Effects of Extreme Drought Years on Radial Growth of Oak and Beech on the Swabian Alb, Germany

Abstract

Climate change is increasing drought frequency and severity in Central Europe, with consequences for the productivity and stability of mixed broadleaf forests. We analyzed annual radial growth of European beech (Fagus sylvatica L.) and sessile oak (Quercus petraea (Matt.) Liebl.) along a water-availability gradient on the Swabian Alb (southwestern Germany). At three nearby sites, 45 trees per species were sampled. Tree-ring chronologies were related to meteorological variables and integrated drought indices (PDSI/scPDSI, SMI). Extreme dry and wet years were identified from detrended growing-season indices, and event responses were quantified using resistance, recovery, and resilience metrics and their humid-year analogues. Annual growth variation was primarily site-driven rather than species-driven, highlighting strong control by local soil water availability. Correlation analyses showed that integrated drought and soil-moisture indices explained radial growth more consistently than single climate variables. Oak showed higher drought resistance than beech in several events, although the magnitude and significance of this difference depended on site and drought year. Resistance generally declined toward recent events, particularly at Site 2, while recovery tended to increase. Recovery did not consistently favor either species, and resilience remained broadly stable, except for a decline in beech during the most recent drought events. Overall, sensitivity to drought is jointly controlled by species identity and fine-scale site conditions: oak has a comparatively robust resistance advantage, whereas post-drought recovery and resilience are strongly context-dependent. These findings support site-specific species mixtures and silvicultural strategies to sustain forest function under increasing drought risk.

1. Introduction

Recent sequences of consecutive warm and dry years increasingly affected forest growth across Central and Western Europe, with substantial ecological and economic consequences [1,2,3]. Drought is now recognized as a major driver of reduced radial stem growth, declining vitality, and elevated mortality risk for many temperate tree species [4,5,6,7]. However, drought impacts are rarely determined by climate alone. They are strongly modulated by local water availability, soil depth, rooting conditions, and stand structure. Therefore, tree-growth responses must be interpreted using both regional climate signals and site-specific indicators of soil moisture and water limitation [8,9].

European beech (Fagus sylvatica L.) and sessile oak (Quercus petraea (Matt.) Liebl.) are particularly relevant in this context. Both species are key components of Central European broadleaf forests and are widely present in mixed stands [10,11,12,13,14]. Current silvicultural guidelines promote mixed broadleaf stands as part of climate-change induced drought adaptation strategies [15,16]. Yet the future role of beech in such mixtures remains uncertain because recent studies report pronounced drought-related reductions in growth and vitality, whereas oak shows comparatively higher drought tolerance [17,18,19,20,21,22,23,24]. It remains unclear whether these species differences are consistent across local water-availability gradients or whether site conditions can amplify, weaken, or even override species-specific drought responses.

The Swabian Alb, a mountain range in southwestern Germany, provides a suitable setting to address this question. Its shallow, calcareous, and partly skeletal soils can generate strong small-scale contrasts in plant-available water [25]. Such heterogeneity allows oak and beech responses to be compared along a dry-to-moist gradient while minimizing broad-scale climatic differences. Tree-ring analysis is especially useful for this purpose because it provides annually resolved growth records and enables retrospective assessment of tree responses before, during, and after extreme years [26].

A challenge is the definition of drought. Single meteorological variables such as precipitation or temperature may not capture the cumulative water deficits that are most relevant for tree growth. Integrated indicators can provide a more biologically meaningful representation of drought stress. The Palmer Drought Severity Index (PDSI) and its self-calibrated form (scPDSI) combine precipitation and atmospheric evaporative demand to estimate cumulative moisture deficits and support comparisons across years and regions [27,28,29,30]. The Soil Moisture Index (SMI) complements these climate-based indicators by representing water availability in the soil profile and thus linking meteorological drought more directly to potential rooting-zone stress [31]. Combining these indicators with tree-ring data enables a more robust assessment of hydroclimatic extremes and their translation into growth resistance, recovery, and resilience.

This study investigates annual radial growth dynamics of sessile oak and European beech at three nearby sites in the Swabian Alb that differ in local soil water availability. We combine chronology statistics, climate-growth correlations, pointer-year analyses, and event-based response metrics to identify drought-vulnerability patterns and derive site-specific implications for the management of drought-prone mixed broadleaf forests.

The specific objectives are:

• to identify extreme dry and wet years using climatic and soil-moisture indicators;
• to quantify interannual growth variability and common growth signals within and between species and sites;
• to determine which climatic and drought-related variables most strongly control radial growth;
• to assess drought response components, particularly resistance, recovery, and resilience, in oak and beech;
• to derive practical recommendations for species selection and stand development under increasing drought risk in southwestern Germany.

Based on previous evidence and regional drought development, we test the following hypotheses:

H1. 

Climate-growth relationships are stronger for integrated drought indicators (PDSI/scPDSI and SMI) than for single meteorological variables.

H2. 

Beech shows stronger growth reductions and lower drought resistance than oak during extreme dry years.

H3. 

Species differences in drought response increase with decreasing site water availability.

H4. 

Recovery and resilience after drought differ between species and site conditions, indicating contrasting adaptation potential under future drought-prone forest conditions.

2. Materials and Methods

2.1. Study Site and Tree Selection

The study was conducted at three sites in a mixed temperate forest northeast of Hechingen (Baden-Württemberg, Germany). Mean annual temperature is 9.2 °C for 1991–2020, which is 1.0 °C higher than for 1961–1990; mean annual precipitation is 806 mm, 8 mm lower than for 1961–1990, with approximately one-third falling between May and July. The sites were selected to ensure close co-occurrence of sessile oak (Quercus petraea) and European beech (Fagus sylvatica) under contrasting local soil water regimes. Although geographically close, the sites differ in topographic position and elevation, resulting in distinct micro-environmental conditions. Site 1 is located on a gentle northwest-facing slope (5°) at 520 m a.s.l., Site 2 on a steeper southwest-facing slope (15–20°) at 550 m a.s.l., and Site 3 near a steep escarpment at 850 m a.s.l. (

Table 1. Site characteristics, stand structure, tree dimensions, minimum tree age in 2025 (based on the oldest measured tree ring), mean tree-ring width, chronology statistics, and tree species composition. Values for tree-level variables are given as mean ± standard deviation, with n = 15 per species and site. AC: first-order autocorrelation on raw tree-ring series, Rbar: mean series correlation, MS: mean sensitivity, and EPS: Expressed Population Signal (Rbar, MS, and EPS on standardized ring-width values).

	Site 1	Site 2	Site 3
Coordinates (WGS84)	48.3691° N, 9.0004° E	48.3634° N, 8.9948° E	48.3690° N, 9.0460° E
Altitude [m a.s.l.]	520	550	850
Slope [°]	5	15–20	40
Basal area [${\mathrm{m}}^2{\mathrm{ha}}^{-1}$]	23.43	39.91	38.86
Density [stems ${\mathrm{ha}}^{-1}$]	313	520	589
Average DBH [cm] (oak/beech)	54.5 ± 10.4/43.1 ± 6.8	47.3 ± 10.8/37.3 ± 8.9	44.6 ± 10.5/42.1 ± 12.2
Average tree height [m] (oak/beech)	29.5 ± 2.7/28.2 ± 2.4	23.4 ± 1.8/20.2 ± 3.8	17.4 ± 3.3/17.1 ± 2.8
Average minimum tree age in 2025 [years] (oak/beech)	104.2 ± 38.7/87.0 ± 3.9	111.9 ± 25.1/101.1 ± 6.4	168.2 ± 30.0/149.4 ± 34.1
Average tree-ring width [mm] (oak/beech)	2.30 ± 0.47/2.29 ± 0.34	1.73 ± 0.30/1.71 ± 0.40	1.09 ± 0.18/1.19 ± 0.21
AC (oak/beech)	0.56 ± 0.16/0.78 ± 0.08	0.41 ± 0.19/0.68 ± 0.14	0.57 ± 0.27/0.63 ± 0.14
Rbar (oak/beech)	0.42/0.38	0.59/0.31	0.49/0.45
MS (oak/beech)	0.14/0.16	0.21/0.14	0.15/0.20
EPS (oak/beech)	0.91/0.90	0.96/0.87	0.93/0.92
Tree species composition [%]	Fagus sylvatica (78.3), Quercus petraea (8.0), Picea abies (8.0), Carpinus betulus (3.6), Tilia sp. (1.4), Pinus sylvestris (0.7)	Fagus sylvatica (56.0), Quercus petraea (21.0), Carpinus betulus (10.0), Tilia sp. (6.0), Acer pseudoplatanus (3.5), Aria torminalis (3.5)	Fagus sylvatica (56.0), Quercus petraea (12.8), Fraxinus excelsior (10.1), Sorbus aucuparia (7.3), Acer pseudoplatanus (7.3), Crataegus sp. (4.6), Acer campestre (1.8)
Soil properties	Vertic Cambisol on marl clay, periglacial cover with cambic horizon, clay-rich polyhedral subsoil, high nutrient	Haplic Vertisol on marl clay, eroded and mixed profile with humus-rich polyhedral horizon, high nutrient availability	Rendzic Leptosol on limestone, shallow, skeletal, humus-rich topsoil, limited P and K availability

