Reduced Temperature Sensitivity of Maximum Latewood Density Formation in High-Elevation Corsican Pines under Recent Warming

Maximum latewood density (MXD) measurements from long-lived Black pines (Pinus nigra spp. laricio) growing at the upper treeline in Corsica are one of the few archives to reconstruct southern European summer temperatures at annual resolution back into medieval times. Here, we present a compilation of five MXD chronologies from Corsican pines that contain high-to-low frequency variability between 1168 and 2016 CE and correlate significantly (p < 0.01) with the instrumental April–July and September–October mean temperatures from 1901 to 1980 CE (r = 0.52−0.64). The growth–climate correlations, however, dropped to −0.13 to 0.02 afterward, and scaling the MXD data resulted in a divergence of >1.5 ◦C between the colder reconstructed and warmer measured temperatures in the early-21st century. Our findings suggest a warming-induced shift from initially temperature-controlled to drought-prone MXD formation, and therefore question the suitability of using Corsican pine MXD data for climate reconstruction.


Introduction
Southern Europe is predicted to experience increasing aridity in the 21st century mainly due to rising temperatures [1][2][3], and the intensity, frequency, and duration of heatwaves and droughts are expected to increase [4,5]. The precise assessment of climate changes requires proxy-based reconstructions of pre-instrumental conditions to place recent changes into a long-term context. The employment of trees as high-resolution natural climate archives can provide such valuable insights into past climate fluctuations [6]. Although ancient trees growing under extreme conditions and still preserved in their natural state are difficult to find, dendroclimatic studies in Corsica offer the opportunity to reconstruct southern European climate variability over several centuries [7][8][9]. Due to the great variety of microclimates, the Corsican mountains feature a high ecological biodiversity [14] including two native pine species: Maritime pine (Pinus pinaster AITON) and Corsican Black pine (Pinus nigra ssp. laricio MAIRE). The former mainly covers lower elevations <1300 m a.s.l. that are strongly influenced by anthropogenic activities. In contrast, Black pine forests are in the high elevation belt between 1400-1800 m with remnants of natural stands near the upper treeline ( Figure 2a). The drought-and frost-tolerant P. nigra [15,16] may reach ages up to 850 years [7], ranking these trees among the oldest in the Mediterranean Basin [17]. Particularly resistant to parasites and decomposition due to their high resin content [18], relict pine stems can remain for centuries on the rocky slopes ( Figure 2b). The presence of these well-preserved trunks in the vicinity of old living stands offers the opportunity to develop multi-centennial tree-ring chronologies.
In drought-dominated regions, tree-growth is typically determined by the availability of water resulting from the mutual interaction of temperature, precipitation, and evapotranspiration [19,20]. Due to low water supply in summer, Mediterranean forest productivity is dependent on the frequency and amount of rainfall events before and during the growing season. Consequently, tree-ringwidth (TRW) chronologies have been used as proxies to reconstruct precipitation [21][22][23][24] and hydroclimatic variability across southern Europe [25][26][27][28][29]. High-resolution summer temperature reconstructions are restricted to other tree-ring parameters, such as the maximum latewood density (MXD), as the TRW is less temperature sensitive in Mediterranean environments [30].
MXD chronologies are generally less influenced by biological memory effects and correlate more closely with instrumental temperature data [31]. However, compared to the extensive networks in Fennoscandia [32,33] and the Alps [34], the availability of temperaturesensitive density records spanning 500+ years in the Mediterranean is restricted to the Spanish Central Pyrenees [35,36], the Cazorla Natural Park in southern Spain [37], and the Pindos National Park in north-western Greece [38,39]. With respect to the central Mediterranean region, comparably long-term MXD chronologies are still missing.
Nonetheless, two annually resolved temperature reconstructions provide initial insight into the region's past climate variability: Leonelli et al. [40] developed a 300-year temperature reconstruction based on a multi-site MXD network in Italy, while the only temperature reconstruction for Corsica was derived from stable carbon isotopes (δ 13 C) extending 560 years into the past [8]. Hence, a long-term MXD-based temperature reconstruction from Corsica is still lacking and remains key to unfold the region's climate history into the early second millennium.
