Recent multispecies tree-growth decline reveals a severe aridity change in Mediterranean Chile

Soil moisture (SM) is a crucial factor in the water cycle, sustaining ecosystems and influencing local climate patterns by regulating the energy balance between the soil and atmosphere. Due to the absence of long-term, in-situ measurements of SM, studies utilizing satellite-based data and tree-ring analysis have become valuable in assessing variations of SM at regional and multi-century scales, as well as determining its effects on tree growth. This information is particularly pertinent in biodiversity hotspots made up of semi-arid ecosystems currently threatened by climate change. In the Mediterranean Chile region (MC; 30°–37° S), an ongoing megadrought since 2010 has resulted in a significant decline in the forest throughout the area. However, the impact of SM on tree growth at a multi-species and regional level remains unexplored. We analyzed a new network of 22 tree-ring width chronologies across the MC to evaluate the main spatiotemporal tree-growth patterns of nine woody species and their correlation with SM, using PCA. We also reconstructed the SM variations over the past four centuries and assessed its connection with large-scale climate forcings. Our results indicate that the primary growth patterns (PC1) explained 27% of the total variance and displayed a significant relationship with SM between 1982–2015 (r = 0.91), accurately reflecting the current megadrought. The tree-ring SM reconstruction covers the period 1616–2018 and shows a strong decrease around the year 2007, revealing an unprecedented recent change in aridity with respect to the last four centuries. The intensity of the South Pacific subtropical anticyclone, which primarily owe their existence to the subsiding branch of the Hadley Cell, appears as the primary climatic mechanism correlated with the reconstruction and the present aridity conditions in MC. The current SM conditions align with anticipated aridity changes in MC, providing a bleak perspective of future regional climate.


Introduction
Many sub-tropical regions are currently facing significant environmental changes influenced by the increased frequency, duration, and intensity of droughts (Allen et al 2015, McDowell et al 2020, Cook et al 2022).These changes affect the availability of water resources and the dynamics of forest ecosystems, with implications for carbon and hydrological dynamics (Anderegg et al 2015, Kannenberg et al 2019, Forzieri et al 2022).Located in the subtropical western coast of South America, the Mediterranean region of Central Chile (MC; 30 • -37 • S) is considered a global biodiversity hotspot, and one of the biologically richest and most endangered terrestrial ecosystems on Earth (Mittermeier et al 2011, Fuentes-Castillo 2020).The MC biome contributes to the balance of carbon storage and water regulation in a highly anthropic region that concentrates over 70% of the country's population and is the primary contributor to national agricultural production (Cerda et al 2014, Roco et al 2015, Montoya-Tangarife et al 2017).MC has a high biodiversity and endemism level of flora and fauna (Myers et al 2000), but several studies have reported a significant increase in risk factors threatening it, including widespread land-use change (Schulz et al 2010, Uribe et al 2020), humancaused fires (Urrutia-Jalabert et al 2018, Bowman et al 2020), and the current megadrought (Garreaud et al 2017(Garreaud et al , 2020)), partly attributable to anthropogenic forcings (Boisier et al 2018).This unprecedented event coincides with an increase in air temperature (Burger et al 2018) and a decrease in precipitation in recent decades (Rivera et al 2017, Garreaud et al 2020), which has caused a massive regional decline in sclerophyll forest ecosystems (Venegas-González et al 2019, 2023, Miranda et al 2020).Climate models suggest that aridity conditions will intensify in this region during the 21st century, negatively impacting tree growth (Matskovsky et al 2021) and species distribution (Fuentes-Castillo et al 2019).This will further enhance the vulnerability of biodiversity, ecosystem services, and regional economies in the MC region (Schwabe et al 2013).
Assessing the variability of soil surface water balance is crucial for understanding the environmental impacts of droughts and for assessing possible ongoing climatic changes in densely populated and endangered ecosystems of subtropical semi-arid regions.Tree rings in Mediterranean regions are recognized for their ability to document hydroclimatic conditions at interannual, decadal, and multi-century scales (Cherubini et al 2003, Villalba et al 2011).This enables the assessment of moisture variability's influence on tree growth and the reconstruction of past environmental conditions.Additionally, these environmental archives have been shown to be extremely useful for identifying patterns and changes in past hydroclimatic variability at different spatial and temporal scales in subtropical regions of both hemispheres (LeQuesne et al 2006, Christie et al 2011, Palmer et al 2015, Wilson et al 2016, Garreaud et al 2017, Cook et al 2020, Morales et al 2020, Williams et al 2020).Although tree-ring studies have assessed the connection between climate and tree growth (Venegas-González et al 2023), as well as precipitation and drought reconstructions in MC (LeQuesne et al 2006, Morales et al 2020), the role of soil moisture (SM) changes in relation to tree growth across multiple species and regional scale has yet to be investigated.Retrospective studies of MC hydroclimate and tree growth relations have primarily focused on the impact of low precipitation (LeQuesne et al 2006), rather than thoroughly assessing the effects of hydrological drought at surface and subsurface levels (Mishra and Singh 2010).Satellite-based SM data has proven a valuable tool in showcasing the significance of this parameter for tree growth and forest ecosystem dynamics (Muñoz et al 2014, Kostić et al 2021, Thomte et al 2022).
In this study, we establish a new tree-ring network spanning 22 sites across the MC region, encompassing 9 different species, including 3 new for treering studies.The objectives of the present study are: (i) to evaluate the relationships between tree growth and SM both at local and regional scale across the network, (ii) to determine the common spatio-temporal tree growth patterns from the network and its relation with SM, and (iii) to utilize these registries to develop a SM reconstruction to evaluate the severity of the present megadrought within the context of the last 400 years.A deeper understanding of the relationship between multispecies tree growth and SM across the MC biome will enhance our knowledge of how this highly diverse and threatened ecosystem responds and adapts to ongoing and future climate changes in the MC region.