). All soils developed from the Middle Jurassic Opalinuston Formation as parent material but differed in depth and physical structure: Sites 1 and 2 are clay-rich marl-derived soils (Vertic Cambisol and Haplic Vertisol, respectively), whereas Site 3 is a shallow, stony limestone soil (Rendzic Leptosol). Although soil drainage was not measured directly, these site characteristics imply contrasting water balances. At Site 1, the gentle northwest-facing slope and relatively deep, clay-rich soil likely reduce surface runoff and lateral drainage while increasing soil-water storage. At Site 2, the steeper southwest-facing slope likely enhances runoff and downslope drainage and receives greater solar exposure, which may increase evaporative demand despite the clay-rich soil. At Site 3, the shallow, stony soil has a low water-storage capacity, and its position near the escarpment may facilitate rapid drainage, reducing the amount of water available to roots between precipitation events. Phosphorus availability may be constrained by elevated calcium content at Site 3. The Hechingen station (approximately 518 m a.s.l.) is representative of Sites 1 and 2 and averages 39 snow-cover days and a frost-free period of 174 days per year (1991–2020). In contrast, the higher-elevation Trochtelfingen station (approximately 800 m a.s.l.), which is more representative of Site 3, averages 63 snow-cover days and a shorter frost-free period of 155 days. Site 3 therefore likely experiences a longer freezing season and more persistent snow cover than Sites 1 and 2 [32]. All sites belong to the “DryTrees” monitoring framework and are described in detail by Mauz et al. [33], which also includes a map showing the location of the Swabian Alb and a satellite map of the sites. Immersive 360° panoramas of the locations are available at https://36o.de/drytrees/ (accessed on 29 July 2026) [34].

At each site, 15 oak and 15 beech trees were selected within 0.2–0.45 ha, yielding a total of 45 individuals per species. Stand structure varied markedly among sites: Site 1 had the lowest basal area and stem density but larger mean tree dimensions, whereas Sites 2 and 3 were denser and characterized by smaller tree sizes (stem DBH ≥ 7 cm). Mean tree age was approximately 100 years at Sites 1 and 2, whereas trees at Site 3 were substantially older, with several individuals approaching 200 years. Tree ages were estimated from increment cores taken from all sampled trees (see Section 2.3). Reported minimum ages likely underestimate true establishment ages because many cores did not reach the pith and sampling was conducted at breast height and not at the stem base. Species composition at all sites was beech-dominated (56%–78%), while oak accounted for 8%–21% of stems; hornbeam (Carpinus betulus) and ash (Fraxinus excelsior) occurred locally.

2.2. Climate Data and Indices

Meteorological data for 1947–2024 were obtained from the German Weather Service (Deutscher Wetterdienst, DWD) station Hechingen (station ID 2074; ~520 m a.s.l.; [32]). Daily temperature and precipitation records were aggregated to monthly and annual time series (Figure 1) for drought-index calculations and climate-growth analyses. Because the three forest sites are close to one another, the station data provide a consistent regional climate signal; however, precipitation at Site 3 may be underestimated because of topographic effects near the escarpment.

Drought intensity was quantified using the Palmer Drought Severity Index (PDSI; [27]) and the self-calibrated PDSI (scPDSI; [28]), both of which range from −4 (extreme drought) to +4 (very wet). Compared with single meteorological variables, these indices integrate water supply and atmospheric demand and are therefore well suited for characterizing cumulative moisture deficits relevant to tree growth [30,35]. Potential evapotranspiration (PET) was calculated with the Python package pyet 1.4 [36] following Oudin [37], using daily mean temperature and latitude. Monthly PET and precipitation were then used to compute PDSI and scPDSI with the R package scPDSI. For cross-checking and broader spatial comparability, we additionally used gridded scPDSI products from Van Der Schrier et al. [29] and Dunn et al. [38] (hereafter ${\mathrm{scPDSI}}_{sch}$; Figure 1).

Figure 1. Annual mean air temperature (Tm), precipitation totals (PP), and annual averages for the PDSI, the scPDSI (including global gridded data from Van Der Schrier et al. [29] and Dunn et al. [38], ${\mathrm{scPDSI}}_{sch}$), as well as the topsoil soil moisture index (${\mathrm{SMI}}_{TS}$) and the bulk-soil moisture index (${\mathrm{SMI}}_{BS}$). The colored dashed lines represent the respective linear trend. The data span from 1950 to 2024 (1951 to 2022 for SMI) and refer to the weather station in Hechingen, ~520 m a.s.l.

Figure 1. Annual mean air temperature (Tm), precipitation totals (PP), and annual averages for the PDSI, the scPDSI (including global gridded data from Van Der Schrier et al. [29] and Dunn et al. [38], ${\mathrm{scPDSI}}_{sch}$), as well as the topsoil soil moisture index (${\mathrm{SMI}}_{TS}$) and the bulk-soil moisture index (${\mathrm{SMI}}_{BS}$). The colored dashed lines represent the respective linear trend. The data span from 1950 to 2024 (1951 to 2022 for SMI) and refer to the weather station in Hechingen, ~520 m a.s.l.

To complement climate-based indices, we extracted Soil Moisture Index (SMI) data for the period 1951–2022 from the German Drought Monitor [31]. This product is available at 4 × 4 km spatial resolution and provides estimates for topsoil moisture (${\mathrm{SMI}}_{TS}$, upper 25 cm) and bulk soil moisture (${\mathrm{SMI}}_{BS}$; Figure 1). Including SMI enables a more direct link between meteorological drought and site-level water availability in the rooting zone. Sites 1 and 2 were located within the same 4 × 4 km model grid cell, whereas Site 3 fell within an adjacent grid cell. Accordingly, climate-growth correlation analyses for Site 3 were based on climate data extracted from its respective grid cell. However, the climate time series of the two grid cells were highly similar, with Pearson correlation coefficients of 0.99 and 0.91 for ${\mathrm{SMI}}_{TS}$ and ${\mathrm{SMI}}_{BS}$, respectively. For clarity, Figure 1 presents only the SMI time series for the grid cell containing Sites 1 and 2. Intercorrelations among all climate variables and drought indices used in this study are provided in

Table 2. Intercorrelation matrix (Pearson) of climate parameters and indices shown in Figure 1. Correlations were calculated for 1951–2022 (n = 72).

	Tm	PP	PDSI	scPDSI	scPDSIsch	SMITS	SMIBS
Tm	1	−0.13	−0.32	−0.34	−0.33	−0.29	−0.21
PP	-	1	0.80	0.62	0.48	0.78	0.61
PDSI	-	-	1	0.90	0.72	0.85	0.87
scPDSI	-	-	-	1	0.81	0.73	0.86
scPDSIsch	-	-	-	-	1	0.69	0.83
SMITS	-	-	-	-	-	1	0.86

. As with precipitation, uncertainty may be somewhat higher at Site 3 because shallow, stony soils and local topographic effects are not fully represented by the gridded product.

2.3. Tree-Ring Data

For each tree, two increment cores (5 mm diameter) were extracted at breast height (1.30 m) from opposite directions perpendicular to the slope, using increment borers (Haglöf Sweden AB, Långsele, Sweden). This design was chosen to reduce within-tree variability and strengthen cross-dating reliability. Cores were mounted on wooden supports and surface-prepared with a core microtome to obtain clear annual-ring boundaries [39]. Ring widths were measured using WinDENDRO 2019 (Regent Instruments Inc., Quebec City, QC, Canada). Cross-dating quality and measurement plausibility were checked using xDateR (https://viz.datascience.arizona.edu/xDateR/ (accessed on 29 July 2026)). Cores were retained when 30-year segments reached a minimum correlation of 0.30 with the respective master chronology. If this criterion was met only for part of a core, only the reliable segment was retained. When one core of a tree was not usable, the second core was used alone. Final sample depth and retained series are reported in

Table A1. Overview of sampled tree cores, tree dimensions, chronology start years, and within-tree core correlations.