Here, we present the longest MXD record for the Central-Western Mediterranean based on 35 P. nigra trees growing at three high-elevation sites on the island of Corsica (France). Covering the period 1168-2016 CE, the dataset extends existing MXD measurements (1518-1980 CE) [41] back to medieval times as well as into the 21st century. To evaluate the application of Regional Curve Standardization (RCS), age-band chronologies were produced and different chronology variants were correlated against regional climate data. In this study, we (1) assess climate signals in five chronology variants of the new Corsican MXD compilation, (2) compare new density data with existing records and summer temperature reconstructions across the southern European region, and (3) evaluate the potential of the Corsican MXD network for climate reconstruction.

Wood Density Data and Chronology Development
In 2017, 39 increment cores from 20 living and relict Black pines (Pinus nigra ssp. laricio) were sampled at two high-elevation sites in the adjacent Asco (Asc, 1600 m a.s.l.) and Tartagine valleys (Tar, 1450 m a.s.l.), located near the upper treeline in the north-western mountains of Corsica (Figure 1b). Long-lived trees from these valleys were previously used to develop multi-centennial tree-ring records [7][8][9]; however, no wood density measurements have been performed. To obtain the longest possible time series, mainly old and dominant trees as well as cross-dated relict stems were selected for this study.
Following standard radio-densitometric techniques [42][43][44], all cores were heated in 96% ethanol in a Soxhlet extraction system to remove the hydrophobic components. The purified samples were split into 3-cm segments to saw out 1.2-mm thick laths orthogonally to the longitudinal stem axis. The laths were acclimatized at 20 • C and 50% relative humidity for 4 h to ensure a homogeneous cell wall moisture content of~10%. Subsequently, the samples were radiographed for 14 min at 10 kV and 20 mA and the wood densities determined at 0.01 g/cm 3 resolution using a Walesch high-precision DENDRO2003 X-ray densitometer. The density peaks in each tree ring were used to produce inter-annual MXD data ( Figure 2c).
After aligning the MXD data by cambial age, a Hugershoff-shaped age trend was observed, including an average increase of 0.1 g/cm 3 over the first 70 years of growth and a gradual decrease of 0.13 g/cm 3 over the subsequent 330 years (71-400 years). To remove this characteristic biological trend from the raw MXD data [32,33], a Hugershoff Standardization (HGS) and a Regional Curve Standardization (RCS) were applied in the detrending program ARSTAN [45], resulting in two chronology variants: COR HGS and COR RCS .
After applying a data adaptive power-transformation to minimize heteroscedasticity within the individual series [46], all data were detrended by calculating residuals from the smoothed HGS-/RCS-mean curves and variance-stabilized using a 300-year spline with a fixed 50% cutoff to avoid deviations caused by changing sample replications and interseries correlations [47]. Robust bi-weight mean chronologies were computed [48] and the covariance of each chronology was assessed by calculating the inter-series correlation (Rbar) and the Expressed Population Signal (EPS) over semi-overlapping 30-year periods [49].
The new density data produced in our laboratory in Mainz (Germany) were compared with older measurements (n = 30 series), produced in the early 1980s in Birmensdorf (Switzerland) covering the period 1518-1980 CE [41] (data available from the International Tree-Ring Data Bank, www.ncdc.noaa.gov/data-access/paleoclimatology-data, accessed on 5 October 2020). The latter were derived from Black pines growing at the mountain pass Col de Sorba on Monte Renoso (Sor, 1400 m a.s.l.), which is located~45 km southeast of the new sites (Figure 1b). Given the significant correlation (r avg = 0.46) among the three site chronologies Asc, Tar, and Sor from 1636 to 1980 CE ( Figure 3, see Supplementary Table S1 for detailed site chronology characteristics), we produced a merged dataset integrating all Corsican MXD data (COR). The composite dataset contained 69 MXD series and exceeded a robust EPS of ≥0.85 back to 1425 CE ( Figure S1). When merging the data, the absolute MXD values were adjusted among the sites to mitigate significant (p < 0.001) elevational MXD effects between Tar and Sor (0.96 g/cm 3 ) and Asc (0.89 g/cm 3 ) calculated for the first 200 years of growth ( Figure S2).