Study area
The MC region encompasses a strip of land between the Pacific coast and the central Andes of Chile and Argentina (30 • -37 • S), situated at the transition between the Atacama Desert in the north and the humid-temperate Andes in the south (figure 1).This area spans a latitudinal precipitation gradient from 100 to over 700 mm per year.The Mediterranean climate in the region exhibits a substantial oceanic effect that gradually diminishes from the coast towards the Andes, reaching up to ∼7000 m a.s.l.only 200 km from the Pacific Ocean (Sarricolea et al 2017).Year-to-year precipitation variability is linked to the number of frontal systems arriving from the Pacific Ocean, highly influenced by variations in El Niño-Southern Oscillation (ENSO).ENSO, in turn, modulates the dynamics of the subtropical anticyclone (SPA) and the atmospheric pressure in the Amundsen-Bellingshausen sector of the South Pacific (Aceituno 1988, Montecinos andAceituno 2003).During El Niño (La Niña) events, aboveaverage (below-average) precipitation is expected in the MC during winter (Montecinos and Aceituno 2003).Several studies have linked high-latitude climatic forcing, such as the Southern Annular Mode (SAM, also known as the Antarctic Oscillation or AAO), to seasonal precipitation variability and its long-term reduction trend in MC (Garreaud et al 2009, Quintana and Aceituno 2012, González-Reyes 2016, Boisier et al 2018).Currently, there has been a decrease in precipitation since the late 1970s as a result of climate changes caused by both natural and anthropogenic forcings (Boisier et al 2016, 2018, Villamayor et al 2022).
Several woody species were selected across the MC region to cover different environments within this biome, including the Andean mountains, the central valley, the coastal range, and the coastal lowlands.This encompasses sclerophyllous trees and shrubs, and deciduous forests (Gajardo 1994).We sampled 22 tree-ring sites from 9 species (figures 1, S1 and table S1).Three of these species have not been previously used for dendrochronological studies (Escallonia revoluta, Berberis montana, and Nothofagus glauca).Some of the sampling sites included transitional Andean zones (1400-2696 m.a.s.l.) with Proustia cuneifolia (Huañil) and Berberis montana (Michay) as arboreal shrubs, Kageneckia angustifolia (Frangel) as small trees, and the conifer Austrocedrus chilensis (Ciprés de la Cordillera).Tree-ring samples were collected from 2015 and 2021.Two to three cores were extracted at breast height from trees using an increment borer, and cross-sections were obtained from multi-stemmed shrubs exhibiting eccentric growth patterns.Samples were air-dried in the laboratory and smooth-polished with sandpapers of decreasing grain size.The growth rings were dated following Schulman's convention for the Southern Hemisphere, which assigns to each ring the year when radial growth begins (Schulman 1956).Tree-ring widths were measured with a Velmex measuring system accurate to 0.001 mm.The accuracy of dating of the tree-ring series was verified using the COFECHA software (Holmes 1983).The dplR software package (Bunn 2008) was used to standardize the individual series of each site by fitting either a negative exponential curve or a straight line (Cook and Kairiukstis 2013).This technique effectively eliminates fluctuations caused by biological growth trends or other low-frequency variations not related to climate.The quality of each chronology was assessed using objective measures, including mean sensitivity (MS) to determine year-to-year variability within a ring-width series, first-order autocorrelation (AR1), mean series intercorrelation (RBAR), and the expressed population signal (EPS) which quantifies the degree to which the series portrays a hypothetically perfect chronology (Wigley et al 1984).Only periods in the chronologies with EPS > 0.85 were considered.