Site	Tree	Oldest Year	Radius cm	Height m	Crown m2	Kraft Class	Chron. 1 Year	Chron. 2 Year	r
1	Q1	1855	34.2	27	168.01	2	1855	1859	0.618
1	Q2	1849	27.4	27	116.28	2	1970	1950	0.628
1	Q3	1852	33.1	32	135.61	2	1853	1852	0.561
1	Q4	1938	27.2	30.2	107.87	2	1943	1938	0.648
1	Q5	1957	19.4	26.8	44.26	2	1957	-	-
1	Q6	1933	26.1	29.2	79.5	2	1969	1933	0.596
1	Q7	1949	25.8	23.8	75.8	2	1960	1949	0.573
1	Q8	1949	22.6	28.5	78.16	2	1949	-	0.585
1	Q9	1940	25.8	29.4	84.99	2	1940	1946	0.743
1	Q10	1937	25.6	30.7	106.53	2	1948	1937	0.444
1	Q11	1877	40.6	33.8	247.21	2	1911	1877	0.554
1	Q12	1953	27.5	27.9	110.96	2	1953	1960	0.663
1	Q13	1932	23.1	33.6	73.4	2	1942	1932	0.541
1	Q14	1937	29	32	95.4	2	1937	1937	0.698
1	Q15	1953	21.2	30.8	66.1	2	1966	1953	0.474
1	F1	1939	22.4	31.2	103.5	2	1939	1939	0.724
1	F2	1941	22.8	29.6	87.26	2	1952	1941	0.67
1	F3	1935	18.3	27.6	72.61	2	1939	1935	0.675
1	F4	1937	27.4	27.8	130.07	2	1937	1953	0.701
1	F5	1942	17.5	27.8	49.43	2	1944	1942	0.765
1	F6	1941	21	30.1	105.74	2	1941	1941	0.732
1	F7	1942	20.2	26.5	91.57	2	1942	1943	0.712
1	F8	1932	21.3	26.9	83.17	2	1932	1935	0.652
1	F9	1938	27.7	29.6	113.38	2	1945	1938	0.696
1	F10	1941	15	21.3	89.23	3	-	-	-
1	F11	1937	23.7	29.4	91.24	2	1938	1937	0.652
1	F12	1932	21.8	28.7	68.86	2	1932	1938	0.609
1	F13	1939	19.6	27	71.36	2	1939	1939	0.693
1	F14	1931	18.9	27.2	55.39	2	1931	1945	0.64
1	F15	1942	25.3	31.9	81.62	2	1944	1942	0.654
2	Q1	1931	23.7	25.4	54.34	2	1931	1931	0.817
2	Q2	1920	24.4	23.9	64.02	2	1922	1920	0.755
2	Q3	1914	38.8	23.4	54.01	2	1914	1929	0.669
2	Q4	1903	18.7	21.8	25.29	2	1913	1903	0.633
2	Q5	1826	23.1	25.8	162.07	2	1860	1826	0.748
2	Q6	1938	26.6	20.5	69.77	2	1948	1938	0.645
2	Q7	1923	22.1	24	64.39	2	1923	1924	0.865
2	Q8	1914	20.1	24.1	51.76	2	1927	1914	0.823
2	Q9	1914	31.5	26.5	86.86	2	1936	1914	0.769
2	Q10	1921	22.9	24.2	77.35	2	1923	1921	0.755
2	Q11	1926	20.1	21.8	32.75	2	1929	1926	0.55
2	Q12	1928	16.7	23.6	35.41	2	1980	1928	0.621
2	Q13	1919	22.6	19.8	22.76	2	1919	1927	0.774
2	Q14	1917	26.3	22	48.64	2	1917	1924	0.817
2	Q15	1901	20.5	23.5	60.07	2	1901	1918	0.769
2	F1	1934	22	20.8	23.24	2	1950	-	0.412
2	F2	1927	24.8	21.9	69.32	2	-	-	-
2	F3	1927	16.6	23.6	33.11	2	1929	1927	0.668
2	F4	1926	20.1	20.2	57.36	2	1926	1932	0.584
2	F5	1911	19.3	28.9	86.45	2	1911	1914	0.56
2	F6	1919	25.1	25	60.4	2	1931	1919	0.623
2	F7	1928	14.6	17.3	40.06	3	1943	1950	0.48
2	F8	1922	22.2	14.2	23.27	2	1970	1920	0.412
2	F9	1916	17.2	21	73.91	2	1918	1916	0.628
2	F10	1918	23.4	18.9	45.46	2	1925	1918	0.583
2	F11	1916	22.6	21.2	65.88	2	1945	1916	0.578
2	F12	1929	12.5	18.3	52.44	2	1929	1933	0.568
2	F13	1925	13	17.8	25.72	2	1930	1925	0.645
2	F14	1929	12.4	15.4	25.94	3	1929	-	0.347
2	F15	1923	13.8	18.5	14.34	2	1930	1990	0.314
3	Q1	1850	18.3	14	36.65	2	1853	1850	0.746
3	Q2	1862	38.3	22.3	86.89	2	1876	1862	0.867
3	Q3	1854	23.6	13.8	22.79	2	1854	1881	0.742
3	Q4	1846	24.7	19.1	63.88	2	1848	1846	0.751
3	Q5	1841	19.7	20.8	29.55	2	-	1844	0.582
3	Q6 neu	1906	17	14.8	29.76	2	1911	1906	0.764
3	Q7 neu	1857	18.5	21.4	38.66	2	1920	1920	0.785
3	Q8 neu	1881	22.8	17.3	41.33	2	1935	1881	0.691
3	Q9	1901	15.9	17	23.98	2	1950	1965	0.64
3	Q10	1830	23.9	14.2	29.88	2	1878	1905	0.705
3	Q11	1794	17.1	12.2	25.81	2	1980	1980	0.623
3	Q12	1841	22.9	22.9	39.84	2	1841	1848	0.78
3	Q13	1908	23.7	17.3	56.18	2	1908	1911	0.642
3	Q14	1841	25.3	14.6	58.5	2	1841	1843	0.689
3	Q15	1840	22.8	19.5	37.5	2	1840	1890	0.739
3	F1	1880	20.1	15.7	65.41	2	1906	1880	0.666
3	F2	1908	13.4	15.7	14.14	3	1950	1965	0.639
3	F3	1910	16.6	18.4	36.77	2	1910	1920	0.725
3	F4	1903	14.8	14.5	35.47	3	1903	1904	0.683
3	F5	1904	13	15.7	33.97	3	1957	1904	0.561
3	F6	1880	22	22.1	78.99	2	1880	1900	0.52
3	F7	1847	31.4	20.2	106.97	2	-	1854	0.626
3	F8	1854	19.8	18.5	50.8	2	1935	1935	0.617
3	F9	1829	20.1	15.8	49.51	2	1875	1857	0.55
3	F10	1826	24	16	46.74	2	1875	1875	0.572
3	F11	1837	24	18	59.52	2	1890	1905	0.674
3	F12	1853	31	16.1	81.16	2	1980	1965	0.697
3	F13	1941	19.7	15.1	49.91	2	1841	1980–2021	0.626
3	F14	1855	31.2	22.6	130.61	2	1855	1859	0.715
3	F15	1907	14.5	12.2	28.27	2	1907	1911	0.75

. For tree-level series construction, the two core measurements per tree were combined using the root mean square (RMS). Raw ring-width series were retained for analyses of growth-response metrics (resistance, recovery, and resilience), whereas detrended series were used for chronology statistics and climate-growth analyses.

To remove age- and size-related trends, ring-width series were detrended in two steps: first with a linear function and then with a spline using a 50% frequency cut-off at 30 years. Detrending was performed with dplPy ([40], dplpy Version 0.1.5), a Python implementation inspired by dplR workflows [41]. Site- and species-specific chronologies (RWI = Ring Width Index) were built from standardized tree-ring series after prewhitening to reduce residual autocorrelation. Variance-stabilized chronologies were generated with a minimum moving-window length of 30 years.

Chronology quality and pointer years were calculated using the interval trend (Gleichläufigkeit) following Schweingruber [42,43]. Years with an interval trend greater than 75% were considered pointer years and were compared with extreme years identified from the drought indices.

2.4. Identification of Extreme Years

Extreme dry and wet years were identified using growing-season (April–September) mean values of scPDSI for 1950–2024 and ${\mathrm{SMI}}_{BS}$ for 1951–2022. To reduce bias caused by pronounced recent trends, especially during the last decade (Figure 1), both index series were linearly detrended before threshold classification.

Growing seasons were classified as extreme dry years when detrended scPDSI or detrended ${\mathrm{SMI}}_{BS}$ values fell below the 5th percentile, and as extreme wet years when values exceeded the 95th percentile. Consecutive extreme years were grouped and treated as a single extreme event (

Table 3. List of extreme dry and wet years. Growing seasons were classified as extreme dry years when detrended scPDSI or detrended ${\mathrm{SMI}}_{BS}$ values fell below the 5th percentile, and as extreme wet years when values exceeded the 95th percentile. SMI: Soil moisture index, scPDSI: self-calibrated Palmer Drought Severity Index.

Extreme Dry Years	Climate Index	Extreme Wet Years	Climate Index
1959	SMI	1965–1967	SMI, scPDSI
1972	SMI	1982	scPDSI
2004	scPDSI	1986–1988	SMI, scPDSI
2017–2019	SMI, scPDSI	1994–1995	SMI, scPDSI

).

2.5. Calculation of Growth-Response Metrics

Growth responses to hydroclimatic extremes were quantified using resistance ($R_{\mathrm{t}}$), recovery ($R_{\mathrm{c}}$), and resilience ($R_{\mathrm{s}}$) following Lloret et al. [26]. These indices characterize the magnitude of growth reduction during stress events and subsequent post-event recovery of radial stem growth. All calculations were based on individual-tree ring-width series.

$$R_{\mathrm{t}}=\mathrm{EY}/\mathrm{PreEY}$$

(1)

$$R_{\mathrm{c}}=\mathrm{PostEY}/\mathrm{EY}$$

(2)

$$R_{\mathrm{s}}=\mathrm{PostEY}/\mathrm{PreEY}$$

(3)

Here, $\mathrm{EY}$ denotes tree-ring width during the event year (or mean tree-ring width during multi-year events), $\mathrm{PreEY}$ denotes mean tree-ring width over the 5-year period preceding the event, and $\mathrm{PostEY}$ denotes mean tree-ring width over the 5-year period following the event.