Both detrending techniques applied in this study-HGS and RCS-are able to preserve multi-centennial variability in dimensionless index chronologies [50,51], although HGS is an individual detrending method and is restricted by the "segment length curse" [52]. Accurate RCS detrending, however, requires the mean cambial age of the series combined in a chronology to be constant over time [53,54]. To avoid the typical increase in mean tree age toward the present, age-band chronologies (ABC) [55] were produced by removing tree rings younger than 30 years and older than 200 years (ABC200), 300 years (ABC300), and 400 years (ABC400) of cambial age (see Supplementary Table S2 for chronology characteristics). Accordingly, five methodologically different chronologies were developed considering HGS and RCS with all data, as well as RCS with three age bands: COR HGS , COR RCS , ABC200, ABC300, and ABC400.

Climate Data and Signal Estimation
To assess the climate sensitivity of Corsican Black pine MXD and evaluate its suitability for reconstruction purposes, climate data from the meteorological stations in Ajaccio  [56], and GPCC v2020 precipitation data [57] were compiled using the KNMI Climate Explorer [58]. The selected CRU and GPCC grids extended across 42-42.5 • N and 8.5-9.5 • E, covering the entire study area.
The GPCC data were chosen for the analysis of precipitation signals due to the high station density in Europe and long temporal coverage back to 1891 CE [57]. The temperature means and precipitation totals were converted into monthly and seasonal anomalies with respect to   The local temperatures are, however, closely correlated with the coastal records (r = 0.86) and CRU temperatures (r = 0.84). In contrast, the annual precipitation totals from Calacuccia were less correlated with the coastal records (r = 0.53) and GPCC data (r = 0.67). We, therefore, used the coastal means and CRU temperatures as well as the GPCC precipitation data to assess the climate sensitivity of high-elevation MXD records.
All MXD chronologies were correlated (Pearson's r) with monthly and seasonal climate data over the 1951-1980 CE period to compare the climate signals contained in the old and new density measurements, as well as over the 1951-2016 CE period to account for recent climate variability. To evaluate the temporal robustness of the climate signals, 31-year running correlations were calculated between the density indices and different seasonal averages of meteorological data (April-July and September-October temperatures as well as June-July and August-September precipitation). Further, all MXD chronologies were correlated with 0.5 • gridded monthly and seasonal CRU TS4.04 data using the KNMI Climate Explorer to assess spatial correlation patterns over the Mediterranean Basin from 37-47 • N and −3-21 • E.

MXD Chronology Characteristics
The COR MXD compilation consists of 69 radii from 35 pine trees and spans the period 1168-2016 CE, including an increasing number of measurements toward the present ( Figure 4). Replication falls below five series at 1359 CE, while the maximum of 61 series is reached from 1872 to 1890 CE. COR replication decreases again in the late 20th century from 55 series in 1980 CE to 29 series in 1981 CE, when the Sor data terminate. Over the first 400 years of cambial age, average MXD (AMXD) is 0.91 g/cm 3 (±0.05 g/cm 3 ) with a minimum of 0.78 g/cm 3 in the fifth year of growth and a maximum of 1.00 g/cm 3 in the 107th year of growth. This juvenile AMXD increase is followed by a small but persistent decrease ( Figure S3).
The lengths of the individual series ranged from 67 to 849 years with a mean segment length (MSL) of 325 years, whereas the age classes are unevenly distributed across the sites. Trees sampled at the Asc site were significantly (p < 0.01) older (MSL Asc = 543 years) than those at the two lower sites (MSL Tar = 290 and MSL Sor = 211 years). Consequently, COR was characterized by a heterogenous age structure and site composition over time. All five COR chronologies shared annual to multi-decadal scale variability. Further, COR HGS showed a significant (p < 0.001) agreement in correlation with the site chronologies (r = 0.67-0.82) over the sufficiently replicated (n ≥ 5 series) 1636-1980 CE common period.
Compared to COR, the age-band chronologies-ABC200, ABC300, and ABC400-were characterized by flatter mean age and replication curves (Figure 4b-f). Due to the removal of tree rings older than 200, 300, and 400 years (and younger than 30 years) the age-band chronologies cover different periods. Whereas all chronologies extend back to 1365 CE with n ≥ 5 series, ABC200 already ends in 1980 CE, ABC300 in 1988 CE, and only ABC400 reaches into the 21st century until 2016 CE. Inter-chronology differences in replication and site contributions, thus, increase toward the present. (c-f) Samples (bar plots) and replication (black lines) of the chronologies. Each horizontal bar depicts one individual series. The percentages in brackets indicate the remaining data compared to COR (=100%). Grey-shaded areas mark the periods when replication in at least one record is n < 5 series.
