronology Characteristics
Trees from the Western Mediterranean were older on average (264 years) but formed narrower rings (0.79 mm) than those from central (212 years, 1.46 mm) and eastern sites (263 years, 0.96 mm) (Tables 1 and 3; Figure A1). The year-to-year tree-ring autocorrelation (AC) and interannual mean sensitivity (MS) of growth were similar at the extremes of the west-east dipole, which suggests that climatic constraints on growth are more important for these edge forests than in the Central Mediterranean region (Table 3). Western sites presented higher inter-series correlation and higher mean sensitivity compared to central and eastern regions, respectively, although these comparisons must  Figure A1). The remaining dendrochronological statistics confirmed the coherence among individual sites and further composited series for each region (Table 3).

Climate-Growth Relationships
Mediterranean Abies species showed higher negative correlations with mean temperatures during the previous autumn in A. pinsapo and A. borissi-regis and late spring and early summer temperatures for the three Abies species (Figure 3). TRWi was positively influenced by late spring/early summer precipitation in all sites and species except P. heldreichii in the central region and P. nigra and A. pinsapo in the west, which showed the highest correlation with precipitation in the previous summer and autumn, respectively ( Figure 3). The highest correlation with temperature was observed during the previous autumn in the western P. nigra site, while the lowest was observed in the eastern C. libani ( Figure 3). Total precipitation showed a higher signal in western C. atlantica and central low-elevation A. borissi-regis in the Central Mediterranean ( Figure 3).

Climate-Growth Relationships
Mediterranean Abies species showed higher negative correlations with mean temperatures during the previous autumn in A. pinsapo and A. borissi-regis and late spring and early summer temperatures for the three Abies species (Figure 3). TRWi was positively influenced by late spring/early summer precipitation in all sites and species except P. heldreichii in the central region and P. nigra and A. pinsapo in the west, which showed the highest correlation with precipitation in the previous summer and autumn, respectively ( Figure 3). The highest correlation with temperature was observed during the previous autumn in the western P. nigra site, while the lowest was observed in the eastern C. libani ( Figure 3). Total precipitation showed a higher signal in western C. atlantica and central low-elevation A. borissi-regis in the Central Mediterranean ( Figure 3).   Table 1.

Relationships between Growth SPEI, NAO, and WI
The relationships between TRWi and SPEI showed a significant association between growth and intensity and severity drought in West P. nigra and C. atlantica but not in the A. pinsapo site ( Figure 4). We found a positive and significant correlation with SPEI recorded at a 3-12-month-long scale in these western sites, corresponding to short wet or dry periods at a high frequency occurrence ( Figure 4). In central and eastern sites, significant SPEI and TRWi relationships were found for 1-8-month-long scales, higher in lower elevation (A. borissi-regis) and eastern (C. libani) sites ( Figure 4). Lower drought-growth correlations were observed in Abies species. Finally, we observed similar high-frequency TRWi (growth) evolution patterns in comparison with the mean SPEI3 drought index for May and June, with growth increases and reductions evolving in parallel to SPEI3 evolution ( Figure 5). The composite chronologies for each region (west, central, and east) Figure 3. Bootstrapped tree growth-climate associations for the studied sites assessed by relating ring-width indices (TRWi) to seasonal and monthly climatic variables (mean temperature and precipitation) for the 1946-2010 period. The temporal window of analyses ranges from the previous August (previous-year months and seasons are abbreviated in uppercase letters) to the current December (current-year months are abbreviated in lowercase letters). Significance levels are shown in darker grey color (p < 0.01). Sites' codes are as in Table 1. Table 3. Dendrochronological statistics based on radial growth series for each studied site. The statistics and mean basal tree-ring width values were calculated for the common period of 1930-2010. All dendrochronological statistics were calculated for residual series except autocorrelation (AC), which was obtained for raw tree-ring width data.