In addition, we estimated growth recovery time (GRT) and increment loss due to drought (Loss) following Thurm et al. [44]. GRT indicates the number of years required for annual growth to return to pre-drought levels, and Loss quantifies the cumulative difference between observed growth and the pre-drought reference during that period. Loss was retained as a signed metric: negative values indicate growth below the pre-drought reference, with more negative values representing greater cumulative growth loss, whereas positive values indicate growth above the reference. The maximum GRT was set to 10 years, although no observed recovery time reached this cap. GRT was treated as missing when event-period growth exceeded pre-drought growth because no event-related growth reduction occurred. For the 2017–2019 event, GRT was also set to missing when recovery had not occurred by the end of the available series in 2024.

We also calculated average growth reduction (AGR) and average recovery rate (ARR; [21]). AGR was calculated as Loss divided by GRT and retained the sign of Loss; negative AGR values therefore indicate an average annual growth deficit relative to the pre-drought reference. ARR was calculated as $\left(1-R_{\mathrm{t}}\right)/\mathrm{GRT}$, multiplied by 100, and reported as a percentage. Negative ARR values occur when $R_{\mathrm{t}}$ > 1, indicating that event-period growth exceeded pre-drought growth and that no event-related growth reduction occurred. AGR quantifies the mean annual impact of drought, whereas ARR reflects the proportion of the event-period growth reduction recovered per year.

Resistance ($R_{\mathrm{t}}$), recovery ($R_{\mathrm{c}}$), and resilience ($R_{\mathrm{s}}$) were also determined for extreme wet years using analogous event-year, pre-event, and post-event growth windows. In contrast to drought conditions, wet years are generally not expected to cause growth reductions. Nevertheless, we retained the established terminology to maintain consistency, while acknowledging that the ecological meaning of these metrics differs between dry and wet extremes. These results are provided in Appendix B Figure A1 and Figure A2.

2.6. Statistical Analysis

Relationships between tree growth and climate variables were assessed using Pearson’s correlation coefficient. Correlations were calculated between detrended tree-ring chronologies for each species and site and monthly, growing-season, or annual climate variables. Statistical significance was evaluated at p < 0.05. Differences in tree growth responses between species and among sites were examined separately for each drought event. Because the response variables did not consistently satisfy the assumptions of parametric analysis of variance, aligned rank transform (ART) factorial analyses of variance were used [45]. A separate model was fitted for each response variable and drought event, with species, site, and the species × site interaction included as fixed effects. Significant interactions were followed by ART-C simple-effects contrasts. Species were compared within each site, with Holm adjustment across the three sites, whereas sites were compared pairwise within each species, with Holm adjustment applied separately for oak and beech. When the interaction was not significant, significant marginal effects of species and site were interpreted, and pairwise site comparisons were Holm-adjusted. Results are reported as medians and interquartile ranges. Aligned rank transform (ART) omnibus results for site, species, and their interaction, including Benjamini–Hochberg-adjusted p-values across events, are provided in Appendix A 

Table A2. Aligned rank transform (ART) omnibus tests for the effects of site, species, and their interaction on average growth reduction (AGR), average recovery rate (ARR), growth recovery time (GRT), increment loss due to drought (Loss), recovery ($R_{\mathrm{c}}$), resilience ($R_{\mathrm{s}}$), and resistance ($R_{\mathrm{t}}$) for each drought event. Effect sizes are partial ${\eta }^2$ values based on ranks. $p_{\mathrm{BH}}$ values were adjusted across events within each metric and effect using the Benjamini–Hochberg procedure. Significance codes: . p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.

