Tree growth influenced by warming winter climate and summer moisture availability in northern temperate forests

The role of future forests in global biogeochemical cycles will depend on how different tree species respond to climate. Interpreting the response of forest growth to climate change requires an understanding of the temporal and spatial patterns of seasonal climatic influences on the growth of common tree species. We constructed a new network of 310 tree‐ring width chronologies from three common tree species (Quercus robur, Pinus sylvestris and Fagus sylvatica) collected for different ecological, management and climate purposes in the south Baltic Sea region at the border of three bioclimatic zones (temperate continental, oceanic, southern boreal). The major climate factors (temperature, precipitation, drought) affecting tree growth at monthly and seasonal scales were identified. Our analysis documents that 20th century Scots pine and deciduous species growth is generally controlled by different climate parameters, and that summer moisture availability is increasingly important for the growth of deciduous species examined. We report changes in the influence of winter climate variables over the last decades, where a decreasing influence of late winter temperature on deciduous tree growth and an increasing influence of winter temperature on Scots pine growth was found. By comparing climate–growth responses for the 1943–1972 and 1973–2002 periods and characterizing site‐level growth response stability, a descriptive application of spatial segregation analysis distinguished sites with stable responses to dominant climate parameters (northeast of the study region), and sites that collectively showed unstable responses to winter climate (southeast of the study region). The findings presented here highlight the temporally unstable and nonuniform responses of tree growth to climate variability, and that there are geographical coherent regions where these changes are similar. Considering continued climate change in the future, our results provide important regional perspectives on recent broad‐scale climate–growth relationships for trees across the temperate to boreal forest transition around the south Baltic Sea.


| INTRODUC TI ON
The productivity, composition and functional role of future forests in global biogeochemical cycles will depend on how different tree species respond to climate, competition with neighbours, and local environmental conditions. Climate limitations on forest ecosystems are thought to be greatest in high elevation and latitude environments and also in moisture-limited regions where extreme climate events, such as drought, can cause widespread forest mortality (Allen et al., 2010;Anderegg, Kane, & Anderegg, 2013). However, recent research documents considerable climate change effects on temperate forests where reduced timber yield (Lindner et al., 2014), declining resilience (Cavin & Jump, 2017;Gilliam, 2016;Zang, Hartl-Meier, Dittmar, Rothe, & Menzel, 2014), reduced biodiversity (Thom & Seidl, 2016) and increased disturbance activity (Millar & Stephenson, 2015;Senf & Seidl, 2018) have been attributed to warming temperatures and/or reduced moisture availability in the growing season.
Each of the last three decades has been successively warmer than any preceding decade since 1850, with 1983-2012 likely the warmest 30 year period in the last 1,400 years in the Northern Hemisphere (IPCC, 2014). Climate change effects on vegetation have been documented at annual and seasonal scales, with the most research directed at evaluating the effects of climate in summer months. However, climate warming is more pronounced in winter than summer at locations in the mid-to high-latitudes (Xia et al., 2014), where changing winter precipitation patterns (spatial and temporal variability, type; Wu et al., 2019), extreme cold events (magnitude and frequency; Kodra, Steinhaeuser, & Ganguly, 2011) and temperature (variability in extremes and mean) can independently and collectively influence patterns of vegetation growth and distribution, especially in temperate mixed-wood forests (Kreyling, 2010;Kreyling & Henry, 2011). To better identify the effects of these climatic changes on forest productivity, spatially explicit estimations of tree growth (i.e. radial tree-ring growth) responses to seasonal climate are required. Furthermore, analyses assessing the stationarity of these climate-growth responses are necessary to determine changing climate effects on tree growth.
Recent research using broad-scale tree-ring networks have documented changing effects of seasonal climate on temperate and Northern Hemisphere forests (e.g. Babst et al., 2019). The advancement of methods to integrate tree-ring measures into estimates of hemispherical forest productivity (e.g. Seftigen, Frank, Björklund, Babst, & Poulter, 2018), continental vegetation models (e.g. Klesse, Babst, et al., 2018) and future forest growth relies on the development and evaluation of new tree-ring data networks . Tree-ring networks can be used to characterize the spatial and temporal patterns in the relationship between tree growth and climate and improve our ability to model the impacts of future climate (Charney et al., 2016;Klesse, DeRose, et al., 2018), especially for temperate, complex forests that provide diverse ecosystem services.
Given the uncertainty of future climate, an improved understanding of the influence of climate on tree growth in temperate forest ecosystems is needed, especially for winter forest ecology, which has received comparatively less research attention (Weigel et al., 2018).
In this study we developed a multispecies network of tree-ring width data in the temperate oceanic and continental forests surrounding the south Baltic Sea to assess climate-growth relationships at a subcontinental scale. This region includes hemiboreal forests poleward of the study region but is primarily composed of temperate forest ecosystems. Incorporating existing and new tree-ring data, this network is compiled from site-level chronologies, aiming to reduce the individual tree variability that stems from microsite differences, within-population genetic variation and small-scale disturbances. With this composited dataset, we test the assumption that species-specific responses to climate influence tree-growth variability across this generally temperate region, and that this relation is further characterized by different genus-specific responses. The objectives of our study were: (a) to determine the broad species-specific response to climate variables, (b) to investigate the dominant climate factors influencing the growth of common tree species in this region, (c) to explore the temporal stability of climate-growth relationships identified and (d) to assess the spatial pattern of climate-growth responses.