Relationships between Growth SPEI, NAO, and WI
The relationships between TRWi and SPEI showed a significant association between growth and intensity and severity drought in West P. nigra and C. atlantica but not in the A. pinsapo site ( Figure 4). We found a positive and significant correlation with SPEI recorded at a 3-12-month-long scale in these western sites, corresponding to short wet or dry periods at a high frequency occurrence ( Figure 4). In central and eastern sites, significant SPEI and TRWi relationships were found for 1-8-month-long scales, higher in lower elevation (A. borissi-regis) and eastern (C. libani) sites ( Figure 4). Lower drought-growth correlations were observed in Abies species. Finally, we observed similar high-frequency TRWi (growth) evolution patterns in comparison with the mean SPEI3 drought index for May and June, with growth increases and reductions evolving in parallel to SPEI3 evolution ( Figure 5). The composite chronologies for each region (west, central, and east) showed significant correlations between MJ SPEI3 drought index and TRWi, representing high coherence between concatenated relations among SPEI3, NAO, and WI ( Figure 2, Table 2).
Monthly NAO and WI showed long-term variability during the 20th century, but stationary trends (Table 4). However, SPEI3 showed strong variability between decades and years with wet periods (1960s, 1970s in the west; 1920s, 1940s in the central region; and 1960s, 1990s in the east) alternating with dry ones (1950s, 1980s in the west; 1950s, 1980s in the central region; and 1970s, 1990s in the east) ( Figure 5). However, long-term drought trends were different along the east-west dipole, mainly in the second half of the 20th century (Table 4).
Atmosphere 2020, 11, x FOR PEER REVIEW 11 of 23 showed significant correlations between MJ SPEI3 drought index and TRWi, representing high coherence between concatenated relations among SPEI3, NAO, and WI ( Figure 2, Table 2).   Table 1.   Table 1. Monthly NAO and WI showed long-term variability during the 20th century, but stationary trends (Table 4). However, SPEI3 showed strong variability between decades and years with wet periods (1960s, 1970s in the west; 1920s, 1940s in the central region; and 1960s, 1990s in the east) alternating with dry ones (1950s, 1980s in the west; 1950s, 1980s in the central region; and 1970s, 1990s in the east) ( Figure 5). However, long-term drought trends were different along the east-west dipole, mainly in the second half of the 20th century (Table 4). period. The plotted drought index and NAO (dashed cyan line; winter for west, spring for central, and summer for east) were most strongly related to the tree-ring width indices (TRWi) in most study sites (see Figure 5). Pearson correlations (r) between TRWi and the indices and associated probability values (P) are shown for each site (see Figure 6 for more correlations).

SPEI 3 West Central East
C. atlantica (west), C. libani (east), A. borissi-regis (central), and A. cilicica (east) showed the highest positive associations with SPEI3, followed by P. nigra in the western, eastern, and central regions, respectively ( Figure 6). TRWi was negatively correlated with Febraury NAO in C. atlantica and positively with the previous autumn in A. pinsapo and P. nigra (west) and current summer in Central Mediterranean species (Figure 6). Some eastern sites showed positive relationships with the current summer, autumn, and winter NAO indices (A. cilicica, P. nigra) and the central (P. heldreichii) region ( Figure 6). As expected, the WI index showed the highest signal in the West Mediterranean for C. atlantica (negative) and P. nigra (positively) during autumn and winter, respectively ( Figure 6). However, it is interesting to notice that all the species showed at least one month with significant associations between TRWi and WI index both in the central and in the eastern basin ( Figure 6). were most strongly related to the tree-ring width indices (TRWi) in most study sites (see Figure 5). Pearson correlations (r) between TRWi and the indices and associated probability values (P) are shown for each site (see Figure 6 for more correlations).