SM data
To evaluate the relationships between tree growth and SM, we used the data from by the Famine Early Warning Systems Network Land Data Assimilation System (FLDAS;McNally et al 2017).This global land surface model is driven by various datasets from satellite measurements and atmospheric analyses, which are evaluated using direct comparisons of outputs with independent measurements (ground/satellite) and indirect comparisons, including correlations of SM with vegetation indices (McNally et al 2017).The FLDAS data provides monthly gridded SM data at various depths, with a horizontal resolution of approximately 10 kilometers since the year 1982.The data is expressed as the volume of water per soil volume (m 3 m −3 ).For our research, we focused on the SM layer between 0-10 centimeters depth.

Tree growth and SM relationships
Two methods were used to establish the connection between SM and tree-ring chronologies.First, we examined the connections at the individual chronology level, considering the SM grid at the chronology location.Additionally, we assessed these connections at a regional scale using the mean SM for the entire MC region (30 • -37 • S, see figure 4(a)).The evaluations were conducted at the monthly scale over the common period 1982-2015, covered by the ring width chronologies and the FLDAS.To identify the optimal correlation between SM and tree growth, we evaluated all possible monthly and seasonal combinations using Pearson correlation coefficients by running the KLIMA_trees routine (available at https:// github.com/lonkotrewa/KLIMA_trees) in R project software (R; R Core Team 2020).Additionally, Monte Carlo bootstrap of 2000 simulations were performed to obtain robust Pearson correlations.We analyze the spatiotemporal patterns of radial growth for the nine species using the entire network of tree ring chronologies across the MC region (22 sites), using empirical orthogonal function (EOF) analyses (Wilks 2011).The EOF analysis accurately condenses the dominant growth patterns recorded in the 22 chronologies into a few series, referred as principal components (PCs).
Our study primarily examines the first and second PCs (PC1 and PC2) and their associated spatial expression to characterize the dominant growth patterns.Finally, we determine the relationships between SM and the primary spatiotemporal pattern of MC tree growth by correlating the PC1 amplitudes with the SM mean from the entire MC region.

SM reconstruction
The hydrological year (May-April) SM reconstruction was generated by utilizing a principal component regression (PCR) method.This involved regressing the hydrological year SM (May-April) for the MC region against standard tree-ring chronologies, following the approach detailed by Cook et al (2007).Candidates predictors of SM were selected from tree-ring chronologies that began before 1950 and exhibited a significant correlation (p-value < 0.01) with hydrological year SM.These selected chronologies (see table S1) were then entered into a PCR to reduce the number of predictors and enhance the commonality present in the tree-ring records.Given the relative short SM record for calibration, the reconstruction model was developed using the 'leave-oneout' cross-validation procedure (Michaelsen 1987).In this approach each observation is successively withheld; a model is estimated on the remaining observations, and a prediction is made for the omitted observation.At the end of this procedure, the time series of predicted values assembled from the deleted observations is compared with the observed predictors to compute the validation statistics of model accuracy and error.The goodness of fit between observed and predicted SM values was tested based on the proportion of variance explained by the regression (R 2 -adj), the significance of the linear trend of the regression residuals evaluated by the Mann-Kendall test (Mann 1945, Kendall 1975), the reduction of error (RE) statistic over the verification period (Cook et al 1999), as well as the root-mean-square error (RMSE) statistic as a measure of inherent uncertainties in the reconstruction (Weisberg 1985).
To maximize the length of our tree-ring chronologies we utilized a nested PCR approach.This involved dropping shorter chronologies and repeating the PCR procedure using the remaining longer series (Meko 1997, Cook et al 2004).Calibration was performed on each nested PCR for the period 1982-2015.The final reconstruction was created by splicing the relevant segments from each full-period calibrated nested model.The mean and variance of each nested reconstruction were adjusted to those of the most replicated nest with the highest R 2 to prevent artificial changes in variance during the reconstruction caused by purely statistical, non-climatic factors (Cook et al 2004, Wilson et al 2007).To minimize potential underestimation of extreme hydrological year regional SM values during the reconstruction, we restored the lost variance by rescaling the final reconstruction to match the variance of the observed SM data during the calibration period (Cook et al 2004).After that, we updated the final nested reconstruction, covering the period 1616-2018, with FLDAS regional data until 2021.

Climate forcings and MC SM variability
To determine the main relationships between MC SM variability and large-scale climate, we performed multiple regression analyses using several climate indices as potential predictors of the MC SM record.These climate indices (see table 1) have been widely used to characterize climatic conditions in South Western South America and were thus possibly linked to interannual SM variability across MC.As some of these candidate predictors may not be statistically independent, a stepwise regression approach (F-toenter 0.05, F-to-remove 0.10) was used to avoid multicollinearity among the predictors and develop a reliable regression model.Since most of the seasonal climatic index series exhibit strong serial correlations, we utilized the prewhitened versions of the seasonal climatic indices as predictors of the prewhitened SM record.Additionally, we estimated the spatial correlation pattern between the reconstructed hydrological year SM and the mean annual sea level pressure (SLP) (May-April; 0.25 × 0.25 gridded cell) from the ERA5 reanalysis (Hersbach et al 2020).