The age-band chronologies showed substantial agreement in high-to-low frequency covariance (r avg = 0.91) and correlated significantly at r avg = 0.92 with the untruncated chronology COR RCS over the well-replicated (n ≥ 5 series) 1365-1980 CE common period. On multi-decadal scales, MXD indices of all chronologies increased from around 1400 to 1440, 1600 to 1660, 1700 to 1730 and 1760 to 1790 CE, and decreased from 1360 to 1390, 1440 to 1460, 1580 to 1600, 1730 to 1760, 1790 to 1820, 1860 to 1910 and 1950 to 1980 CE.
In the 20th century, however, the smoothed chronologies diverged considerably: COR RCS and ABC400 both showed a minor increase from 1975-1990 CE and a sharp decline thereafter; the smoothed ABC300 ends in 1990 CE and showed a striking post-1975 CE increase due to low sample replication; while ABC200 only extends until 1980 CE and showed no remarkably deviations from the other chronologies in the 20th century, but did in periods of n < 10 series, such as the late 17th century and before~1510 CE. Furthermore, large differences were observed between the HGS-and RCS-indices prior to~1425 CE, as evidenced by a downward shift in the RCS-values in the earliest density measurements. These differences were slightly higher between the COR HGS and the agebands than between the COR HGS and COR RCS .
The comparison of COR HGS with other long-term density records from southern Europe revealed a generally decreasing correlation with increasing geographic distance (Table 1, Figure 5a). COR HGS correlated significantly (p < 0.001) with the pine MXD records from the Spanish Central Pyrenees (GER, r = 0.53) [35] and southern Italy (CRI, r = 0.42) [60] from 1360-2004 CE, while weaker associations were found with more distant (>1000 km) records from southern Spain (CAZ, r = 0.33) [37] and north-eastern Greece (SMO, r = 0.28) [38]. Correlations with western Mediterranean sites were slightly higher than with equidistant sites in the eastern Mediterranean.  However, the weakest association was observed with the nearest record from the Swiss Alps (LOE, r = 0.23) [34], located only~450 km north of COR HGS , which is the only chronology based on Larix decidua instead of pine species. The smoothed density records from the Mediterranean outline both periods of high and low inter-regional correlations (Figure 5b). High covariance was primarily found in periods of pronounced variability, such as 1440-1500, 1580-1610, 1650-1670, 1810-1830, and 1850-1890 CE, whereby the zscore levels differed among the sites. Moreover, some chronologies indicated substantially deviating behavior, e.g., CAZ from 1760-1810 CE.
Further comparison with tree-ring based temperature reconstructions (Figure 5c) from Corsica (COR δ13C ) [8] and the Spanish Pyrenees (PYR MXD ) [36] demonstrated that the inter-annual MXD variability of P. nigra was more strongly associated with reconstructed May-September temperatures from the Iberian Peninsula (r 1582-2005 = 0.55) than with reconstructed August-September temperatures from Corsica (r 1582-2005 = 0.16), although the latter were derived from stable carbon isotopes of six Black pines sampled close to the COR sites [8]. In contrast, the reconstruction from the Pyrenees was based on a multiple-site Pinus uncinata MXD network [36].

Climate Sensitivity of Corsican MXD
MXD in Mediterranean conifers was primarily influenced by the warm-season temperatures (Table 1). In Corsica, correlations between the site chronologies and regional observations revealed local differences in temperature sensitivity as well as temporal changes in signal strength ( Figure 6). The site chronologies showed weak to moderate associations to the regional April-July and September-October mean temperatures (AMJJ and SO) over the 1951-1980 CE period, ranging from r = 0.17 at Tar to r = 0.47 at Sor. Although correlations with regional July temperatures were low overall (r = 0.03-0.15), analysis with the CRU temperatures indicated a significant (p < 0.01) association of r = 0.42 at Sor ( Figure S4). Thirty-one-year running correlations between the chronologies and seasonal AMJJ and SO temperatures confirmed these findings (Figure 6c): COR HGS indicated a temporally stable temperature signal until 1983 CE against station (r 1966-1983 = 0.34-0.47) and CRU data (r 1916-1983 = 0.42-0.67). These results were higher than the observations for Asc and Tar but slightly lower than Sor. However, the running correlations of all chronologies decreased sharply after 1983 CE, and this decline was stronger in Tar and weaker in Asc, whereas COR HGS remained in-between.