Metric	Event	n	Effect	ART Test	p	Sig.	Partial ${\mathit{\eta }}_{\mathbf{rank}}^2$	${\mathbf{p}}_{\mathbf{BH}}$
AGR	1959	71	Site	${\mathrm{F}}_{2,65}$ = 1.86	0.164	–	0.054	0.218
AGR	1959	71	Species	${\mathrm{F}}_{1,65}$ = 0.21	0.652	–	0.003	0.652
AGR	1959	71	Species × Site	${\mathrm{F}}_{2,65}$ = 0.73	0.487	–	0.022	0.487
AGR	1972	85	Site	${\mathrm{F}}_{2,79}$ = 6.28	0.003	**	0.137	0.012
AGR	1972	85	Species	${\mathrm{F}}_{1,79}$ = 7.11	0.009	**	0.083	0.019
AGR	1972	85	Species × Site	${\mathrm{F}}_{2,79}$ = 1.02	0.364	–	0.025	0.485
AGR	2004	83	Site	${\mathrm{F}}_{2,77}$ = 1.21	0.304	–	0.030	0.304
AGR	2004	83	Species	${\mathrm{F}}_{1,77}$ = 0.90	0.345	–	0.012	0.460
AGR	2004	83	Species × Site	${\mathrm{F}}_{2,77}$ = 3.18	0.047	*	0.076	0.188
AGR	2017–2019	59	Site	${\mathrm{F}}_{2,53}$ = 4.73	0.013	*	0.151	0.026
AGR	2017–2019	59	Species	${\mathrm{F}}_{1,53}$ = 7.43	0.009	**	0.123	0.019
AGR	2017–2019	59	Species × Site	${\mathrm{F}}_{2,53}$ = 1.20	0.31	–	0.043	0.485
ARR	1959	71	Site	${\mathrm{F}}_{2,65}$ = 3.75	0.029	*	0.103	0.029
ARR	1959	71	Species	${\mathrm{F}}_{1,65}$ = 3.69	0.059	.	0.054	0.237
ARR	1959	71	Species × Site	${\mathrm{F}}_{2,65}$ = 3.48	0.037	*	0.097	0.049
ARR	1972	85	Site	${\mathrm{F}}_{2,79}$ = 10.00	<0.001	***	0.202	0.001
ARR	1972	85	Species	${\mathrm{F}}_{1,79}$ = 0.77	0.384	–	0.010	0.384
ARR	1972	85	Species × Site	${\mathrm{F}}_{2,79}$ = 7.15	0.001	**	0.153	0.006
ARR	2004	83	Site	${\mathrm{F}}_{2,77}$ = 7.42	0.001	**	0.162	0.002
ARR	2004	83	Species	${\mathrm{F}}_{1,77}$ = 0.87	0.355	–	0.011	0.384
ARR	2004	83	Species × Site	${\mathrm{F}}_{2,77}$ = 3.44	0.037	*	0.082	0.049
ARR	2017–2019	59	Site	${\mathrm{F}}_{2,53}$ = 4.03	0.023	*	0.132	0.029
ARR	2017–2019	59	Species	${\mathrm{F}}_{1,53}$ = 1.49	0.228	–	0.027	0.384
ARR	2017–2019	59	Species × Site	${\mathrm{F}}_{2,53}$ = 1.10	0.339	–	0.040	0.339
GRT	1959	71	Site	${\mathrm{F}}_{2,65}$ = 3.66	0.031	*	0.101	0.042
GRT	1959	71	Species	${\mathrm{F}}_{1,65}$ = 1.20	0.277	–	0.018	0.386
GRT	1959	71	Species × Site	${\mathrm{F}}_{2,65}$ = 0.52	0.595	–	0.016	0.595
GRT	1972	85	Site	${\mathrm{F}}_{2,79}$ = 6.81	0.002	**	0.147	0.004
GRT	1972	85	Species	${\mathrm{F}}_{1,79}$ = 24.51	<0.001	***	0.237	<0.001
GRT	1972	85	Species × Site	${\mathrm{F}}_{2,79}$ = 10.17	<0.001	***	0.205	<0.001
GRT	2004	83	Site	${\mathrm{F}}_{2,77}$ = 10.37	<0.001	***	0.212	<0.001
GRT	2004	83	Species	${\mathrm{F}}_{1,77}$ = 0.23	0.634	–	0.003	0.634
GRT	2004	83	Species × Site	${\mathrm{F}}_{2,77}$ = 0.59	0.556	–	0.015	0.595
GRT	2017–2019	59	Site	${\mathrm{F}}_{2,53}$ = 1.24	0.297	–	0.045	0.297
GRT	2017–2019	59	Species	${\mathrm{F}}_{1,53}$ = 1.14	0.29	–	0.021	0.386
GRT	2017–2019	59	Species × Site	${\mathrm{F}}_{2,53}$ = 0.96	0.388	–	0.035	0.595
Loss	1959	71	Site	${\mathrm{F}}_{2,65}$ = 1.43	0.246	–	0.042	0.271
Loss	1959	71	Species	${\mathrm{F}}_{1,65}$ = 0.01	0.929	–	0.000	0.929
Loss	1959	71	Species × Site	${\mathrm{F}}_{2,65}$ = 0.35	0.709	–	0.011	0.709
Loss	1972	85	Site	${\mathrm{F}}_{2,79}$ = 6.45	0.003	**	0.140	0.010
Loss	1972	85	Species	${\mathrm{F}}_{1,79}$ = 14.92	<0.001	***	0.159	0.001
Loss	1972	85	Species × Site	${\mathrm{F}}_{2,79}$ = 5.22	0.007	**	0.117	0.030
Loss	2004	83	Site	${\mathrm{F}}_{2,77}$ = 1.33	0.271	–	0.033	0.271
Loss	2004	83	Species	${\mathrm{F}}_{1,77}$ = 1.27	0.264	–	0.016	0.352
Loss	2004	83	Species × Site	${\mathrm{F}}_{2,77}$ = 1.97	0.146	–	0.049	0.292
Loss	2017–2019	59	Site	${\mathrm{F}}_{2,53}$ = 4.35	0.018	*	0.141	0.036
Loss	2017–2019	59	Species	${\mathrm{F}}_{1,53}$ = 3.16	0.081	.	0.056	0.162
Loss	2017–2019	59	Species × Site	${\mathrm{F}}_{2,53}$ = 0.83	0.442	–	0.030	0.589
$R_{\mathrm{c}}$	1959	85	Site	${\mathrm{F}}_{2,79}$ = 5.17	0.008	**	0.116	0.015
$R_{\mathrm{c}}$	1959	85	Species	${\mathrm{F}}_{1,79}$ = 3.24	0.076	.	0.039	0.101
$R_{\mathrm{c}}$	1959	85	Species × Site	${\mathrm{F}}_{2,79}$ = 3.76	0.027	*	0.087	0.029
$R_{\mathrm{c}}$	1972	90	Site	${\mathrm{F}}_{2,84}$ = 2.20	0.117	–	0.050	0.121
$R_{\mathrm{c}}$	1972	90	Species	${\mathrm{F}}_{1,84}$ = 3.56	0.063	.	0.041	0.101
$R_{\mathrm{c}}$	1972	90	Species × Site	${\mathrm{F}}_{2,84}$ = 9.61	<0.001	***	0.186	0.001
$R_{\mathrm{c}}$	2004	90	Site	${\mathrm{F}}_{2,84}$ = 10.62	<0.001	***	0.202	<0.001
$R_{\mathrm{c}}$	2004	90	Species	${\mathrm{F}}_{1,84}$ = 9.75	0.002	**	0.104	0.010
$R_{\mathrm{c}}$	2004	90	Species × Site	${\mathrm{F}}_{2,84}$ = 8.29	<0.001	***	0.165	0.001
$R_{\mathrm{c}}$	2017–2019	90	Site	${\mathrm{F}}_{2,84}$ = 2.17	0.121	–	0.049	0.121
$R_{\mathrm{c}}$	2017–2019	90	Species	${\mathrm{F}}_{1,84}$ = 0.00	0.966	–	0.000	0.966
$R_{\mathrm{c}}$	2017–2019	90	Species × Site	${\mathrm{F}}_{2,84}$ = 3.69	0.029	*	0.081	0.029
$R_{\mathrm{s}}$	1959	85	Site	${\mathrm{F}}_{2,79}$ = 3.49	0.035	*	0.081	0.035
$R_{\mathrm{s}}$	1959	85	Species	${\mathrm{F}}_{1,79}$ = 1.50	0.225	–	0.019	0.225
$R_{\mathrm{s}}$	1959	85	Species × Site	${\mathrm{F}}_{2,79}$ = 0.18	0.833	–	0.005	0.833
$R_{\mathrm{s}}$	1972	90	Site	${\mathrm{F}}_{2,84}$ = 5.43	0.006	**	0.115	0.012
$R_{\mathrm{s}}$	1972	90	Species	${\mathrm{F}}_{1,84}$ = 1.87	0.175	–	0.022	0.225
$R_{\mathrm{s}}$	1972	90	Species × Site	${\mathrm{F}}_{2,84}$ = 3.92	0.024	*	0.085	0.031
$R_{\mathrm{s}}$	2004	90	Site	${\mathrm{F}}_{2,84}$ = 3.69	0.029	*	0.081	0.035
$R_{\mathrm{s}}$	2004	90	Species	${\mathrm{F}}_{1,84}$ = 4.33	0.041	*	0.049	0.081
$R_{\mathrm{s}}$	2004	90	Species × Site	${\mathrm{F}}_{2,84}$ = 3.94	0.023	*	0.086	0.031
$R_{\mathrm{s}}$	2017–2019	90	Site	${\mathrm{F}}_{2,84}$ = 12.94	<0.001	***	0.236	<0.001
$R_{\mathrm{s}}$	2017–2019	90	Species	${\mathrm{F}}_{1,84}$ = 20.37	<0.001	***	0.195	<0.001
$R_{\mathrm{s}}$	2017–2019	90	Species × Site	${\mathrm{F}}_{2,84}$ = 4.85	0.01	*	0.104	0.031
$R_{\mathrm{t}}$	1959	85	Site	${\mathrm{F}}_{2,79}$ = 8.09	<0.001	***	0.170	0.001
$R_{\mathrm{t}}$	1959	85	Species	${\mathrm{F}}_{1,79}$ = 0.31	0.582	–	0.004	0.582
$R_{\mathrm{t}}$	1959	85	Species × Site	${\mathrm{F}}_{2,79}$ = 1.93	0.151	–	0.047	0.303
$R_{\mathrm{t}}$	1972	90	Site	${\mathrm{F}}_{2,84}$ = 13.93	<0.001	***	0.249	<0.001
$R_{\mathrm{t}}$	1972	90	Species	${\mathrm{F}}_{1,84}$ = 13.30	<0.001	***	0.137	0.001
$R_{\mathrm{t}}$	1972	90	Species × Site	${\mathrm{F}}_{2,84}$ = 0.73	0.483	–	0.017	0.644
$R_{\mathrm{t}}$	2004	90	Site	${\mathrm{F}}_{2,84}$ = 11.14	<0.001	***	0.210	<0.001
$R_{\mathrm{t}}$	2004	90	Species	${\mathrm{F}}_{1,84}$ = 6.31	0.014	*	0.070	0.019
$R_{\mathrm{t}}$	2004	90	Species × Site	${\mathrm{F}}_{2,84}$ = 7.08	0.001	**	0.144	0.006
$R_{\mathrm{t}}$	2017–2019	90	Site	${\mathrm{F}}_{2,84}$ = 13.04	<0.001	***	0.237	<0.001
$R_{\mathrm{t}}$	2017–2019	90	Species	${\mathrm{F}}_{1,84}$ = 17.85	<0.001	***	0.175	<0.001
$R_{\mathrm{t}}$	2017–2019	90	Species × Site	${\mathrm{F}}_{2,84}$ = 0.22	0.802	–	0.005	0.802

. To assess the temporal development of growth responses around drought events, average growth deviations were calculated for the drought year and for the five years preceding and following each event. Growth deviations were expressed relative to the mean tree-ring width of the five years preceding the drought event, which served as the reference period. Uncertainty around mean growth deviations was quantified using 90% bootstrap confidence intervals based on 9999 bootstrap iterations. Confidence intervals were derived separately for each species and site and were used to illustrate the variability and robustness of observed growth responses. Exploratory temporal trends across the four event periods were assessed using linear regressions of site-level median response metrics against event year, separately for each species and site. Because only four event periods were available, these trends were interpreted descriptively; line intensity in the trend figures reflects the statistical significance of the fitted slopes. Aligned rank transform factorial analyses and ART-C contrasts were conducted in R using the ARTool package 0.11.2 [46]. All remaining statistical analyses were conducted in Python 3.10.13 using pandas 2.3.3 [47], NumPy 1.24.0 [48], and SciPy 1.15.3 [49]. Data visualization and figure generation were performed using Plotly 5.22.0 [50].

3. Results

3.1. Tree-Ring Chronologies

Across all sites, individual tree-ring series showed pronounced interannual variability, while site-level mean chronologies displayed coherent common signals (Figure 2). A marked increase in ring width occurred around 1950 (vertical dashed line in Figure 2), most clearly at Sites 2 and 3. Thereafter, growth trajectories diverged among sites: Site 3 showed a long-term decline from approximately 1960 onward in both species, whereas Sites 1 and 2 remained comparatively stable through much of the late twentieth century. Mean annual ring width was highest at Site 1 (oak/beech: 2.30 ± 0.47/2.29 ± 0.34 mm), intermediate at Site 2 (1.73 ± 0.30/1.71 ± 0.40 mm), and lowest at Site 3 (1.09 ± 0.18/1.19 ± 0.21 mm). Differences in stand structure provide relevant context for these patterns. Trees at Site 3 were substantially older (oak/beech: 168.2 ± 30.0/149.4 ± 34.1 years) than trees at Site 1 (104.2 ± 38.7/87.0 ± 3.9 years) and Site 2 (111.9 ± 25.1/101.1 ± 6.4 years), and also had narrower and less variable rings (Table 1). Overall, temporal variability in annual radial growth was similar for both species, although pronounced interspecific differences occurred in 1966 and 1967 at Sites 1 and 2, when oak growth was substantially lower than beech growth. From 2008 onward, radial growth declined across all sites, with the strongest reductions at Site 2 for both species and at Site 1 for beech. In contrast, oak at Site 1 and both species at Site 3 showed weaker recent declines (Figure 2). Detrended tree-ring chronologies for both species across the three study sites were free of long-term growth trends and confirmed these between-site contrasts and species-specific synchronicity (Figure 3). Intercorrelation analysis (

Table 4. Site and species intercorrelations (Pearson) calculated for 1950–2024 (n = 75). Q and F denote oak (Quercus petraea) and beech (Fagus sylvatica), respectively; numerals identify Sites 1–3. Same-species chronologies were positively correlated across all site pairs, while within-site oak-beech correlations were strongest at Sites 2 and 3. n.s.: not significant, *: p < 0.05, **: p < 0.01, ***: p < 0.001.