Tree growth and SM relationships
Twenty-two tree-ring sites were analyzed to assess the link between tree growth and SM variability at both local and regional scales, utilizing FLDAS SM data spanning the period 1982-2015.The results reveal a generally strong and positive relationship between tree growth and May-April SM at both local (tree-ring site vs. respective SM grid) and regional (tree-ring site vs. entire MC region) scales, as illustrated in figure 2. Significant positive correlations were recorded at both local and regional scales for most sites, with stronger relationships identified at the local scale (figure 2).The highest correlation coefficients were observed for K. angustifolia (PYL) at both local and regional scales, C. alba (PLC) and N. macrocarpa (ACH) at the local scale, and B. miersii (BLC) and K. angustifolia (MET) at the regional scale.Several C. alba sites displayed significant correlations with both in-situ and regional May-April mean SM values in the MC.Similarly, a majority of B. miersii chronologies exhibited strong relationships with SM values (r > 0.55).Furthermore, the three species E. revoluta (SOB), N. glauca (HPL), B. montana (PAR), utilized for the first time in treering studies, demonstrated significant correlations with May-April mean SM at local and regional scales, demonstrating their utility for dendrochronological studies.

Spatiotemporal patterns of tree growth
Utilizing 22 ring-width chronologies from nine species of shrubs and trees distributed throughout the MC region, the first two PCs explain 37% of the total variance in tree growth over the period 1970-2015 (figure 3).The first principal component (PC1) accounts for 27% of the total variability, indicating a significant contribution from the entire tree ring network located between 32 • and 34 • S (figure 3(a)).Several species chronologies, including those from B. miersii, C. alba, and K. angustifolia exhibit high PC1 loadings (figure 3(a)).PC1 amplitudes above the average prevail during the intervals of 1977-1984, 1990-1992, and 2000-2006.In contrast, years below the mean are recorded during the intervals of 1988-1990, 1994-1999, and the extended period of low growth between 2007-2015, as illustrated in figure 3(b).The PC2 accounts for 10% of the total variance, with a significant contribution from the P. cuneifolia and N. macrocarpa chronologies situated at the northern boundary of the MC region (∼30 • S; figure 3(c)).Temporal variations in PC2 reveal above-average tree growth intervals primarily from 1971 to 1976 and 1987 to 1991, while belowaverage periods were identified mainly from 1977 to 1979, 1983 to 1988 and 1997 to 2003 (figure 3(d)).Additionally, a remarkable interannual growth variability has been observed since 2004.

Regional pattern of tree growth and SM relationships
We examined the correlation between temporal changes in PC1 and the gridded mean May-April SM in the MC region from 1982 to 2015 using Pearson correlation coefficients.The May to April intervals were selected due to the high significant correlations obtained during these months (figure S2), and also not observed in another climatic variables such as May-April precipitation and mean air temperature during the same months (figure S3).We found strong   spatial correlations between PC1 and the annual variability of SM, particularly in the 33 • -36 • S region, where Pearson correlation coefficients were notably high and significant (r > 0.8; figure 4(a)).In temporal terms, both time series exhibit synchronicity during several years, displaying consistently low values of yearly regional SM starting from 2007 (figure 4(b)).Furthermore, a robust correlation (r = 0.91) was recorded between SM and the first PC of the radial growth patterns for the nine different species in the MC from 1982 to 2015.We also identified that correlations between the PC1 and May-April SM are highest compared to ERA5 Precipitation and mean air temperature during May-April (figure S3).

Reconstruction of SM in mediterranean Chile
The robust correlation between PC1 and SM content in the MC region during May to April months (r = 0.91, 1982-2015; figure 4) represents a novel finding not previously reported in tree-ring studies conducted in South America.Building upon this significant result, we conducted a nested May to April SM reconstruction spanning the last four centuries (1616-2018) to characterize the temporal variability and changes in SM in the MC region.To include the years 2019 to 2021, we included actual FLDAS measurements into the reconstructed series.Our May-April regional SM reconstruction is based on seventeen chronologies from various tree and shrub species.These species exhibit a significant correlation (p-value < 0.01) with May-April SM content at both local and regional scales within the MC region (figure 2).We only selected tree and shrub chronologies with an EPS greater than 0.85 since 1950.Our new findings demonstrate the robustness of our reconstruction and its significant correlation with observed May-April SM variations from 1982 to 2015 (r = 0.91; figure 5(a)).Additionally, the RMSE values were low, and there is not any significant trend in the residuals of the regression models (table S2).The final nested reconstruction effectively reproduced total variance values between 0.43 and 0.81 (figure 5(c)).For a complete summary of the reconstructions, as well as their corresponding statistics and intervals needed to generate the final nested reconstruction, refer to table S2.Since several chronologies contain information beyond 2015 (table S1), we perform reconstructions up to 2018 using the PC reconstruction.The SM reconstruction displays extended periods of low and high May-April SM values from 1616 AD to the present day.For example, there were periods of exceptionally high values found during 1650-1675, 1725-1750, 1820-1860, 1890-1910, and 1960-1990 (figure 5(b)).Conversely, intervals of low SM values were observed during 1680, 1751-1770, 1790-1820, 1911-1950, and since 2007 onwards.