| Study region
The south Baltic Sea region includes the postglacial lowland landscapes (~0−400 m above sea level [a.s.l.]) of Denmark, Sweden, Germany, Poland, Kaliningrad region of Russia, Lithuania, Latvia and Estonia surrounding the south Baltic Sea (within ~300 km), south of the gulfs of Bothnia and Finland and east of Skagerrak, lying between ~51-60°N and 9-28°E ( Figure 1). Agricultural cropland, managed and protected forests and peatland complexes typify the land cover types outside urban and settled areas. Both maritime and continental weather patterns modulate climate across the study region set between the continental interior of Europe and the northern Atlantic Ocean. Cold, moist winters and warm, wet summers characterize the regional climate, with an average 604 mm of annual precipitation and a mean annual temperature of 7.8°C in the southwest of the region (Rostock; 1981-2010 climate normals; CRU TS 4.01) and 704 mm and 5.9°C in the northeast of the region (Tallinn; 1981-2010 climate normals; Estonian Weather Service, 2018). Over the past 100 years, air temperature in the study region has increased in both the warm and cold seasons ( Figure S1). Winter precipitation in some poleward areas of the study region have experience increased precipitation, and summer precipitation at some equatorward areas record less precipitation in the late 20th and early 21st centuries ( Figure S1).
Pedunculate oak (hereafter referred to as 'oak') is the most widely dispersed oak species in the south Baltic Sea region. Oak trees generally tolerate a wide range of edaphic conditions with optimal growth most frequently observed on fertile and well-watered sites (Arend, Kuster, Günthardt-Goerg, & Dobbertin, 2011;van der Maaten et al., 2018). Scots pine is the most broadly distributed pine species found throughout Eurasia. A high tolerance to poor soils, drought and frost events permits Scots pine trees to grow in a wide variety of environments ranging from 0 to 2,600 m a.s.l., and also commonly occurring as an early seral species in disturbed settings (Durrant, De Rigo, & Caudullo, 2016). European beech is found growing from southern Italy to southern Norway and Sweden, occurring in various habitats from mountainous regions to lowlands (Seynave, Gégout, Hervé, & Dhôte, 2008), with cold winter temperatures and low moisture availability restricting beech from more northern and/ or drier locations (Peters, 1997).
At all sites, including those not previously published (Table S2), one or two increment cores were collected ~1.3 m above the base of ≥10 trees with an increment borer. Tree cores were dried and fixed to wooden mounting boards and sanded with progressively finer sand paper or cut with a sharp blade so the ring boundaries could be clearly identified. Ring widths were measured to the nearest 0.01 or 0.001 mm using either digital measuring software F I G U R E 1 The distribution of Quercus sp. ('Oak'), Pinus sylvestris ('Scots pine') and Fagus sylvatica ('E. beech') in Europe (left panels; species distribution: http://www.eufor gen.org/speci es/), and the location of tree-ring width data sites in the network (circle and cross markers) included in the analysis (cross marker) (right panels). Inset black boxes in left panels mark the study region (e.g. WinDENDRO, CooRecorder) or a measuring table system (e.g. Lintab, Velmex).
To assign each ring to a calendar year, ring widths were visually crossdated and statistically verified using COFECHA (Holmes, 1983) or CDendro (Cybis Electronik & Data, Sweden). In the case of two cores per tree, annual ring measurements were averaged between cores to yield one series per tree. All raw, crossdated ring-width series were detrended to remove age-related trends and other nonclimatic influences on growth (e.g. growth releases). The variance in each ring-width series was stabilized by adaptive power transformation to produce homoscedastic indices (Cook & Peters, 1997), and then a spline function with a 50% frequency cut-off at 32 years was fit to each time series using ARSTAN (Cook, 1985). Residuals from the detrending functions were computed via subtraction (Helama, Lindholm, Timonen, & Eronen, 2004). For each site, individual, detrended ring-width series for each species were combined into an annual chronology using a biweight robust mean (Cook, 1985). To  (Table S1). This period was the longest interval we could select that included the most chronologies.
The complete Baltic Sea tree-ring network presented here includes 136 oak, 119 Scots pine and 55 beech sites (Table S1), including series from 6,121 individual trees. Detrended standardized chronologies for each site are presented in Table S2. In this study, we present 42 new and previously unpublished chronologies that include five oak, 18 Scots pine and 19 European beech new treering width datasets that span intervals within 1647-2016 (Table S1).
Interpolated climate datasets are calculated by this frontend software using the Parameter-elevation Regressions on Independent Slopes Model (Daly et al., 2008) for precipitation and ANUSplin for temperature, and are based on the CRU TS 3.1 climate dataset (Mitchell & Jones, 2005). Monthly precipitation sums and monthly mean, maximum and minimum temperature for each month from January to September in the growth year and June to December in the previous year were obtained. Seasonal temperature and precipitation variables were also computed by averaging monthly values (summer: June-August; autumn: September-November; winter: December-February; spring: March-May). We calculated the standardized precipitation-evapotranspiration index (SPEI; Beguería, Vincente-Serrano, Reig, & Latorre, 2014) from the monthly temperature and precipitation datasets. SPEI is a multiscalar climatic drought index that can be computed for multiple temporal scales and considers precipitation and the effect of temperature on drought severity by including evapotranspiration (Vicente-Serrano, Beguería, & López-Moreno, 2010). In this study, we used the Thornthwaite equation to estimate potential evapotranspiration (Thornthwaite, 1948) and calculated SPEI integrated over 1-, 3-and 6 month periods (SPEI1, SPEI3 and SPEI6; Beguería & Vicente-Serrano, 2017). Both precipitation and SPEI climate variables are used to describe 'moisture availability'.