	Q2	Q3	F1	F2	F3
Q1	0.61 ***	0.52 ***	0.32 **	0.25 *	0.22 n.s.
Q2	1	0.57 ***	0.31 **	0.63 ***	0.45 ***
Q3	-	1	0.46 ***	0.39 ***	0.66 ***
F1	-	-	1	0.54 ***	0.54 ***
F2	-	-	-	1	0.56 ***

) showed strong positive correlations among all site chronologies. For oak, the correlation was highest between Sites 1 and 2 (0.61), followed by Sites 2 and 3 (0.57) and Sites 1 and 3 (0.52). Beech correlations varied little among site pairs (0.54–0.56). Within-site oak-beech correlations were stronger at Sites 2 and 3 than at Site 1.

3.2. Impact of Climate on Growth

Climate-growth correlations were analyzed to identify the seasonal windows and hydroclimatic variables most strongly associated with radial growth of beech and oak at each site (Figure 4). Correlations were computed from July of the previous year to August of the current year and for aggregated periods, including the previous and current growing seasons and the previous and current full years. Across sites and species, integrated drought indicators (scPDSI and ${\mathrm{SMI}}_{BS}$) showed clearer and more frequent significant relationships with annual radial growth than single meteorological variables (precipitation and temperature). In general, correlations with current-year growing-season hydroclimatic conditions were stronger than correlations with previous-year conditions, indicating that growth was primarily controlled by moisture availability during the current vegetation period. Species-specific differences were consistent along the site gradient: beech generally exhibited stronger climate sensitivity than oak, particularly at Site 2. At this intermediate site, both species tracked variability in scPDSI and ${\mathrm{SMI}}_{BS}$ most closely, with multiple significant correlations in growing-season windows. At Site 1, strong relationships were more pronounced for beech than for oak, whereas at Site 3, correlations were present but overall weaker and more variable among months. For precipitation alone, significant correlations were less consistent and mostly site-specific. Temperature showed predominantly negative relationships with growth in July and August of the previous year, especially at Site 3, suggesting enhanced growth limitation under warm conditions when water availability was low. Annual radial growth of oak was consistently negatively correlated with ${\mathrm{SMI}}_{BS}$ in July of the previous year, with a statistically significant relationship at Site 1.

3.3. Growth Responses to Extreme Years

Pointer years were irregularly distributed from 1950 to 2024 and showed no clear long-term trend (Figure 5). Major negative pointer years, detected in at least three of the six site- and species-specific chronologies, occurred in 1962, 1964, 1976, 1983, 1989, 1995, and 2023. Positive pointer years frequently followed negative pointer years, indicating renewed growth after strongly reduced increments. Conversely, the negative pointer years in 1964, 1983, and 1995 followed positive pointer years and may therefore partly represent relative growth declines after exceptionally wide rings.

Correspondence between pointer years and hydroclimatic extremes was event-specific. The extremely dry edaphic conditions in 1959 and 1972 were not associated with negative pointer years. In contrast, the 2004 drought coincided with negative pointer years for oak at Site 2 and beech at Site 3, while the 2017–2019 drought was reflected only in oak at Sites 1 and 2. Exceptionally wet periods, which were generally consistent between SMI and scPDSI, were often associated with positive pointer years, particularly at their onset and largely irrespective of site or species.

Average growth deviations during the five years before and after drought events showed that beech generally responded with stronger growth reductions than oak across the three sites (Figure 6). At Site 2, the magnitude of the growth reduction was comparable between oak and beech. In contrast, oaks at Site 3 did not exhibit an immediate growth reduction during the drought event; however, a pronounced decline in radial growth became apparent over the subsequent five years. Similarly, at Site 2, a strong growth reduction occurred during the five years following the drought event. Although both species initially showed signs of recovery during the first two post-drought years, this was followed by a marked decline in growth.

Resistance, recovery, and resilience exhibited substantial variability among drought events and study sites (Figure 7). Resistance ($R_{\mathrm{t}}$) was lower at S2 than at S1 and S3 in 1959. In 1972, oak was more resistant than beech across sites, and resistance followed the order S3 > S1 > S2. A significant species × site interaction occurred in 2004 (${\mathrm{F}}_{2,84}$ = 7.08, p = 0.001): oak was more resistant than beech at S3, resistance was lowest at S2 within oak, and lower at S3 than at S1 within beech (all $p_{\mathrm{Holm}}$ < 0.001). In 2017–2019, oak was again more resistant than beech, while resistance at S2 was lower than at S1 and S3 (all $p_{\mathrm{Holm}}$ < 0.001).

Recovery ($R_{\mathrm{c}}$) showed significant species × site interactions in all drought events (all p ≤ 0.029). Oak recovered more strongly than beech at S1 in 1959. In 1972, beech recovery exceeded that of oak at S1 and S3, whereas oak recovered more strongly at S2. In 2004, recovery was highest at S2 and S3 in beech and at S2 in oak; beech also recovered more strongly than oak at S3. In 2017–2019, the only significant follow-up difference was greater beech recovery at S3 than at S1.

Resilience ($R_{\mathrm{s}}$) differed among sites in 1959 (p = 0.035), although no pairwise contrast remained significant after Holm adjustment. Significant species × site interactions occurred in 1972, 2004, and 2017–2019 (all p ≤ 0.024). In 1972, beech resilience was lower at S2 than at S1 and S3 and lower than oak resilience at S2. In 2004, oak resilience was lower at S3 than at S1 and S2, without significant species differences. In 2017–2019, beech resilience was highest at S3, oak resilience was lowest at S2, and oak was more resilient than beech at S1.

Together, Figure 7 and Figure 8 show that lower drought resistance was often offset by stronger recovery. Consequently, many tree-level observations and site medians remained close to the resilience-1 curve in 1959, 1972, and 2004 despite significant species and site effects. This compensation weakened during 2017–2019, when low resistance coincided with low recovery, most clearly for beech at Site 2.

GRT and Loss differed most clearly in 1972 (

Table 5. Growth recovery time (GRT), increment loss due to drought (Loss), average growth reduction (AGR), and average recovery rate (ARR) for the selected drought events. Q and F denote oak (Quercus petraea) and beech (Fagus sylvatica), respectively; numerals identify Sites 1–3. Values are medians with interquartile ranges in brackets. Species, site, and species × site effects were evaluated separately for each drought event and response variable using aligned rank transform factorial analyses of variance. For significant interactions, bold values indicate significant ART-C species contrasts within a site, while different uppercase and lowercase letters indicate significant site contrasts within oak and beech, respectively. Follow-up p-values were adjusted using the Holm procedure. Omnibus test results are presented in Table A2.

Drought Event	Q1	F1	Q2	F2	Q3	F3
	GRT [years]
1959	1.0 [1.0–1.8] A	1.5 [1.0–2.0] a	2.0 [1.0–4.0] A	2.0 [2.0–5.5] a	1.0 [1.0–1.5] A	1.5 [1.0–2.5] a
1972	1.0 [1.0–1.0] A	2.5 [2.0–3.0] a	1.0 [1.0–1.0] A	5.0 [3.0–6.0] a	1.0 [1.0–1.8] A	1.0 [1.0–2.5] b
2004	2.0 [1.0–2.5] B	1.0 [1.0–3.0] b	1.5 [1.0–2.0] B	1.0 [1.0–3.0] b	2.5 [2.0–4.0] A	3.0 [2.5–3.5] a
2017–2019	1.0 [1.0–2.5] A	2.0 [1.5–4.0] a	2.0 [1.0–4.0] A	3.0 [1.2–4.0] a	1.0 [1.0–3.5] A	1.0 [1.0–3.0] a
	Loss [mm]
1959	0.00 [−0.16–0.26] A	0.49 [−0.23–0.81] a	−0.04 [−0.52–0.07] A	−0.07 [−1.82–0.43] a	0.20 [0.00–0.33] A	0.08 [−0.26–0.51] a
1972	0.19 [−0.20–0.56] A	−0.71 [−1.17–−0.10] b	−0.13 [−0.36–0.12] A	−1.93 [−3.89–−0.46] b	0.22 [0.08–0.45] A	0.13 [−0.30–0.56] a
2004	0.03 [−0.64–0.33] A	−0.45 [−1.28–0.01] a	−0.66 [−0.83–−0.30] A	−0.67 [−0.85–0.52] a	−0.28 [−0.34–−0.17] A	−0.58 [−1.22–−0.32] a
2017–2019	0.09 [−0.86–0.26] AB	−1.16 [−1.75–−0.40] ab	−1.11 [−2.21–−0.05] B	−2.62 [−3.53–−0.76] b	−0.06 [−0.25–0.00] A	−0.13 [−0.56–0.02] a
	AGR [mm ${\mathbf{year}}^{\mathbf{-}\mathbf{1}}$]
1959	0.01 [−0.14–0.26] A	0.37 [−0.12–0.81] a	−0.04 [−0.19–0.07] A	−0.03 [−0.34–0.30] a	0.20 [0.03–0.33] A	0.05 [−0.09–0.51] a
1972	0.19 [−0.16–0.50] AB	−0.26 [−0.38–−0.05] ab	−0.10 [−0.27–0.12] B	−0.32 [−0.63–−0.15] b	0.22 [0.05–0.45] A	0.13 [−0.12–0.56] a
2004	0.01 [−0.26–0.23] A	−0.29 [−0.56–0.01] a	−0.35 [−0.56–−0.23] B	−0.24 [−0.63–0.52] a	−0.09 [−0.14–−0.07] A	−0.21 [−0.35–−0.13] a
2017–2019	0.09 [−0.28–0.26] AB	−0.53 [−0.53–−0.22] ab	−0.24 [−0.66–−0.05] B	−0.88 [−0.90–−0.52] b	−0.05 [−0.12–−0.00] A	−0.13 [−0.26–0.01] a
	ARR [%]
1959	4.33 [−1.29–6.57] A	−5.08 [−21.34–−0.56] a	3.47 [−0.90–8.42] A	2.59 [0.29–4.44] a	−3.08 [−9.85–0.13] A	−0.16 [−10.09–2.39] a
1972	−0.19 [−6.13–3.60] B	4.48 [2.26–6.80] a	11.45 [5.40–20.04] A	3.55 [1.42–7.17] a	−12.35 [−26.73–−1.11] C	1.73 [−6.84–7.71] a
2004	−0.16 [−3.82–10.47] B	11.36 [3.30–17.16] a	24.99 [18.12–28.33] A	9.24 [1.64–40.13] a	5.70 [1.40–9.41] B	13.86 [9.49–15.38] a
2017–2019	2.15 [−2.52–4.04] B	8.10 [3.64–12.32] b	7.41 [5.65–14.87] A	13.06 [10.79–14.70] a	2.75 [2.11–6.26] B	5.27 [−5.58–10.45] b