C. atlantica (west), C. libani (east), A. borissi-regis (central), and A. cilicica (east) showed the highest positive associations with SPEI3, followed by P. nigra in the western, eastern, and central regions, respectively ( Figure 6). TRWi was negatively correlated with Febraury NAO in C. atlantica and positively with the previous autumn in A. pinsapo and P. nigra (west) and current summer in Central Mediterranean species (Figure 6). Some eastern sites showed positive relationships with the current summer, autumn, and winter NAO indices (A. cilicica, P. nigra) and the central (P. heldreichii) region ( Figure 6). As expected, the WI index showed the highest signal in the West Mediterranean for C. atlantica (negative) and P. nigra (positively) during autumn and winter, respectively ( Figure 6). However, it is interesting to notice that all the species showed at least one month with significant associations between TRWi and WI index both in the central and in the eastern basin ( Figure 6).   Figure 7 shows moving correlations between TRWi and SPEI3, NAO, and WI for the months with the best long-term correlation between these indices and TRWi (see Table A1 and Figure 6). The most stable correlations are obtained when SPEI3 series are analyzed, in particular, in the Western and Eastern Mediterranean regions. In the Central Mediterranean area, only A. borissi-regis is significantly associated with SPEI3. On the other hand, NAO and WI show a much more variable association with TRWi series. In the Western Mediterranean area, the WI shows a stable negative correlation with A. pinsapo and a (mostly significant) negative association with C. atlantica during the second half of the analyzed period. On the other hand, NAO only shows a relatively stable positive association with C. atlantica during the second half of the analyzed period.
In the Central Mediterranean region, no clear associations were observed. The WI was positively associated (mostly significantly) with A. borissi-regis and, during some decades in the middle of the analyzed period, positively associated with P. heldreichii. Associations with P. nigra were not significant. On the other hand, NAO changed the sign of its association with A. bosissi-regis along the analyzed period but showed a more stable positive association (near or above the significance level during a significant part of the analyzed period) with P. nigra and P. heldreichii.  Figure 7 shows moving correlations between TRWi and SPEI3, NAO, and WI for the months with the best long-term correlation between these indices and TRWi (see Table A1 and Figure 6). The most stable correlations are obtained when SPEI3 series are analyzed, in particular, in the Western and Eastern Mediterranean regions. In the Central Mediterranean area, only A. borissi-regis is significantly associated with SPEI3. On the other hand, NAO and WI show a much more variable association with TRWi series. In the Western Mediterranean area, the WI shows a stable negative correlation with A. pinsapo and a (mostly significant) negative association with C. atlantica during the second half of the analyzed period. On the other hand, NAO only shows a relatively stable positive association with C. atlantica during the second half of the analyzed period.
In the Central Mediterranean region, no clear associations were observed. The WI was positively associated (mostly significantly) with A. borissi-regis and, during some decades in the middle of the analyzed period, positively associated with P. heldreichii. Associations with P. nigra were not significant. On the other hand, NAO changed the sign of its association with A. bosissi-regis along the analyzed period but showed a more stable positive association (near or above the significance level during a significant part of the analyzed period) with P. nigra and P. heldreichii.
Finally, in the Eastern Mediterranean, NAO and WI showed very similar associations. Both of them were positively associated with A. cilicica and P. nigra, coming in and out of the level of significance, and both of them were negatively correlated with C. libani during the final half of the analyzed period; however, the WI was positively correlated during the early decades of this period.  Table 1, and the selected period for each site is shown in Table A1. Sites' codes are as in Table 1, and the selected period for each site is shown in Table A1.

Exploring WI Reconstructions from Tree Rings
We explored whether TRWi could emerge as a potential proxy for the WI reconstruction in the Mediterranean Basin (Table 5, Figure A2). The regression models for C. atlantica, the site with higher correlations for SPEI12, NAO, and WI (Figures 3, 4 and 6), were adjusted explaining 52% (SPEI), 29% (NAO), and 28% (WI) variability during the period of 1903-2001 (Table 5). The calibration/verification statistics indicated that there is reconstructive potential in the WI model, even if the WI signal is weak (Table 5; Figure A2). The two subperiods for calibration and verification of the regression models showed significant correlations for summer SPEI12

Discussion
Regional climatic changes affect societies and natural ecosystems, endorsing the basis for efficient adaptation measures [1]. Several studies support that regional climates are the result of the interaction of large-scale dynamics with the orography and physical properties at the regional and local scales. Notwithstanding, improving our knowledge about regional and local climate linkages, including extremes, remains highly relevant [7,51]. Here, we obtained reliable information on the amplitude and changes of past climate dynamics at regional scales using tree rings in relation with drought variability (SPEI) and atmospheric circulation patterns (NAO, WI). Despite significant recent research efforts, detailed interactions among regional drought variability and atmospheric patterns are not sufficiently understood yet [8]. Furthermore, the analysis of long-term climate series at large spatial scales provides relevant information on climate variability and trends in relation to climate change [1, 52,53]. Since the climate of the Mediterranean Basin is characterized by seasonal drought during the dry and warm summer conditions [54,55], this region is potentially sensitive to climate change, with periods of water shortage negatively affecting the tree growth and dynamics of forest ecosystems, as well as to several socio-economic implications within and beyond the region [56,57]. Our results showed higher drought sensitivity in the western areas, compared to the eastern part, according to significant tree growth sensitivity due to siring drought intensity caused by recent changes on large atmospheric oscillations patterns [11].