| Tree growth responses to climate
To identify the climate-response patterns across sites and species in the study region, the response of the ARSTAN standard chronologies to monthly climate variables was calculated over the 16 month time window (i.e. from the previous June to current September) and for seasonal climate variables. We performed bootstrapped Pearson's correlation analysis with 5,000 replicates to identify the predominant factors contributing to the observed growth variability. Climate-growth relationships were calculated over the common periods of 1943-2002 (60 years), 1943-1972 (30 years) and 1973-2002 (30 years). We then focused on two climate-growth relationships for each tree species (oak, Scots pine and European beech): (a) the dominant climate variable influencing the regional growth of a species across the study region ('dominant climate response'), and (b) the climate-growth relationship that exhibited the greatest change in strength and/or direction when we compared the response for the two 30 year periods ('greatest change climate response'). The strongest average response among sites for each species and climate variable was used to determine these relationships. We then assessed the temporal stability of these two climate-growth relationships for each tree species in more detail by employing bootstrapped correlation analysis with 5,000 replicates in 31 year moving windows lagged by 1 year over the 1943-2002 period. To assess the effect of autocorrelation in these analyses, we conducted the same climate-growth analyses described above using the ARSTAN residual chronologies that were processed using autoregressive modelling to remove autocorrelation (Cook & Holmes, 1986).