Climate forcings and MC SM variability
The multivariate stepwise regression trials revealed that, except the latitudinal position of the Southern Pacific anticyclone (SPAL), all climatic indices for December to February (austral summer) exhibit consistent and significant relationships with the observed regional May-April SM derived from FLDAS product from 1983 to 2021 (table 1).During autumn (March to May; MAM), significant relationships present only significant values with the intensity of the Southern Pacific anticyclone (SPAI).Significant correlations were also detected during winter (June to August; JJA).Significant relationships were found with Niño 3.4, SOI, and the TPI, with the strongest correlation observed with the SPAI.In winter, utilizing the full range of climate indices, the model was able to replicate 33% of the overall variance.On an annual scale, the stepwise regression model was able to replicate 31% of the total variability, where the SOI and SPAI was the most relevants climate  S1.
Table 1.Multivariate regression trials between 1983-2021 prewhitened versions of the observed May to April soil moisture average for MC and seasonally averaged estimates for the six climatic indices.Four 3-month seasons starting in May of the previous year were used to current April, and candidate predictors were selected following a stepwise regression approach (F-to-enter 0.05, F-to-remove 0.10).The r denotes the correlation coefficient between the observed May to April soil moisture average for MC and seasonal averages for each climatic index; Beta denotes the standardized regression coefficients used to compare the relative strength of the various predictors within each model; Adj R 2 denotes the coefficient of determination adjusted for the number of predictors in the model, and # indicates that none of the predictors passed the threshold value for F-to-enter.The climate indices acronyms mean N3.4: SST anomalies for the Niño3.4region (Trenberth 1997), SOI: Southern Oscillation Index (Ropelewski and Jones 1987), TPI-IPO: Interdecadal Pacific Oscillation (Henley et al 2015), SAM: Southern Annular Mode (Marshall 2003), SPAI: Intensity of the South Pacific Subtropical Anticyclone, and SPAL: latitudinal position of the South Pacific Subtropical Anticyclone.Both indexes based on Barrett and Hameed (2017)