The five MXD composite chronologies all showed similar temperature sensitivities from 1901-1980 CE (Figure 7). Correlations with the April-July CRU temperatures (r = 0.43-0.53) were slightly higher than with the September-October means (r = 0.33-0.44), but the AMJJ and SO season fit best (r = 0.52-0.64). The strongest association with temperature was observed for the most truncated age-band chronology ABC200 (r = 0.64, p < 0.01). Spatial analyses using gridded CRU temperatures revealed strong correlations with the April-October means across the western and central Mediterranean regions (Figure 7a). The highest correlations (r ≥ 0.40) were found with the June-July temperatures over Corsica, Iberia, and Maghreb. In August, however, correlations over the study area declined (r = 0.21). As for the site chronologies, a systematic decrease in signal strength was observed in the composite records when the calibration period was extended into the 21st century (Figure 7b). All MXD chronologies reaching 2016 CE with n ≥ 5 series (COR HGS , COR RCS , ABC400) showed non-significant low correlations with the AMJJ and SO temperatures over the full 1901-2016 CE period. A simple scaling of COR HGS against post-1900 CE temperatures highlights a substantial divergence of 1.56 • C between the colder reconstructed (t = −0.26 ± 0.78 • C) and warmer measured temperatures (t = +1.30 ± 0.38 • C) from 2000 to 2016 CE, although the MXD and temperature records display similar interannual to decadal scale variability over most of the 20th century (Figure 7c).
The observed 21st century divergence is caused by opposing low-frequency trends starting in the late 1990s and is even greater when considering COR RCS and ABC400, respectively. The 31-year running correlations support these findings (Figure 7d), as all composite chronologies showed temporally stable temperature correlations ≥0.45 before 1967 CE, followed by r = 0.30-0.45 from 1968 to 1983 CE and a sharp decrease thereafter. Simultaneously, running correlations with June-July precipitation increased since the 1980s (Figure 7e).
Analyses using the GPCC precipitation data showed that the Corsican composite records were only weakly associated with precipitation ( Figure S5). Between 1891 and 1980 CE, all chronologies correlated significantly (p < 0.05) negatively with the June-September precipitation (r = −0.28 to -0.36). Correlations were lower and insignificant over the full 1981-2016 CE calibration period for COR HGS , COR RCS , and ABC400; however, the running correlations with the June-July precipitation showed a sharp post-1980 CE increase, while the correlations with the August-September precipitation remained stable.

Chronology Variants and Characteristics
The combination of MXD data from three different valleys enabled the development of a regional record (COR) characterized by a high sample replication (n = 69) and a robust common signal (EPS ≥ 0.85) back to 1425 CE. The five methodologically different composite chronologies share high fractions of high-to-low frequency variance from 1500 to 1980 CE. Noticeable trend differences were recorded before~1500 CE and in the late 20th and early 21st centuries (Figure 4). The age-band datasets more closely meet the requirements for RCS detrending [50,61], as these chronologies are characterized by more balanced replication and age curves.
Thus, our findings imply that age-band decomposition-as introduced by Briffa et al. [55] for a large network of Northern Hemisphere MXD data to improve the preservation of long-timescale variability and successfully applied in recent studies by Esper et al. [33,37]produces a sample structure that is likely more applicable for RCS detrending. However, the differences between ABC200, ABC300, and ABC400 remained small and restricted to periods of weaker sample replications (n < 10 series). This is particularly the case in ABC200, as this chronology was most affected by the systematic removal of old tree rings.
The substantial data reduction together with the marked discrepancies with all other chronologies during low-replication periods raises questions regarding the overall reliability of ABC200. ABC300 and ABC400, on the other hand, are characterized by less variable replication curves. Even after the removal of tree rings older than 400 years, ABC400 extends until 2016 CE with sufficient data replication. Yet removing data from the comparably small Corsica MXD dataset remains challenging.
Large differences between the HGS and RCS indices before~1425 CE and smaller deviations after 1900 CE raises doubts regarding the suitability of RCS for our dataset. While HGS, as a tree-by-tree standardization technique, is barely able to preserve multicentennial variability and produces a quasi-stationary chronology fluctuating around a mean index of one, the composite RCS technique retains more low frequency variance and produces constantly lower values at each chronology end (Figure 4).