| Spatial patterns of changing climate-growth responses
We employed spatial segregation analysis to describe any broad spatial patterns in the stability of the growth responses for the two types of climate-growth relationships (see (a) and (b) in previous section). More commonly applied in epidemiological research of infectious disease outbreaks (e.g. Diggle, Zheng, & Durr, 2005;Tarr et al., 2018), spatial segregation analysis can be used to describe structure in spatial point patterns over space. First, for each site we classified the climate response as 'stable' or 'unstable' by comparing the significance of the correlation coefficients at the 0.95 confidence level for the 1943-1972 and 1973-2002 periods for both of the climate-growth relationships.
An 'unstable' change in climate response between the two periods was determined if one of the following criteria was met: (a) the sign (positive, negative) of a significant response changed while responses remained significant, (b) the response changed from statistically insignificant to significant in either direction, (c) the response changed from statistically significant to insignificant. If any one of those criteria were not met, a 'stable' classification was assigned. Considering the site-level stability estimates were based on bootstrapped correlations, and in order to prevent spurious correlations that we could possibly expect at eight of the 152 sites per each bootstrap iteration (95% confidence level) from affecting the further analysis, we iterated the entire bootstrap procedure 999 times. Sites were finally classified if a consistent classification was assigned in at least 65% of the bootstrapped iterations. This threshold was selected to include as many sites as possible, while excluding the sites that classified inconsistently-most likely due to spurious correlations. Sites that did not meet this criterion were excluded from this analysis. Two sites from dominant climate response and five sites from the greatest change climate response were excluded.
The site-level classifications of the climate correlation stability were pooled into two datasets (regardless of species) representing the two climate-growth relationships (dominant and largest change climate responses). For each of the datasets we evaluated if spatial synchrony in 'stable' and 'unstable' changes in the climate responses was present using spatial segregation analysis with 1,000 Monte Carlo iterations (Zheng & Diggle, 2013). For any point in space, spatial segregation analysis determines the probability that a point is 'stable' or 'unstable' based on the stability classification of the sites.
The probability that a point will be stable was estimated by averaging the probability distributions obtained by kernel smoothing estimations of each site (Diggle et al., 2005), where 1 is a 'stable' site and 0 is 'unstable'. Finally, the p-values of the Monte Carlo spatial segregation test were mapped across the study region highlighting the probability of similar changes in climate-growth responses for both the dominant and greatest change climate responses.

| Monthly and seasonal climate responses
Over the period 1943-2002, correlations between climate variables and tree growth varied among the species examined ( Figure S2).
Although variability in correlation strength and direction was detected among sites for each climate variable and for each tree species, we report here (a) the broadly dominant relationship between tree growth and climate, and (b) the climate-growth relationship exhibiting the greatest change (strength and/or direction in correlation value) over the 1943-2002 period. Overall, the climategrowth relationships reported are generally weak to moderate (r = −0.4 < 0 < 0.4; Figure S2), which is consistent with other treering width studies in this generally temperate region (cf. Babst et al., 2013;Scharnweber et al., 2011;Sohar et al., 2014).
Across the oak sites, growth was most strongly and significantly related to June and summer precipitation, as well as monthly measures of previous summer/autumn SPEI1 from 1943 to 2002 (Figure 2; Figure S2). June precipitation in the growth year was most consistently and positively related to oak ring width across the study region and this relationship remained generally stable for many sites over the 60 year period when the 1943-1972 and 1973-2002 intervals are compared (Figures 2 and 3), and in moving window correlation analysis ( Figure 4). When correlation values for each climate variable are compared between the two 30 year periods, the greatest change in growth response is with average February temperature (Figures 2 and 3).  (Figures 2 and 3).
Across the European beech sites, growth was most strongly and significantly related to June precipitation in the growth year from 1943 to 2002 (Figure 2; Figure S2). Over the 60 year period, beech growth exhibits a slight increase in the positive response to June precipitation at many sites (Figures 2, 3 and 4). Despite this increase, we considered June precipitation to have the most consistent and strong effect on beech growth. When the climate-growth responses for intervals 1943-1972 and 1973-2002 are compared, the greatest change was detected for average February temperature (Figures 2 and 3). At  Figure S3).
Here we focus on growth responses to mean temperature; similar responses are found with minimum and maximum temperature ( Figure S4). For all studied tree species and all climatic drivers, very similar relationships between growth and climate parameters were found using the residual chronologies (where autocorrelation is removed; Figure S5).  1943-1972 and 1973-2002. Months are labelled with the first letter of the month, with lower letter case indicating previous year and upper letter case indicating current year (e.g. current year January = 'J', previous year August = 'a'). Box plots show median, lower and upper quartiles (25% and 75%) of bootstrapped correlations. Black dots are outlier correlation values that fall outside these quartiles F I G U R E 3 Spatial distribution of the correlations between tree growth and the dominant monthly climate variable influencing tree growth (a), and the climate-growth relationship that exhibited the greatest change (b) when the two subperiods of 1943-1972 and 1973-2002 Figure 5). The high density of sites and variability in 'stable'

| Spatial patterns in changing climate-growth responses
and 'unstable' classifications in northeastern Germany resulted in no significant segregation of sites in that area of the study region ( Figure 5).

