Introduction

Calcium (Ca) is the most common alkaline element in the Earth’s crust1. Ca is relatively easily weathered as a soluble cation from primary and secondary minerals. When liberated, Ca moves readily into the soil solution, where it might be adsorbed onto the soil cation exchange complex, taken up by plants and microorganisms, or leached through the soil profile.

The biogeochemistry of Ca in forest ecosystems is complex because its availability depends on the interplay between supply from atmospheric deposition, cation exchange, mineral weathering, mineralization of soil organic matter, and losses through leaching and biomass accumulation. Anthropogenic influences on the Ca cycle in forests are mainly manifested by Ca removal through tree harvest and soil acidification2,3,4,5,6. Although rates of acidic deposition substantially declined in North America and Europe7, depletion of soil Ca and subsequent mobilization of aluminum (Al) remain a persistent threat to the functioning and productivity of forest ecosystems, especially in areas underlined by acid-sensitive bedrock8,9,10. Differences in litter Ca concentration due to intrinsic differences in species-specific physiology may also cause profound changes in soil acidity and fertility11. Ca-rich tree species, like most angiosperms, contain higher foliar Ca, require higher soil pH, and have a faster turnover of soil organic matter than gymnosperm species12,13,14.

From a whole-plant perspective, Ca uptake into roots occurs principally by passive movement in the mass flow of soil water driven by the transpiration stream. Thus, tissue Ca supply is often found to be tightly linked to transpiration rates15. In plants, calcium acts as a second messenger in regulating various physiological and metabolic processes. Cytosolic Ca2+ in leaves, which presents a relatively small fraction compared to the total foliar Ca content, controls leaf gas exchange by regulating stomatal guard cell turgor and thus stomatal opening and closing16,17,18,19. Stomatal closure is tightly associated with the early activation of anion channels in guard cells. Activation of these channels has been linked with abscisic acid and an enhancement of Ca2+ in the cytosol20,21. Specifically, the production of non-protein amino acid γ-aminobutyric acid (GABA) is necessary and sufficient to reduce stomatal opening and transpirational water loss, improving water-use efficiency and drought tolerance of plants22. GABA production in plants is upregulated by stresses23, including acidic conditions24. In this respect, root activity in obtaining Ca could be potentially indirectly linked to carbon (C) assimilation in the canopy through concurrent regulation of water and carbon fluxes through stomates25.

The ratio of the minor (13C) to the major (12C) isotopes in plant tissues serves as a valuable natural tracer that can provide important insights into the exchanges of C and water between plants and the atmosphere. The plant tissue ratio of 13C/12C—thereafter referred δ13C (relative deviation from standard)—depends on stomatal conductance to water vapor (gs) and photosynthesis, which varies according to how plants physiologically respond to changes in their environment. δ13C in tree rings is widely used to infer time-integrated temporal changes of intrinsic water-use efficiency (iWUE), which is defined as the ratio of net photosynthesis (A) to gs26,27,28:

$${i{WUE}}=\frac{{A}}{{{g}}_{{{{{{\rm{s}}}}}}}}=\frac{{{c}}_{{{{{{\rm{a}}}}}}}(1-\frac{{c_{{{{{\rm{i}}}}}}}}{{{{{{\rm{c}}}}}}_{{{{{\rm{a}}}}}}})}{1.6}$$
(1)

where ci and ca are the CO2 intercellular and ambient air concentrations, respectively. As can be seen from Eq. 1, an increase in the atmospheric CO2 concentration, ca, will cause a proportional increase in A/gs of leaves, so long as ci/ca remains approximately constant. This means that all else being equal and with stable ci/ca, iWUE of terrestrial vegetation is expected to increase in direct proportion to increasing atmospheric CO2. Evidence from δ13C in tree rings suggests that this situation has been approximately realized26,29,30. However, environmental factors distinct from ca, such as temperature and precipitation31, plant functional type32,33, tree size/age34, nitrogen (N) deposition35,36, acidic deposition37,38, have independent effects on net photosynthesis (A) and stomatal conductance (gs) and, therefore, may modulate the response of iWUE to rising ca39.

Given the important, though often neglected, the physiological role of Ca in plants, its soil availability may affect ecosystem water losses through altered tree growth and/or stomatal conductance. At the Hubbard Brook Experimental Forest, Green et al.40 found that Ca amendment temporarily increased annual evapotranspiration, thus decreasing water runoff from that forest, presumably due to stimulated primary production. Other ecosystem experiments41,42 observed acidification-induced soil Ca depletion, accompanied by transpiration increases. Higher transpiration could increase the upward movement of dissolved solutes in the xylem and thus enhance Ca delivery from the soil solution to the root surface. Whatever the mechanism of increased transpiration is, one intriguing question remains: “Does calcium availability affect the intrinsic water-use efficiency of trees?”.

To answer this question, we sampled wood, foliage, and soil from one broadleaf deciduous and two evergreen conifer species in mixed forests along a gradient of soil acidity and nutrient availability. Our soil chemistry gradient arose from natural environmental effects such as different tree species (one angiosperm Fagus sylvatica and two gymnosperms—Picea abies and Abies alba) combined with different histories of acidic air pollution (Supplementary Table 1).