Whether these low frequency trends are valid is difficult to estimate; however, we hypothesize that the regional curves established in the RCS runs [50] do not adequately represent the earliest and latest density measurements, particularly during the low-replication period prior to~1425 CE. This bias might originate from the selective sampling approach that focuses on old living trees and leads to younger trees being underrepresented in all chronologies [62]. However, to clarify these methodological limitations, a larger MXD dataset including different age classes is required.

Bimodal Temperature Signal
Various studies have reported warm-season temperatures to be the main climatic driver of MXD formation in high-elevation Mediterranean conifers [30,[35][36][37][38][39][40]. Regarding the calibration with pre-1980 CE instrumental data, our results are consistent with these findings revealing significant temperature correlations throughout the growing season from April to October including a lack of forcing in August. In contrast, the influence of precipitation on MXD appears to be minor. The resulting bimodal seasonality of temperature signals is comparable to the findings reported by Büntgen et al. [35] and Esper et al. [37] from Spanish sites.
Whereas bimodality at the Cazorla Natural Park in southern Spain was characterized by a weakened response during mid-summer months from June-August [37], it was less pronounced in the Pyrenees [35], as well as in Corsica, where only one month (July and August respectively) was insignificant. Such bimodal response patterns have only been observed at western Mediterranean sites thus far; eastern Mediterranean sites, such as SMO [38,39], as well as MXD chronologies from central [34,63] and northern European regions [32,33], have not been found to exhibit bimodality.
Given the temperature insensitive MXD formation in Morocco [64] and the bimodal temperature signals at the two Spanish sites [35,37], our findings suggest a tendency toward increased seasonal bimodality with decreasing latitude and/or greater aridity throughout the Western Mediterranean. However, additional studies need to be conducted across the region to identify the mechanisms and climatic thresholds responsible for bimodal temperature signals.
The physiological reasons for this lack of response during peak warmth are not yet fully understood but are potentially related to a temperature-decoupled carbohydrate production [37] and/or an effect of summer drought limiting xylogenesis [65,66]. The latter assumption is in line with previous work by Trouet [67], who found the MXD of P. nigra to be more sensitive to August precipitation than to temperature at several sites in southern Italy and Greece, whereas Esper et al. [64] reported the MXD formation in xeric regions of Morocco to be primary controlled by drought variability.
Since the high mountain regions of Corsica can also be subject to drought from June-August [13], our results imply that high evapotranspiration rates limit the July-August soil water contents, thus, temporally altering the sensitivity of latewood formation toward precipitation. This hypothesis is supported by dendrometer data obtained along an elevational transect of pines in Corsica revealing fast changes in stem circumference after precipitation events in summer and rainfall-induced cambial reactivation in autumn [68]. The latter might also explain the negative correlations of MXD with September precipitation ( Figure S5).
The wood anatomical traits of P. nigra latewood, such as the tracheid diameter, lumen width, and cell wall thickness, are strongly determined by the trees' water status, as their expression results from trade-offs between the hydraulic requirements of the xylem and the avoidance of embolism [69][70][71]. Protecting the trees from cavitation, smaller tracheids with thicker cell walls are formed in summer, while larger tracheids with thinner cell walls are needed to improve hydraulic efficiency [9,72]. Since the trees vascular system can be severely damaged by xylem embolism, cavitation protection appears to be primarily important. Thus, latewood cells tend to be denser in years with higher monthly temperatures and adequate water supply [69,73].
Temperatures at the beginning and end of the growing season appear to have a strong influence on latewood formation, as sufficient precipitation falls in spring and autumn. The positive relationship between April-May temperatures and MXD is likely related to a combination of early snowmelt, reinforced photosynthetic activity, and the production of non-structural carbohydrates prior to the onset of cambial activity [7,35]. In contrast, correlations in September-October demonstrated a direct influence of temperature on secondary cell wall deposition and lignification [73,74]. Although latewood cell formation generally proceeds from July to September, wall thickening and lignification of terminal tracheids can continue into October [7,75,76], leaving discernible imprints in MXD.