| D ISCUSS I ON
This tree-ring network is the largest tree-ring dataset for the south Baltic Sea region and presents a balanced perspective of tree growth by relying on site-level, tree-ring measurements collected for different purposes including ecological, climatological and management applications. Our interpretations focus on broad climatic speciesspecific controls on tree growth across the study region. Across the species examined, growth responses to early spring and summer climate were identified as regionally dominant, while responses to some winter climate variables exhibited the greatest change. We focus our interpretations on these two broad patterns.

| Dominant climate-growth relationships
June precipitation had the greatest overall influence on the growth of deciduous species (albeit weakly for oak) suggesting that moisture variability broadly influences the growth of oak and European beech in the study region from 1943 to 2002 (Figure 2; Figure S2). The absence of a stronger and more spatially coherent relationship between oak growth and a dominant climate factor is consistent with previous studies near the study region that report an influence of site conditions on tree growth (e.g. Friedrichs, Trouet, et al., 2009;Rybníček et al., 2016). Furthermore, Haneca, Čufar, and Beeckman (2009) summarize the diverse climate-growth associations made between oak growth and climate variables in European studies, which include growth sensitivities to both temperature and precipitation.
Despite the variability in the growth response of oak to climate and site conditions, prior studies indicate that summer moisture availability broadly controls oak tree growth in Europe (Friedrichs, Büntgen, et al., 2009;Haneca et al., 2009;Neuwirth, Schweingruber, & Winiger, 2007). In our study, there is some evidence for an increasing sensitivity of oak growth to moisture availability in more recent decades in northeastern Germany, with the strength of the correlated relationship between growth and June precipitation increasing at some sites in the latter 30 year period (Figure 3; Figure S10).
This result lends support to the findings presented in Babst et al. June precipitation in the growth year exists (Figures 2, 3 and 4). This result indicates that similar to oak, moisture availability in the growing season predominantly influences annual incremental growth of beech (Babst et al., 2019) and is further supported by studies conducted elsewhere throughout the range of beech (e.g. Cavin & Jump, 2017;Hacket-Pain, Cavin, Friend, & Jump, 2016;Jump, Hunt, & Penuelas, 2006). Variability in the growth responses to precipitation in the early part of the common period appears to decrease over time ( Figure 4). This could be attributed to the early-life growth associated F I G U R E 5 Results of spatial segregation analysis for the dominant climate-growth relationship and the climate-growth relationship that exhibited the greatest change in strength and/or direction over the 1943-2002 interval. In this analysis, a probability surface was determined by kernel-based estimation of spatial segregation. Dark red shading marks areas with a significantly higher probability of similar unstable growth change responses (segregated unstable), while dark blue shading suggests regions with a significantly higher probability of stable responses (segregated stable). See Section 2 for criteria used to classify correlation stability with the late-20th and early 21st century start dates of some beech chronologies (Table S1) and the response change to environmental variability with age. Our results also indicate stronger positive responses of beech growth to June precipitation and monthly SPEI1 ( Figure 2) in recent decades, suggesting that the role of moisture availability in the growing season is becoming increasingly important for growth. Recent studies on the sensitivity of European beech trees to drought events in Europe have documented an increasing response in recent decades (Latte, Lebourgeois, & Claessens, 2015), especially in the central range of the species (Cavin & Jump, 2017) and lowland settings (Latte et al., 2015). Perhaps warming temperatures in the early spring extend the growing season in the south Baltic Sea region (Vitas, 2011), promoting productivity and growth. Scots pine, as an evergreen species, can capitalize on the early onset of these winter climate conditions conducive to growth (Antonova & Stasova, 1993;Gruber, Strobl, Veit, & Oberhuber, 2010). At sites further north (i.e. northern Norway) late winter and early spring climate conditions likely remain too harsh to support growth and summer temperature remains the primary factor limiting growth (Rossi et al., 2008).