Tree-growth and SM relationships
Interannual variations in radial growth of nine tree and shrub species at 22 sites in the MC (30 • -37 • S) region are closely related to variations in May to April SM (0-10 cm depth), derived from FLDAS observations over the period 1982-2015.In addition, a previous study developed by Thomte et al (2022), has used this product to assess the relationship between tree growth and SM levels for Pinus kesiya in the southern Himalayas, indicating the utility of this product in tree-ring analyses.We used three new species for dendrochronological studies (B.montana, E. revoluta, and N. glauca), thus increasing the number of species that compose the current South American network of dendrochronological records (Boninsegna et al 2009, Morales et al 2020).Our correlations results show that the relationships between tree and shrub chronologies and May to April SM are exceptionally strong, demonstrating the high dependence between vegetation growth and SM along the MC region.These positive relationships are remarkable at both local (chronology site) and regional scales using mean May to April SM across the MC.Our results show that the nine study species are sensitive to SM variations across the MC region.C. Alba (PLC and PML sites; figure 2) show the highest correlation values among the other species used in this study.
Our reconstruction demonstrates the exceptional nature of the present MC megadrought within the context of the last 400 years.Recent studies have reported the severe growth decline in several plant species in the MC as a result of the current unprecedented megadrought during the last decade (Garreaud et al 2017, Miranda et al 2020, 2023, Venegas-González et al 2018, 2023).These plant species and probably many others from the MC region, have confronted extreme climate and environmental conditions due to the persistent severe aridity conditions after 2010 (Garreaud et (Miranda et al 2023).This biome provides a variety of fundamental ecosystem services to the society (Gajardo-Rojas et al 2022, Smith-Ramírez et al 2023), and have been proposed as a key element to implement nature based solutions for water security for the region (Schneider-Valenzuela et al 2023).These pervasive drought conditions in MC are well reflected in the SM changes during the last decade, representing the most severe hydroclimatic change in Mediterranean ecosystems worldwide, with profound effects on forest ecosystems (Miranda et al 2023).Conservation efforts are crucial to protect and preserve the remaining MC forests to maintain their resilience capacity and provide mitigation measures at the local level during future climate.
In spatial terms, we record that tree-ring sites located between 32 • -34 • S present the highest loadings factors values within the PCA analysis, being particularly relevant the decline of growth patterns from around the year 2007 onwards exhibited by the first dominant pattern of radial growth in both trees and shrubs.Several sampling sites of B. miersii and C. alba are located at these latitudes in Central Chile, where we registered the strongest relationship with MC SM variability and changes.The first and second temporal growth patterns exhibit contrasting patterns.We also identified a decline in growth in the PC2 pattern but since 2004 and with a marked interannual variability.This behavior aligns with the hydroclimate variability toward the northern sector of the MC region at the transition with the Atacama Desert (Garreaud et al 2009, González-Reyes et al 2017).The year showing the most severe reduction in growth over the 1982-2015 interval was during the strong 1998 La Niña event, following the strong 1997 El Niño event.Our results also indicate a persistent and severe decline in radial growth across the MC region around the year 2007 onwards depicted by the PC1, in parallel to the SM reduction indicated by the FLDAS dataset.The strong decline around the year 2007 aligns with the regional megadrought identified by Garreaud et al (2017), it initiates several years earlier.This may suggest that SM and tree-growth in MC recorded aridity changes beyond precipitation scarcity, integrating the combined effect of precipitation reductions and the increment of heatwaves and extreme temperatures during the last decades (González-Reyes et al 2023).This suggests that environmental and aridity changes in the MC region started around the year 2007, preceding the marked precipitation reduction in 2010, and today stand up as the most severe drought between Mediterranean ecosystems worldwide (Miranda et al 2023).Our findings also indicate that SM is the main factor controlling plant growth patterns at the regional scale in the MC region, and is the main forcing behind the so-called MC forest decline (Miranda et al 2020, 2023, Venegas-González et al 2023).
Heatwaves severely affect SM during the springsummer growing season in MC (González-Reyes et al 2023).This could be amplified during drought episodes such as the current megadrought in Central Chile.Recent studies have addressed the role of droughts in amplifying the effects of heatwaves, leading to severe SM decline (Miralles et al 2014).A study conducted by Piticar (2018) using weather stations across MC identified an increase in heatwaves during the period 1961-2016.Using a high-resolution maximum temperature gridded, González-Reyes et al (2023) reported similar results.The severe forests-browning (Miranda et al 2020(Miranda et al , 2023) ) and tree-growth decline (Venegas-González et al 2023 and this work), could be related to the combined effect of the moisture decline reported in the present study, the precipitation decline during the megadrought (Garreaud et al 2017), and the regional increase in heatwaves in the MC region (González-Reyes et al 2023).