Moderate inter-site correlations ( Figure 3) and varying temperature correlations at Asc, Tar, and Sor ( Figure 6) suggest that the temperature sensitivity of MXD was co-influenced by local site conditions. Since the duration and intensity of summer drought generally decreases with altitude, trees at lower sites generally contain stronger hydroclimatic signals [68,77,78]. This tendency might also impact the reduced temperature correlations at Tar (1450 m a.s.l.), as this site is located approx. 150 m beneath Asc in the adjacent valley.
The MXD chronology from Sor at only 1400 m a.s.l., however, indicates the highest temperature sensitivity, and thus elevational differences appear less influential overall. Previous studies pointed to the combined biogeographical effects of local topography, slope exposure, soil formation, and water availability when discussing signals in Corsican TRW networks [7], tree ring stable isotopes [8,79], and intra-annual density fluctuations [69]. This might also be applicable for MXD; however, further measurements and a network across the island are necessary to disentangle the impact of local site conditions on P. nigra MXD formation.

Fading Temperature Sensitivity
In environments where temperature is the limiting factor of tree growth, MXD has proven ideal for reconstructing past summer temperature variability [33,34,53]. However, hemispheric studies have reported a decreasing temperature sensitivity since the second half of the 20th century [80,81]. This phenomenon, known as the "divergence problem", describes the inability of temperature-sensitive tree-ring chronologies to adequately capture 20th century warming. The lack of mirroring recent warming questions the reliability of tree-ring-derived temperature reconstructions during putative pre-instrumental warm periods.
Despite widespread evidence in a network of MXD data obtained from multiple sites across the Northern Hemisphere [80], divergence appears to be more prevalent at northern latitudes [82]. As most records were developed in the 1980s and 1990s, respectively, and only a few extend until the second decade of the 21st century [80,83], there is a need to update MXD records toward the present for further in-depth-investigations of the divergence problem.
Our Corsican MXD chronologies reveal a similar decoupling from temperature in the late 20th century, resulting in an underestimate of ≥1.56 • C between the warmer instrumental and colder reconstructed temperatures in the 21st century. In contrast to the results by Briffa et al. [80,81], this decoupling starts in the late 1990s and not in the 1960s. We suggest that the increasing temperatures throughout recent decades have resulted in more severe summer drought periods with higher evapotranspiration rates at our study sites.
Instrumental observations support this assumption and reveal that warm-season temperatures have continuously increased over the past 100 years, particularly after 1980 CE (Figure 7c). While the AMJJ and SO temperatures from 1980-2000 CE are comparable to those of the 1940s or 1950s, warming amplified rapidly thereafter surpassing nearly all 19th century observations. Since this temperature increase is not accompanied by increasing rainfall, the offset between precipitation and potential evapotranspiration is widening toward the present [84].
Drought episodes have begun to occur more frequently and persist longer, particularly since the 1990s when hydroclimate extreme events were most severe in the Western and Central Mediterranean [85,86]. In Corsica, the frequency of dry years since intensified rapidly from 20% to 50%, while heavy rainfall events occurred more frequently [84] further promoting summer drought, as water infiltration is impeded by the rocky slopes and large portions of precipitation are lost to surface runoff [68].
We therefore conclude that the P. nigra trees growing at the upper treeline have suffered from enhanced drought stress since the 1990s, causing the decrease in MXD sensitivity to warm-season temperatures toward the end of the 20th century. At the same time, running correlations revealed that the June-July precipitation became increasingly important for MXD formation, although the precipitation signals were weak overall. The 1990s are identified as the turning point in MXD climate sensitivity. It appears that the succession of several severe drought events caused a shortage of carbohydrate reserves due to inhibited photosynthetic activity during summer [70,87].
Although Black pines feature an effective stomatal control mechanism [88] and indicate a plastic response to dryness with a rapid recovery thereafter [15], trees at Tar have suffered particularly from prolonged drought stress since the 1990s. This could explain the anticorrelations with temperature over the most recent calibration period. The combined effects of high temperatures and low water availability stimulate stomatal closure, limit the absorption of carbon dioxide, and further a shortage of non-structural carbohydrates required for cell wall formation [87,89,90].
This conclusion is consistent with previous dendroclimatological studies detailing drought as an important driver of Mediterranean tree growth and suggesting that the increasing frequency and duration of summer aridity has reduced the temperature-growth relationship at southern European sites [91][92][93][94][95]. The Corsican pine forests are also affected by the regular wildfires characteristic for Mediterranean forests [96,97]. Wildfires have been recognized as disturbance events in dendroclimatic studies as they can leave distinct fingerprints in the tree-ring series of surviving pine trees [98][99][100].