| Unstable growth responses to winter climate
In many of the European beech and oak populations we studied, the relationship between tree growth and late winter temperature weakens in recent decades (Figures 2 and 3). We suggest that in the early portion of the study interval , winter cold may have caused reduced growth due to occurrence of winter xylem embolism (Cochard, Lemoine, Améglio, & Grainier, 2001;Pederson, Cook, Jacoby, Peteet, & Griffin, 2004;Sperry & Sullivan, 1992) and frost events (Kollas, Körner, & Randin, 2014). WInter embolisms disrupt the conductance of water in the stem of a tree leading to xylem dysfunction, a process documented in beech and oak trees in Europe including temperate and 'warm' regions (e.g. Cavender- Bares et al., 2005). Pederson et al. (2004) postulated that in anomalously cold years with more frequent or severe embolism occurrence, resources that could be used for early growth would instead be used for recovery. A recent study by Weigel et al. (2018) reports a relationship of beech growth reductions and winter cold that was enhanced towards cold-marginal populations. Furthermore, Weigel (2019) conducted a winter manipulation experiment in beech forests in northern Germany and Poland that shows winter cold, through sublethal root damage or reduced root nutrient uptake, affects beech growth negatively, a finding that is also supported by results presented in Reinmann, Susser, Demaria, and Templer (2019). In the context of our study, this finding could indicate that winter soil warming was more important in the early period  and is becoming less important with recent climate warming ( Figure S6). Not all previous research supports a positive relationship between deciduous tree growth and late winter temperatures, with some studies documenting fast acclimation to freezing, positing that winter temperature plays a small role in shaping the cold range limit of beech (e.g. Lenz, Hoch, & Vitasse, 2016;Matisons, Purina, Adamovičs, Robalte, & Jansons, 2017).
A notable change in climate-growth sensitivity was found when the response of Scots pine growth to January temperature was com-  Figure S1) and over the interval of analysis ( Figure S6) show a generally increasing trend that is consistent with recorded climate warming for this region (IPCC, 2014). Therefore, we suggest an increasing influence of January temperature as a factor affecting the growth of Scots pine trees in this region, a finding supported by conifer growth responses to winter climate by Babst et al. (2013). The physiological rationale for a late winter growth relationship has been discussed for some tree species (Pederson et al., 2004), however, the biological processes underpinning this persistent growth response of Scots pine trees are not well explained. The winter growth sensitivity of Scots pine could be driven by winter photosynthesis (positive carbon gain when needles are not frozen; Havranek & Tranquillini, 1995) and/or frost damage to green foliage during winter (Ensminger et al., 2004;Pederson et al., 2004). Winter photosynthesis in this region is probable provided that the freezing depth of soils in the winter is shallow (Ensminger, Schmidt, & Lloyd, 2008). Additionally, warm earlygrowth season temperatures (i.e. warmer January temperatures in recent decades) likely advance the onset of the vegetation period influencing the overall length of the growing season (Vaganov, Hughes, Kirdyanov, Schweingruber, & Silkin, 1999), where shorter cold seasons result in increased productivity (wider tree-ring) and conversely, longer cold seasons would lead to decreased productivity (narrower tree-ring). Therefore, we postulate that enhanced winter photosynthetic activity driven by warmer late winter/early spring conditions coupled with possible longer growing seasons as plausible explanations for this climate-growth response.