Four hundred years SM dynamics and changes
Our results indicate that the recent multispecies treegrowth decline reveals a severe and unprecedented aridity change in Mediterranean Chile within the context of the last 400 years.The developed SM reconstruction represents the first of its type in South America and records several low and high multidecadal intervals from 1616 to the present in the MC region (figure 5).These periods are also identified in other MC hydroclimate reconstructions, such as the South American Drought Atlas (SADA; Morales et al 2020), from which our SM reconstruction is almost totally independent, sharing just one of our seventeen chronologies utilized as predictors.In addition, the PC1 nests exhibit a robust spatial pattern in terms of Pearson correlations with SADA during the 1928-2000 (figure S5).On the other hand, the remarkable relationship between SM and MC woody species growth indicates the high vulnerability of MC forest ecosystem dynamics to SM variations and changes.Based on these relationships, the present SM reconstruction reproduces by far the highest observed amount of variance compared to any hydroclimate reconstruction in South America (i.The regional MC May-April SM reconstruction presents a strong negative correlation with the mean annual intensity of the Southern Pacific Anticyclone (SPAI; table 1, figure 6).The reconstruction shows a close relationship with the intensity of pressure fields across the Pacific Ocean, particularly at the Southern Hemisphere mid-latitudes where the SPA is located, indicating the key role of this forcing on SM dynamics in the MC region.Conversely, positive and significant correlation values are identified along the Austral Ocean, specifically over the Amundsen-Bellingshausen Sea.Over this circum-Antarctic region, variations in the intensity of the low-pressure system called Amundsen Sea Low occur (Hosking et al 2016), playing a key role as a blocking center of the storm tracks at higher latitudes favoring the incursion of wet air masses at the MC latitudes.Precipitation in MC is mostly a result of frontal passages and moisture advection associated with midlatitude cyclones, therefore weaker SPAI allow for greater precipitation (Barrett and Hameed 2017).The strong negative (positive) correlations between MC SM and SLP all across the mid-(high-) latitudes highlights the previous relation (figure 6), demonstrating the key role of SPAI allowing the passage of storm tracks over MC.In addition, significant and an inverse relationship between the SPAI developed by Barrett and Hameed (2017) and our SM reconstruction are also recorded during the 1982-2021 period (figure S6).
ENSO also plays a role on MC SM variations, with the Southern Oscillation Index (SOI) correlating during winter SM and the N3.4.region sea surface temperatures (SST) during summer.During El Niño (La Niña) years, an enhanced (decreased) subtropical moisture transport leads to larger values of precipitable water and heightened zonal integrated water vapor transport, resulting in greater (lesser) precipitation in MC (Campos and Rondanelli 2023).Spatial correlations between the regional May-April SM reconstruction and SST show positive correlations in the tropical Pacific, particularly toward the subtropical east in front of the Chilean coast, and in the Amundsen-Bellingshausen Sea in the Austral Ocean (figure S7).During spring-summer, the interdecadal Pacific Oscillation (TPI-IPO index) also plays a relevant role in the variability of MC SM (table 1).In addition, the decadal to multidecadal variability highlighted in our reconstruction is likely influenced by IPO variability (figure S8).The strong negative correlations of MC SM with SLP over the subtropical Pacific Ocean and the SPAI variability are mainly determined by the subsiding branch of the Hadley Cell in the Southern Hemisphere (Fahad et al 2020, Flores-Aqueveque et al 2020, Villamayor et al 2022).These correlations are in concordance with the expected influence of this circulation pattern over the subtropics (Nguyen et al 2013(Nguyen et al , 2015, figure , figure 3).The descending branch of the Hadley Cell can be considered the 'edge of the tropics' , and determines the subtropical high-pressure belt, influencing the latitudinal distributions of precipitation, clouds cover, relative humidity, and drought severity of the subtropical region where MC is located.Over the last four decades the Hadley Cell has been expanding poleward, contributing to increased droughts over many subtropical regions (Lucas et al 2012).This expansion has been attributed to a combination of natural and anthropogenic climate forcings, with increasing greenhouse gasses concentrations and ozone depletion over Antarctica playing a key role (Nguyen et al 2015, Waugh et al 2015).Presently, anthropogenic forcing appears as one of the leading factors of precipitation and aridity change in MC (Boisier et al 2016(Boisier et al , 2018)).

Conclusions
We determined the relationships between May-April SM content derived from the FLDAS global model and radial growth of nine woody species distributed in 22 sampling sites across Mediterranean Chile.Three of these species have been used for the first time in ring-width studies (B. montana, E. revoluta, and N. glauca).Chronologies from several tree species, such as B. miersii and C. alba, reveal a strong relationship with SM variations during the period 1982-2015 (r > 0.80).The first mode of interannual variability in radial growth from the treering network shows a remarkable correlation of 0.91 with soil water content, demonstrating the vulnerability of these species to changes in SM across the MC region.
The ring-width-based SM reconstruction shows a severe SM decline from around the year 2007, revealing a recent unprecedented regional aridity change within the context of the past four centuries.The increase in the pressure intensity of the South Pacific Subtropical Anticyclone appears as the main largescale forcing instigating the present aridity conditions in MC.Our results are in agreement with previous studies indicating that MC is suffering the worst persistent drought of the last centuries (Garreaud et al 2017, Cook et al 2022).The present browning and forest decline in Central Chile stands out as the most severe of its type among Mediterranean ecosystems worldwide (Miranda et al 2023), posing enormous challenges for the management and conservation of this global biodiversity hotspot.The last decade mean regional precipitation of MC is similar to the expected regional precipitation amount for the period 2070-2090 under RCP8.5 scenario (Boisier et al 2018).Detailed assessments of MC future SM variability would be relevant to forecast the potential effects of climate change over the threatened forest ecosystems of Central Chile.

Figure 1 .
Figure 1.Left: Elevation map of the Mediterranean region of Chile (MC) in the south-western coast of South America and the location of 22 tree-ring sites from the nine sampled species: Austrocedrus chilensis (tree), Beilschmiedia miersii (tree), Berberis montana (shrub), Cryptocarya alba (tree), Escallonia revoluta (shrub), Kageneckia angustifolia (tree), Nothofagus glauca (tree), Nothofagus macrocarpa (tree), and Proustia cuneifolia (shrub), see species pictures in figure S1.Elevation (in meters) is based on the Shuttle Radar Topography Mission SRTM-DEM at 90 × 90 m product.A detailed characterization of the tree-ring chronologies is presented in table S1.The black line indicates the Chilean-Argentinian border which follows the continental water divide along the Andes.Right: Contrasting images showing the severe effects of the current megadrought over the evergreen sclerophyll forests from the Altos de Cantillana natural reserve (34 • S) during the summer of 2018 (top) and 2020 (bottom).