Since the Mediterranean Basin's climate is modulated by atmospheric patterns [7], their changes influence the local climate and drought conditions [58]. In terms of the modulation of the temperature, positive phases of the winter NAO show a high correlation with cold air temperatures in the Southwestern Mediterranean area, while the opposite response is observed during negative NAO phases [3,4]. Similarly, the Central-eastern European and Northern Turkish climates showed positive associations with NAO, while the NAO influenced the winter temperatures in Northern Italy (Figure 2) [20,24,58]. Several studies showed positive relationships between winter NAO and lower precipitation in the Western and Northern Mediterranean areas [13,59]. Conversely, the NAO presented weak relationships during other seasons, except for the NAO changes which translate into variations in precipitation in the Southern Mediterranean [7,11,13,23,58].
The influence of NAO on drought severity (SPEI) and its relationships with the WI was confirmed in the Mediterranean Basin [23]. The WI index accounts for climate variability during the summer season and higher temperatures during winter in Western Europe (Figure 2). A positive WI was related with lower temperatures in Northern Europe and warmer conditions in the Eastern Mediterranean area [20]. However, high frequency of westerly winds (positive WI) increases cloudiness and humidity in Northern Europe, with decreasing precipitation and humidity towards the Mediterranean area. Winter and summer seasons show an extended decrease of precipitation, covering the west and east of the Mediterranean Basin, while decreasing spring precipitation prevails mainly in the west of the Mediterranean [20,23]. Many previous studies showed the relationships between NAO and drought in Southern Europe and North Africa [23], although the WI index captures a similar signal to the NAO during summer and winter seasons, especially during dry spring-summer seasons in Western and Central Europe [23]. While the NAO is a pattern that considers wind strength, the WI is an index that only considers the prevalence of westerly winds [20]. This difference is important since it has been proven that the WI, unlike the NAO, is a better indicator of precipitation with robustness throughout the year in Northern and Central Europe. However, the WI presents difficulty to capture temperature anomalies that are determined by the advection and intensity of air masses, whereas NAO can capture it more easily. Nevertheless, our study is the first attempt to understand the combined influence of these patterns in the effects of drought across Mediterranean ecosystems. As [20] showed, NAO and WI have many similarities in behavior and effects on the European continent and can complement each other to improve our knowledge on both short-and long-term drought effects across European forests [23].
Based on our findings about the relationships among tree-ring data (TRWi) and NAO, WI, and SPEI, we conclude that drought, as reflected by the SPEI, is the main limiting factor of tree growth in agreement with previous regional studies [60]. This pattern has been already observed in different studies showing higher correlation between SPEI and tree growth variation than NAO or other atmospheric patterns [11,12,16]. This result might be related to the temporal scale of drought. Hence, while SPEI is able to capture short-term weather conditions and extreme weather events more accurately when seasonal droughts occur, NAO and WI reflect large spatial and temporal atmospheric patterns, mainly sensitive to intra-annual periods, such as winter, where tree growth stops and drought conditions remain low [10,13,15].