However, whereas the number of recorded fires increased throughout the 20th century in many Mediterranean regions [101], there was no indication of a change in Corsican fire regime coinciding with the altered climate sensitivity in high elevation MXD data as reported here. The striking coherence among TRW chronologies from multiple Corsican pine sites, extending well into the 21st century [7,9] supports this conclusion that local fire events are of minor importance to the climate signals recorded in tree-ring growth proxies.
Despite possible uncertainties in climate signal estimation emerging from the application of gridded and instrumental station data from the coastal plain, the Corsican composite chronologies contain crucial temperature information, as COR HGS indicated high coherence in low-frequency variability with the world's best replicated (n = 414) MXD record from the Spanish Pyrenees (GER). The GER chronology correlated significantly (r 1950-2014 = 0.72, p < 0.01) with the May-June and August-September temperatures over large parts of the Western Mediterranean [35].
Similar spatial correlations were recorded between the ABC200 and AMJJ and SO temperatures prior to 1980 CE. Surprisingly, despite the greater geographical distance, COR HGS was more closely related to a May-September temperature reconstruction based on MXD from Pyrenean mountain pines [36] than to the August-September temperature reconstruction derived from stable carbon isotopes of Corsican Black pines [8]. This discrepancy could result from both the different seasonalities and/or the climate proxies, as the carbon isotopes are more closely influenced by the mutual interactions of temperature, precipitation, and transpiration and carry a mixed climate signal [8].
Based on these findings, we hypothesize that inter-annual MXD variability derived from Corsican Black pines can reliably represent large-scale temperature patterns across the Central-Western Mediterranean over more than 500 years. However, due to the decreasing sensitivity since the late 20th century, a formal reconstruction of past temperatures is not warranted. The development of more and better-replicated MXD chronologies are required to determine how the increasing drought sensitivity affects P. nigra growth and to improve our understanding of past climate dynamics in Corsica.

Conclusions
By combining 69 MXD measurements from three high-elevation sites in Corsica, we updated previously established density data by 36 years into the 21st century and by 350 years back to the Middle Ages. After applying two detrending techniques that were able to preserve multi-centennial variability, as well as age-band decomposition, all chronologies of the now longest MXD record for the Central-Western Mediterranean indicated similar associations with climate data. The strongest correlations were observed with April-July and September-October temperatures over calibration periods prior to 1980 CE. August temperatures were found to play a minor role in MXD formation, accentuating a bimodal seasonality in temperature response that is similar to observations at Spanish sites in the Central Pyrenees and Andalusia. The bimodality in temperature sensitivity is likely related to (1) an unimpeded carbohydrate production during mid-summer, when temperatures do not fall below the physiological thresholds; while, at the same time, (2) distinct drought conditions cause stomatal closure inhibiting photosynthetic activity. Comparisons with other temperature-sensitive MXD chronologies from the Mediterranean showed that the Corsican density record contains large-scale climate information over more than 500 years.
However, the weakening temperature associations since the 1990s caused by differing low-frequency trends between wood density and instrumental temperature data raise doubts regarding the reliability of P. nigra MXD for climate reconstruction. We demonstrated that the post-1990 CE divergence was likely due to a shift from thermal to hydroclimate forcing. In context of recent warming, our results underline the significance of updating tree-ring density data into the present to evaluate whether MXD-based temperature reconstructions derived from 20th century calibration periods remain valid into the 21st century. For a better understanding of underlying physiological mechanisms and the influence of drought on MXD formation, additional studies of wood anatomical traits would be helpful.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/10 .3390/atmos12070804/s1, Table S1: MXD site chronology characteristics; Table S2: MXD composite chronology characteristics; Figure S1: Effects of merging site measurements, Figure S2: Effects of mean adjustment; Figure S3: Sample replication and regional curves of the composite MXD data; Figure S4: Temperature signal estimation using CRU TS4.04 data; Figure S5: Precipitation signal estimation using GPCC v2020 data.  Data Availability Statement: Data will be made available at the International Tree-Ring Data Bank (https://www.ncdc.noaa.gov/data-access/paleoclimatology-data/datasets/tree-ring) at acceptance.