Figure 2 .
Figure 2. Pearson correlation coefficients between each from the 22 tree-ring width chronology from nine species and 10 × 10 km gridded monthly averaged during hydrological year (May-April) FLDAS soil moisture (SM) at local (SM grid corresponding to chronology location) and regional (Mediterranean Chile region see figure 4(a)) scales for the 1982-2015 period.In both cases, the Pearson correlation coefficients were calculated using 2000 bootstrap simulations.Chronologies are ordered from north (left) to south (right) latitudinal gradient and details about species, locations and specific statistics are indicated in tableS1.

Figure 3 .
Figure 3. Main spatio-temporal -growth patterns of the 22 tree-ring chronologies from nine woody species.The (a) and (c) panels correspond to the spatial loadings of the leading two empirical orthogonal function (EOF) modes for the 22 tree-ring chronologies (the size of the symbols is in accordance with the value of the loading).Time series of the corresponding leading modes (PC1 and PC2) explaining 27% and 10% of the variance of the entire tree-ring network during the 1970-2015 common period.

Figure 4 .
Figure 4. (a) Spatial correlation field between the PC1 time-series derived from the 22 tree-ring chronologies from nine woody species and hydrological year (May to future April) regional soil moisture (SM) average of FLDAS product.The Pearson correlation coefficients were calculated using 1000 bootstrap simulations.(b) Temporal relationships between PC1 and the regional average SM of the entire Mediterranean Chile domain.The Pearson 'r' coefficient reported has been based on 2000 bootstrap simulations.

Figure 5 .
Figure 5.The tree-ring reconstruction of the hydrological year (May-April) soil moisture (SM) for the Mediterranean Chile region during the period 1616-2018.(a) Observed and tree-ring predicted SM utilizing 17 selected tree-ring chronologies from 8 species.Calibration and verification statistics: explained variance (adj-R 2 ) over the calibration period, the Pearson correlation coefficient (r) between observed and reconstructed values, the residual error (RE), and the root-mean-square error (RMSE) are indicated.(b) May to April SM reconstruction covering the 1616-2018 period smoothed with a 20-year cubic spline.The shaded area denotes the 1 ± RMSE envelope.(c) Explained variance (adj-R 2 ) over the calibration period and the respective number of tree-ring chronologies utilized to develop each reconstruction nest based on its first principal component.The selected chronologies utilized to develop the SM reconstruction are indicated in tableS1.

Figure 6 .
Figure 6.Spatial correlation field between the May to April Mediterranean Chile tree-ring based soil moisture (SM) reconstruction and 0.25 × 0.25 gridded May to April averaged sea level pressure for the interval 1982-2021 (ERA5 reanalysis).The location of the reconstructed SM region is indicated by the green star.Pearson's r coefficients are significant with a p-value < 0.05 when r = ± 0.31.
figure2) show the highest correlation values among the other species used in this study.Our reconstruction demonstrates the exceptional nature of the present MC megadrought within the context of the last 400 years.Recent studies have reported the severe growth decline in several plant species in the MC as a result of the current unprecedented megadrought during the last decade(Garreaud et al 2017, Miranda et al 2020, 2023, Venegas- González et al 2018, 2023).These plant species and probably many others from the MC region, have confronted extreme climate and environmental conditions due to the persistent severe aridity conditions after 2010(Garreaud et  al 2020, Matskovsky et al 2021, Venegas-González et al 2023).The propagation of this megadrought through the water cycle has led to a substantial decline in snowpack, glacier area and streamflow with the subsequent negative effects in water availability across Central Chile (Cordero et al 2019, Masiokas et al 2019, Ayala et al 2020, Muñoz et al 2020, Garreaud et al 2020, Alvarez-Garreton et al 2021), triggering unfavorable consequences for the Mediterranean forests in Central Chile (Miranda et al 2023).This biome provides a variety of fundamental ecosystem services to the society (Gajardo-Rojas et al 2022, Smith-Ramírez et al 2023), and have been proposed as a key element to implement nature based solutions for water security for the region e. LeQuesne et al 2006, LeQuesne et al 2009, Christie et al 2011, Muñoz et al 2020, Hadad et al 2021).
. For the respective references see Methods section 2.4.
(Hersbach et al 2020)le 1).At a geographic level, the reconstruction of SM and the mean SLP field during May to April, which was obtained from the ERA5 reanalysis(Hersbach et al 2020), show substantial negative correlations, primarily over the Tropical and mid-latitudes of the Pacific Ocean (figure6).Positive correlations are observed at high latitudes over the Amundsen Sea low domain, demonstrating the role of circum-Antarctic blocking centers on MC hydroclimate variability.