An important issue raised by our study is the lack of stationary relationships among NAO, WI, and the climate sensitivity observed in the different tree species (TRWi) (Figure 7). It is important to note that, although the correlations might be significant for the entire time period, decreasing correlations may be present at some shorter time spans, where the relationship is weakened. Here, sequential coupling and de-coupling of climate sensitivity could be related to contrasting and sometimes opposite local environmental responses over time and also among nearby forest ecosystems during a given time period [11]. Non-stationary climate sensitivity relies on ecosystem development and its complex response to climate changes, even in the same region, supporting that these responses are highly influenced by the landscape and the environmental evolution of the area [58,61,62]. In addition, these shifting environmental responses might be amplified in our tree species during the last centuries, suggesting a different resilience and sensitivity to climate variations between the studied mountain forest ecosystems [2,11,53]. Anthropogenic influences should be taken into account, as they also appear to modulate the effects of solar and atmospheric cycles in the tree rings' environmental signals [62]. The obtained climate sensitivity of the Mediterranean forests studied here provides paleoenvironmental records likely influenced by several drivers, including climate variations but also human-induced environmental changes.
However, our results showed that depending on the Mediterranean zone, the inter-annual NAO variations can explain large-scale growth variability ( Figure 5). For the western area, we found higher associations with TRWi for autumn and winter NAO for A. pinsapo and P. nigra ( Figure 6). These results are consistent with other studies that associate an increase in the temperatures during the previous winter with an increase in the growth rates due to an early bud break [63]. This relationship has been proven in studies that show how the NAO strongly modulates the inter-annual variability of precipitation in the Western Mediterranean [62]. This situation explains the negative impact of spring NAO in C. atlantica that translates into a lower soil water reserve and availability, which acts as a limiting factor in the tree radial growth during the growing season. In contrast, a positive NAO phase in summer increases growth rates in Central Europe [64]. Regarding the Eastern Mediterranean region, the summer NAO conditions influence water availability during the growing season [10,16,23], as does the autumn NAO, consequently affecting tree-ring growth of eastern forests [10,62].
Regarding the relationship between tree growth and WI, we observed higher correlations with Western Mediterranean forests ( Figure 6). For instance, P. nigra and A. pinsapo showed positive and negative WI signals in tree growth during winter, summer, and spring seasons, respectively ( Figure 6)-periods in which there was an increase in temperature and a decrease in precipitation in western regions [11]. In this case, the increase in temperature is related to the onset of growing season of P. nigra and C. atlantica (Figure 3). However, autumn WI negatively affected the tree growth of C. atlantica in relation with the influence of WI on the precipitation and temperature patterns (Figures 2 and 6). The autumn WI was related with drought that consequently affected the growth [5,13]. On the other hand, the relationships between WI and growth in Eastern Mediterranean forests were higher with previous winters (A. cilicica and C. libani) and March and July in P. nigra ( Figure 6). This relationship with winter WI was consistent with drought-growth relationships in this area. Our results showed the ability of the WI to explain how the tree growth in the Central Mediterranean area is mainly limited by spring conditions; with a positive effect on growth during autumn. This finding is consistent with those obtained in other studies on the WI, where we can see how its potential is located in Western and Northern Europe, presenting difficulties in explaining the behavior of droughts in Southern Europe [23].
The combination of SPEI, WI, and NAO allows us to study the effects of large-scale atmospheric patterns in the regional growth of tree species in drought-prone Mediterranean areas. While the NAO is better related with large-scale drought periods, the WI allows us to study shorter drought effects, having a stable seasonal influence on drought severity [23]. This combination of characteristics between both atmospheric patterns seems promising for a better understanding and reconstruction of drought periods in the Western Mediterranean area ( Table 5). The potential of using TRWi as a proxy in annual climate reconstructions has been proven in numerous studies [65,66]. However, to our knowledge, this is the first study using TRWi information to understand and reconstruct long-term regional climate patterns in the Mediterranean Basin related with the WI. Our results evidence a strong connection between drought and large atmospherics patterns, especially the interconnected WI and NAO, which is reflected in annual tree-ring chronologies (Table 5). Although tree-ring information from the Mediterranean basin is not scarce, many of the currently available chronologies do not cover centuries to address long-term climate reconstructions, due to the scarce long-lived trees [67]. The oldest trees in our study (C. atlantica) were found in the Atlas Mountains in Morocco [24,67]. This species showed the highest correlations between TRWi and summer precipitation in the Western Mediterranean [68], while negative correlations between TRWi chronologies and summer precipitation were obtained from some Eastern Mediterranean mountains [11], which have been demonstrated to reflect a distinct temperature signal, most likely reflecting the inverse relationship between summer temperature and precipitation ( Figure 5). Nonetheless, Eastern Mediterranean drought responses are also available, for instance, in Southern Turkey [33,69]. Hence, the use of multiple species with different intra-annual periods of climate-growth sensitivity provides complementary information. Even with these shortcomings, TRWi data in our study proved to reconstruct local climatic variables and to detect significant relationships with large-scale atmospheric oscillations such as NAO or WI [23].