). Beech had a longer GRT than oak at S1 and S2, and beech GRT was longer at these sites than at S3 (${\mathrm{F}}_{2,79}$ = 10.17, p < 0.001; all $p_{\mathrm{Holm}}$ ≤ 0.014). In 2004, GRT was longer at S3 than at S1 and S2 (both $p_{\mathrm{Holm}}$ ≤ 0.001). The 1972 interaction for Loss (${\mathrm{F}}_{2,79}$ = 5.22, p = 0.007) reflected more negative values for beech than oak at S1 and more negative beech Loss at S1 and S2 than at S3 (all $p_{\mathrm{Holm}}$ ≤ 0.049). In 2017–2019, Loss was more negative at S2 than at S3 ($p_{\mathrm{Holm}}$ = 0.014).

AGR was higher in oak than beech and lower at S2 than at S3 in both 1972 and 2017–2019 (p = 0.009; $p_{\mathrm{Holm}}$ ≤ 0.011). In 2004, oak AGR was lower at S2 than at S1 and S3 (${\mathrm{F}}_{2,77}$ = 3.18, p = 0.047; both $p_{\mathrm{Holm}}$ = 0.004). ARR showed species × site interactions in 1959, 1972, and 2004 (all p ≤ 0.037): beech was lower than oak at S1 in 1959, higher at S3 in 1972, and lower at S2 in 2004 (all $p_{\mathrm{Holm}}$ ≤ 0.020). Oak ARR followed S2 > S1 > S3 in 1972 and was highest at S2 in 2004; ARR was also higher at S2 than at S1 and S3 in 2017–2019 (both $p_{\mathrm{Holm}}$ = 0.041).

Contrasting temporal trajectories for $R_{\mathrm{t}}$, $R_{\mathrm{c}}$, and $R_{\mathrm{s}}$ across drought years are shown in Figure 9. Resistance ($R_{\mathrm{t}}$) generally declines from 1959 toward recent events, with the strongest negative trends at Site 2 for both species (dashed lines) and at Site 1 for beech (red solid line); Site 3 shows weaker and less consistent declines. Recovery ($R_{\mathrm{c}}$) exhibits the opposite pattern, with increasing values over time at most sites and the steepest positive trends for beech at Sites 2 and 3 (red dashed and dotted lines). Oak also shows increasing $R_{\mathrm{c}}$, but generally with lower slopes than beech. In contrast, resilience ($R_{\mathrm{s}}$) remained comparatively stable around the reference level ($R_{\mathrm{s}}$ = 1), with a slight negative trend for beech at Site 1 (red solid line), whereas beech at Site 3 showed a positive tendency and reached the highest late-period $R_{\mathrm{s}}$ values. Overall, the data indicate a shift from higher resistance in earlier drought years toward lower resistance but stronger post-drought recovery in recent drought years, especially for beech. GRT increased slightly from the 1959 drought event to the 2017–2019 event. For beech at Sites 1 and 2, significant trends indicated increasing growth losses (lower Loss values) and stronger average growth reductions (lower AGR values).

4. Discussion

4.1. Long-Term Growth Patterns and Site Controls

Radial tree stem growth increased sharply after 1945 (Figure 2), particularly at Sites 2 and 3, whereas the increase at Site 1 was less pronounced. This abrupt rise is most likely related to post-war stand thinning than to climate effects, given the severe timber shortage in post-war Germany [51]. Since then, growth at Site 3 has declined steadily, probably reflecting the combined effects of increasing tree age and competition for water on the shallow, rocky soils.

Since 2008, growth has declined strongly across all sites, with the most pronounced reductions in beech (Figure 2). This contrasts with recent observations from Czechia, where most beech and oak trees showed no growth decline. Kašpar et al. [52] reported that negative growth trends in beech were mainly associated with lower elevations, whereas individuals at higher elevations were more likely to show positive trends. The strongest positive trend occurred at approximately 800 m, which is close to the elevation of Site 3. Likewise, approximately 80% of oaks showed positive growth trends, with a slightly stronger tendency at higher elevations. The recent decline observed in our study therefore suggests that local edaphic and stand conditions can override elevation-related expectations.

Overall, growth rates were highest at Site 1 and lowest at Site 3. Site 1 is located at lower foothill elevations and probably offers better water availability than the other sites. This interpretation is consistent with the view that drought impacts in temperate forests are strongly modulated by site-specific water availability [53] and rooting-zone constraints, even under a shared regional climate signal. In contrast, Site 3 combines poorer soil water availability, an elevation approximately 300 m higher than the other sites, lower temperatures, and substantially older trees (approximately 60 years older than at Site 1 and 50 years older than at Site 2). This age difference most likely contributes to reduced ring widths through age-related trends [54].

Same-species growth chronologies were strongly correlated across all site pairs (Table 4). For oak, the strongest relationship occurred between Sites 1 and 2 (0.61), followed by Sites 2 and 3 (0.57) and Sites 1 and 3 (0.52). Beech correlations were nearly uniform, ranging from 0.54 to 0.56. Thus, despite the contrasting elevation, soils, and stand structure of Site 3, annual growth variability retained a strong regional signal, and the differences in synchrony among site pairs were relatively small.

Oak and beech growth responses were highly correlated at Sites 2 (0.63) and 3 (0.66), suggesting that local site constraints can dominate over species identity. In contrast, the lower oak-beech correlation at Site 1 (0.32) points to greater species-specific differentiation under less limiting conditions. At Sites 2 and 3, stronger water limitation and terrain constraints may restrict growth sufficiently to reduce interspecific differences. At Site 1, stand structure may have further amplified oak-beech differences: three oaks located directly along a roadside were substantially older than the remaining trees, resulting in an average age difference of approximately 17 years compared with beech.

Basal area was also substantially lower at Site 1 ($23.43 {\mathrm{m}}^2{\mathrm{ha}}^{-1}$) than at Site 2 ($39.91 {\mathrm{m}}^2{\mathrm{ha}}^{-1}$) and Site 3 ($38.86 {\mathrm{m}}^2{\mathrm{ha}}^{-1}$), which may have influenced the species-specific growth correlations because tree species differ in their responses to stand density.

In general, trees tend to exhibit enhanced growth and improved drought recovery under reduced stand density. Steckel et al. [55] demonstrated this effect for sessile oak, although the response was less pronounced than that observed for ponderosa pine. Similarly, Plauborg [56] reported that Fagus sylvatica showed a moderate response to thinning, stronger than that of Pinus sylvestris L. and Acer pseudoplatanus L., but weaker than that of Picea abies (L.).

4.2. Climate-Growth Relationships and Drought Indicators

The climate-growth analyses support H1: integrated drought indicators explained annual growth variability more consistently than the single meteorological variables temperature and precipitation. In particular, scPDSI and SMI of the bulk soil captured growth-relevant moisture deficits across seasons and years better than temperature or precipitation alone (Figure 4). This likely reflects the cumulative nature of drought stress, where growth limitation emerges from interactions between atmospheric demand, precipitation deficits, and soil water storage rather than from isolated monthly anomalies. The combined use of climate-based and soil-moisture-based indices therefore appears especially useful for dendroecological assessments in heterogeneous landscapes such as the Swabian Alb.

Consistent with our findings, Mattes [57] showed that Fagus sylvatica is more strongly influenced by water availability than Quercus petraea. In that study, temperature also had a negative effect on growth, whereas our results provide only limited evidence for a direct temperature effect. Similarly, Meyer et al. [20] reported greater moisture sensitivity in beech than in oak, although that analysis focused on Quercus robur L. In Switzerland, Weber et al. [58] found that beech at dry sites was more drought-sensitive than beech at more mesic sites. Since around 1980, however, trees at formerly mesic sites have also shown increasing drought sensitivity, a pattern that is consistent with our results and may reflect shifts in the seasonal distribution of precipitation.