| Spatial segregation of climate-growth responses
Significant segregation of 'stable' sites in the northeast of the study region indicates the dominant climate growth responses have remained stable between the two 30 year periods ( Figure 5). This finding suggests that the growth of Scots pine trees in the northeast of the region has been consistently influenced by mean March temperature and the growth of oak by June precipitation. Furthermore, it could be considered that the influence of recent climate change on the dominant climate factors influencing tree growth in forests in the northeast of the study area is lower, at least for the climate variables considered here. A unit of segregated 'unstable' sites centred mainly in northeastern Germany was detected, and we suggest that the slightly increasing sensitivity of deciduous tree sites to June precipitation over the interval of analysis (Figure 3) resulted in many sites being classified as 'unstable' prior to the segregation analysis.
For the greatest change climate response, sites classified as 'unstable' in eastern Poland, Lithuania, and eastern Latvia aggregate to form an eastern unit of primarily Scots pine and some oak sites, which suggests a high probability for changing winter climate drivers of growth ( Figure 5). We suggest that as winter climate conditions have warmed ( Figures S1 and S6), trees previously limited by cold winter or spring conditions in this unit have responded similarly. This finding suggests that within the south Baltic Sea region, tree-ring sites in the eastern part of the study region may exhibit a higher probability of unstable climate-growth relationships to the winter climate variables examined in this study. In southern Sweden, most sites were classified as 'stable', indicating the effects of winter climate have remained spatially and temporally consistent, although the spatial distribution of sites is notably lower here ( Figure 5).
Our application of spatial segregation analysis combines species responses to different climate variables for the two types of climate-growth responses, and applies generalized rules to classify climate-growth correlation stability. A partitioning of the network and subsequent runs of this analysis with (a) only Scots pine sites and (b) beech and oak sites combined ( Figures S7 and S8), supported the interpretations from the multispecies-based spatial analysis. In order to gain a more precise estimation of spatial segregation patterns, more work is needed to buffer edge effects by increasing the spatial distribution of sites and adapt the analysis for a discontinuous spatial domain (i.e. remove the small-scale nonterrestrial landscape elements) for the kernel estimation. Therefore we present this descriptive analysis as a system-level perspective of the forest response to 20th century climate. It also represents a new approach in linking temporal variability in biological responses, represented as tree-ring chronologies, in the spatial domain.

| Possible age effects in tree-ring networks
In large-scale tree-ring network analyses, recent research has indicated that the relationships between climate and tree-ring chronologies could be inflated ('climate sensitivity bias'; Babst et al., 2018), caused by including preferentially sampled datasets that targeted older, bigger trees and/or at sites to maximize their sensitivity to a single climate parameter. Unlike recent network studies that rely heavily on sites from the ITRDB and dendroclimatological studies sampled for specific climate purposes (e.g. Babst et al., 2019), the network presented here includes sites that were collected for ecological and stand management projects in addition to dendroclimatological investigations. Furthermore, the age distribution of individual trees (inferred from site-level mean tree series length) was generally similar among sites classified as 'stable' and 'unstable' in the spatial segregation analysis for the greatest change climate variable ( Figure S9).
A possible exception to this interpretation includes 14 beech sites with average series lengths between 240 and 295 years that were all classified as 'unstable' ( Figure S9) when the climate responses of growth to February temperatures were compared for 1943-1972 and 1973-2002. Previous research has documented an increasing sensitivity of larger trees to climate reflected in growth responses (Mérian & Lebourgeois, 2011;Trouillier et al., 2019), therefore this finding could suggest that bigger, and inherently older, trees at these 14 sites exhibit a more pronounced response to recent winter warming, or simply that smaller trees are more sensitive to frost damage.
Nonetheless, due to the generally similar age-structure of our chronologies we consider the size-and/or age-related effects on climate responses to be reasonably small with a minimal influence on the broad interpretations presented from this network.

| Summary and outlook
The role of tree growth sensitivity to climate in forest carbon dynamics and changing community composition has been highlighted in many types of forests (e.g. Anderegg et al., 2015;Babst et al., 2013;Charney et al., 2016;Clark et al., 2016), including ecosystems that contain species (and populations) that are not typically considered climatically limited (e.g. Cavin & Jump, 2017). In this study we sought to provide insights on tree growth responses to recent climate change in the south Baltic Sea region. Our findings indicate summer moisture availability limits oak and European beech trees, with an enhanced effect in the latter 20th century, and considering the projected climatic drying in sub-boreal and temperate forest ecosystems, a strong influence of moisture availability in the current and previous growing seasons will likely persist. The stationary response of Scots pine growth to February and March temperature over time, and increasing positive response to January temperature likely points to critical early season photosynthetic gains. The documented unstable growth responses to winter temperature for all species examined suggest winter climate change might release some forest trees from cold season climatic limitations in the future. How the responses we documented will contribute to ecosystem level changes in composition and patterns of distribution remains unclear at this point. However, insights from this study can provide valuable information, especially for winter climate responses, to guide the selection of tolerant provenances and refine projections of the expected changes in forest growth and terrestrial carbon budget models needed to support climate change mitigation practices. The authors thank Flurin Babst for contributing data and all ITRDB data contributors for making their tree-ring data publicly available.

ACK N OWLED G EM ENTS
We are grateful to two anonymous reviewers for their constructive feedback to improve this paper. I.D. acknowledges support from CLIMECO funded by the Swedish Institute, Grant 10066/2017.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available in the supplementary material of this article.