Overall, our results support significant relationships among Mediterranean tree-ring chronologies and the SPEI3, in relation with atmospheric oscillations ( Table 5), suggesting that TRWi datasets may be widely suitable proxies to perform reliable reconstructions of the spatial and temporal patterns of drought frequency and intensity in relation with NAO and WI oscillations in the Western Mediterranean area. Thus, despite the NAO and WI presenting periods related to tree growth, the reliability seems much lower than that of the SPEI. Then, although we obtained limited adjustment, the use of drought-sensitive tree-ring chronologies would improve the suitability of TRWi to reconstruct millennial WI patterns. In summary, our understanding of variations in short-duration and extensive drought across the Mediterranean basin is still limited, as a small number of well-dated high-temporal-resolution proxies are available [7] and due to the non-stationary relationships among TRWi, NAO, WI, and SPEI (Figure 7). In this context, in order to improve climate reconstructions, further studies could identify new potential long-term tree-ring series with higher sensitivity to regional-scale circulation patterns, such as WI and NAO indices [70]. Despite the limitations of our results using TRWi, this is a first attempt to understand the spatiotemporal associations among drought patterns and the interconnected NAO and WI oscillations over the Mediterranean Basin, although these relationships should be considered carefully given the relative level of non-stationarity (Figure 7). Further research is needed to understand past and future spatiotemporal variability of the tree growth associations with NAO and WI.

Conclusions
Although our results are still limited to draw detailed conclusions on the spatiotemporal high-resolution links among drought patterns and TRWi, NAO, and WI, we provide valuable information about the west-east dipole effects on how local climate variables, atmospheric patterns, and drought affect forest growth in the Mediterranean Basin. The NAO and WI explained high-frequency growth variability of western C. atlantica and P. nigra forests, and partially of Central and Eastern Mediterranean forests. Our study also shows that tree-ring widths are a potential proxy to long-term reconstructions of WI variability in the Western Mediterranean region. Based on our results, the climate change towards drier conditions could influence western, central, and eastern mountain forests differently in relation with changes in non-stationary atmospheric oscillation patterns. To improve the accuracy of future climate scenarios, local and large-scale climate drivers must be reconstructed using complementary proxies such as high-temporal-resolution tree rings, which allow to extend back centuries. Funding: This research was financially supported by the project LESENS (RTI2018-096884-B-C33) from the Spanish Ministry of Science, Innovation, and Universities and by the VULBOS project (UPO-1263216) from the ERDF funds of the Andalusian regional government 2014-2020. P.C.-G. was funded by the Environmental predoctoral grant by the Fundación Tatiana Pérez de Gúzman El Bueno 2018. This research was supported by the research groups RNM-356 "Ciencias de la Tierra y de la Atmósfera" and RNM-313 "Estructura y Función de Ecosistemas Mediterráneos".   Table 5 for details).    Figure A1. Site chronologies (tree-ring width indices, TRWi; continuous lines) obtained for the studied sites from western (left column), central (central), and eastern (right) regions of the Mediterranean and the number of measured series (grey areas) for each site. Figure A2. Comparison between the reconstructed (color lines) and actual (black line) summer SPEI at a 12 month scale, previous December NAO (NAOdt−1), and previous November-December Westerly Index (WIndt−1) for the Cedrus atlantica site (see Table 5 for details).  Figure A2. Comparison between the reconstructed (color lines) and actual (black line) summer SPEI at a 12 month scale, previous December NAO (NAOd t−1 ), and previous November-December Westerly Index (WInd t−1 ) for the Cedrus atlantica site (see Table 5 for details).