The comparatively strong climate sensitivity at Site 2 may be related to its intermediate position along the site gradient, but the underlying mechanisms cannot be resolved with the available data. One possibility is that trees developed under historically less limiting moisture conditions and now respond strongly to increasing summer drought. This interpretation remains tentative because long-term site-specific measurements of soil moisture, rooting patterns, and tree mortality were unavailable. Moreover, the analysis included only surviving trees, which may introduce survivorship bias and lead to an underestimation of drought effects.

4.3. Species-Specific Drought Responses

Oak showed higher drought resistance than beech in several events, but the magnitude and significance of the species difference depended strongly on site and drought year. Recovery did not consistently favor beech, and species differences in resilience were restricted to particular site-event combinations (Figure 7). Comparable results have been reported from Belgium [18], Catalonia [59], and Switzerland [19], although the oak species differed among studies. Together, these findings indicate that higher resistance is a comparatively robust advantage of oak, whereas the capacity of beech to compensate through post-drought recovery is more dependent on local conditions.

Beech required more time to recover than oak only at Sites 1 and 2 following the 1972 drought, whereas recovery time did not differ consistently between species in the other events (Table 5). This partly agrees with González De Andrés et al. [59], who reported recovery times of 0.5–1.5 years for Quercus pubescens and 1–3 years for beech, compared with 1–2.5 years for Quercus petraea and 1–5 years for beech in our data. Greater drought-related growth loss in beech was likewise confined to a specific site-event combination rather than representing a general species difference.

The distinction between growth resistance and mortality risk is also evident in North American forests, where other species of Quercus and Fagus occur. Oaks (Quercus spp.) are often considered relatively drought resistant, but this classification does not necessarily imply a lower risk of mortality. In their review, Coble et al. [60] highlighted empirical observations from North America showing that species regarded as drought tolerant may, under certain conditions, exhibit higher mortality than more drought-sensitive species such as Fagus grandifolia. One possible explanation is that drought-sensitive species employ more conservative water-use strategies, including tighter stomatal regulation and earlier leaf shedding. Although these responses reduce carbon assimilation and therefore limit growth during drought, they may prevent critical declines in plant water potential and thereby reduce the risks of hydraulic failure and mortality.

4.4. Site Effects, Temporal Trends, and Recovery Dynamics

Site effects and species × site interactions were at least as pronounced as overall species differences (Figure 8). Site 2 repeatedly showed low resistance and strong growth reductions, which may reflect an intermediate exposure: trees there appear to receive less water than those at Site 1 but may be less acclimated to chronic drought than trees at Site 3. However, the variation among events shows that drought response did not follow a simple linear water-availability gradient and was probably also shaped by stand structure, rooting conditions, and drought history.

Resistance generally declined toward recent drought events, while recovery tended to increase and resilience remained comparatively stable, with contrasting trajectories among sites and species (Figure 9). Vanhellemont et al. [18] also observed declining resistance over time, although the decline in their study was slightly stronger in oak. Although the present data do not indicate a general decline in resilience, low resilience in particular site-species combinations may still indicate increasing mortality risk because reduced resilience has been linked to subsequent tree mortality [61].

During the drought years 1959, 1972, and 2004, many resilience values remained close to 1 (dashed curve in Figure 8) because lower resistance was often compensated by higher recovery, although significant site and interaction effects were present. During 2017–2019, this compensation was less consistent, particularly for beech at Site 2, but resilience responses remained heterogeneous among sites rather than declining uniformly. Consecutive drought years after 2019 likely restricted the opportunity for recovery in the most affected trees. Under such conditions, the resistance-recovery framework reaches a methodological limit because recovery can only be assessed meaningfully when favorable post-drought conditions occur.

4.5. Additional Drivers and Methodological Considerations

Mast years, in which trees produce an above-average number of fruits and seeds, can alter carbon allocation and reduce radial growth. Mast years may also coincide with drought events, potentially strengthening growth reductions and delaying recovery [62,63,64]. We compared our data with records from the MASTREE+ database [65] for southern Germany but found no correlation with the drought events or pointer years identified for these species. Mast years therefore do not appear to explain the main patterns observed here. Biotic disturbances such as pests and pathogens may also influence tree growth patterns and cause deviations in radial increment. However, detailed records of pest or disease occurrence were unavailable, so their effects on the observed growth patterns could not be explicitly tested. To our knowledge, no wildfires occurred at any of the study sites during the study period, making fire-related disturbances unlikely. Air pollution and acid deposition associated with the forest-decline (“Waldsterben”) phenomenon of the 1970s and 1980s may also have affected growth, particularly in conifers. For silver fir, Elling et al. [66] identified ${\mathrm{SO}}_2$ pollution as an important driver of decline and subsequent recovery in interaction with climatic and biotic factors. Because the period of elevated air pollution overlapped with several dry periods, its independent effect cannot be isolated in our data.

Competition for water may have additionally influenced drought responses. A review and meta-analysis by Castagneri et al. [67] showed that competition affects tree growth resistance to drought, although the direction of this effect is not uniform. Hein and Dhôte [68] found that basal area increment of oak (Quercus sp.) decrease with an increase in stand density, and for Quercus petraea, Schmitt et al. [69] found that decreasing stand density favored resistance, recovery, and resilience to extreme drought. The tendency toward lower resistance at Site 2 may therefore be partly explained by higher stem density compared with Site 1. However, Site 3 had even higher stem density and generally higher resistance than Site 2 (except for beech in 2004), indicating that stand density alone cannot explain the site differences.

The response to extreme wet years was evaluated only in Appendix B because these events were not expected to impose stress through waterlogging or oxygen limitation at the study sites. The drought-response metrics therefore require an inverse interpretation under humid conditions: high resistance indicates that event-year growth exceeded mean pre-event growth, while low recovery indicates that growth after the humid event was lower than during the event itself. Overall, wet-year responses showed less pronounced species separation than drought responses, and many medians remained close to the reference value of 1 (Appendix B Figure A1 and Figure A2). The clearest site-related pattern was increasing resistance at the drier or more limiting Sites 2 and 3, suggesting that humid years temporarily enhanced growth where water availability usually constrains radial increment. Recovery tended to decline over time across most site-species combinations, particularly at Sites 2 and 3, because post-event growth was lower relative to the high growth during humid years. Resilience remained comparatively stable overall, although beech at Site 3 showed relatively high late-period resilience. Thus, extreme wet years mainly appear to represent short-term growth releases rather than stressful events, and their interpretation should remain distinct from drought resistance, recovery, and resilience.

The event-based results also depend on how drought years are selected. For example, Merlin et al. [70] selected drought events for sessile oak and Scots pine based on the Standardized Precipitation Evapotranspiration Index (SPEI). We also tested SPEI-based classifications, but they did not identify the same drought years as our scPDSI- and SMI-based approach (except for 1959), nor could they be consistently related to growth responses. Similarly, although 1976 and 2003 are widely recognized as drought years in Central Europe, they did not fall below the 5th percentile in our study area for either scPDSI or ${\mathrm{SMI}}_{BS}$. In addition, our event classification was based on growing-season averages and did not explicitly account for drought timing within the growing season. Early-season and late-season droughts may affect cambial activity, carbon allocation, and recovery differently, and this temporal aspect should be considered in future analyses.

5. Conclusions

This study demonstrates that drought responses of oak and beech on the Swabian Alb are shaped by the interaction between tree species and local site conditions. Growth was highest at the lower site with good soil water availability and lowest at the higher, shallow-soil site, where older tree age and limited soil water storage likely reinforced the observed long-term growth decline. Integrated drought indicators, particularly scPDSI and SMI, explained growth variability more consistently than air temperature or precipitation alone. This supports H1, which predicted stronger climate-growth relationships for integrated drought indicators than for single meteorological variables, and highlights the importance of cumulative moisture deficits for radial stem growth.

Oak showed a drought-resistance advantage in several events, whereas recovery and resilience depended strongly on the combination of species, site, and drought year. H2 was therefore partly supported because beech frequently showed stronger growth reductions and lower resistance, but these differences were not consistent across all events and sites. The pronounced variation in recovery and resilience supports H4 and indicates that post-drought compensation cannot be assigned generally to either species. Earlier drought events were often followed by compensatory recovery, whereas the 2017–2019 drought period showed that repeated droughts and shortened recovery windows can weaken this mechanism in vulnerable site-species combinations. H3 was only partially supported because interspecific differences did not increase consistently along the site water-availability gradient. Local stand structure, rooting conditions, and drought history may therefore override simple expectations based on dry-to-moist site conditions.

For forest management, these findings support site-specific adaptation strategies rather than uniform species recommendations. Oak can provide a drought-resistant component on dry, shallow, or otherwise water-limited sites, although the strong site and event effects caution against assuming a consistent advantage under all conditions. Beech can remain an important component of mixed stands where soil water storage is sufficient, but its proportion and competitive status should be carefully managed when facing an increasing drought risk. On mesic and intermediate sites, mixed oak-beech stands remain promising if stand density is regulated to reduce competition for water, maintain crown vitality, and preserve structural diversity.

Overall, drought adaptation in mixed broadleaf forests cannot rely on species choice alone. Effective management should combine drought-tolerant species mixtures with microsite-specific stand regulation that accounts for soil depth, slope position, local moisture availability, and the shortening intervals available for recovery under more frequent drought events.