More specifically, this study seeks to link contemporary ecosystem Ca availability with calculated century-long iWUE changes derived from tree-ring δ13C values to unravel the potential implication of Ca biogeochemistry through effects on tree physiology.

Results

Temporal changes in iWUE

A total of 1448 individual tree– -ring derived iWUE estimates reveal a steady increase over the 20th century (Fig. 1). We identified a statistical breakpoint in iWUE in 1991, after which iWUE increased less strongly in European beech, leveled off in Norway spruce and even reversed in Silver fir. These breakpoints coincide with the reversal of sulfur (S) and N deposition and with tree growth recovery across our sites (Supplementary Fig. 1). Due to the contrasting magnitude of iWUE increases amongst individual trees (F74, 1298 = 4.735, p < 0.001), the resulting iWUE change over the study period increased in order: European beech (0.24 ± 0.019 µmol CO2 mol−1 H2O y−1), Silver fir (0.26 ± 0.032 µmol CO2 mol−1 H2O y−1), Norway spruce (0.36 ± 0.021 µmol CO2 mol−1 H2O y−1). To address the influence of atmospheric CO2 concentration (ca), climate (precipitation, air temperature, and dryness over the vegetation period), tree growth, and air pollution (S and N deposition) on tree iWUE, we used linear mixed effect (LME) models (see “Methods” for details). A significant positive relationship was indicated between iWUE and environmental factors (ca, S deposition, tree-ring width index, air temperature, PDSI) and a negative relationship with precipitation (Table 1). LME model results explained 67% of the variance in iWUE (conditional R2), with 34% of the variance explained by the environmental factors (marginal R2). Marginal R2 substantially increases when tree species and sites enter the LME analysis as fixed effects. Norway spruce had significantly higher iWUE by 9.4 ± 2.5 µmol CO2 mol−1 H2O and Silver fir by 9.9 ± 2.4 µmol CO2 mol−1 H2O compared to European beech (Table 1 and Supplementary Fig. 2a). Furthermore, site-specific conditions significantly modulated individual iWUE (Table 1 and Supplementary Fig. 2b), iWUE being lower at less polluted (Novohradské hory) and colder (Orlické hory, Jizerské hory) sites. Complementary to the LME analysis, hierarchical partitioning (HP) indicated that, of the 59% explained variance by the “full” (including species and sites) LME model, ca account for 43.9%, followed by species (21.3%), dryness index (11.1%), site (10.8%), and S deposition (10.3%).

Fig. 1: Species-specific intrinsic water-use efficiency (iWUE) trends.
figure 1

Species-specific tree-ring derived iWUE with smoothed (generalized additive model) lines and confidence intervals (α = 0.95). Dashed lines indicate significant breakpoints in the data, and point color refers to the estimated sulfur (S) deposition across sites.

Full size image
Table 1 Linear mixed effect (LME) model and hierarchical partitioning (HP) results.
Full size table

The linkages across iWUE, diWUE/dc a, and ecosystem nutrient biogeochemistry

Due to the high degree of inter-correlated quantitative variables representing soil and foliar chemistry, climate, mean iWUE, and diWUE/dca (the mean iWUE change with respect to ca) across our sites and species, we employed principal component analysis (PCA) to identify directions along which the variation in the data is maximal. The first principal direction (Dim 1) represents 46.1% of the variances in our data set. It is associated with variables connecting the acid–base status of soils, foliar Ca content, and diWUE/dca (Supplementary Fig. 3). The second principal direction (Dim 2) represents an additional 25.3% of the variance. It is composed of variables connecting mean iWUE, soil, and foliage C–N–P stoichiometry, and climate (Supplementary Fig. 3). Thus, diWUE/dca and mean iWUE are highly independent of each other and associated with distinct ecosystem properties.

The PCA results were consistent with LME model analysis (Table 1), showing that tree iWUE in broadleaved European beech (high foliar N/P and low C/N, Supplementary Fig. 4) was lower compared to both conifer species. Concurrently, sites with increasing disbalance between soil N and available P (high soil N/P and C/P, such as Orlické hory, Novohradské hory, and Jizerské hory, Supplementary Fig. 3) are those associated with lower mean iWUE, again consistent with LME model results (Table 1).

In the iWUE trend, we identified a breakpoint in 1991, which coincided with the reversal of acidic air pollution in central Europe (Supplementary Figure 1). Consequently, the rate of response of iWUE to CO2 after 1991 weakened or diminished, notably in both conifer species (Fig. 1). Therefore, the resulting diWUE/dca over the entire period reflected an apparent rate change of iWUE within the last three decades. The tight association among soil total exchangeable acidity (TEA), Ca availability, and diWUE/dca across our environmental gradient, which emerged from PCA, we further demonstrated by regression analysis. In the combined dataset, soil available Ca explained 69% of the variability in diWUE/dca (Fig. 2a), with individual relationships further significant for Norway spruce (R2 = 0.84, p = 0.028, n = 5 sites) and Silver fir (R2 = 0.85, p = 0.026, n = 5 sites). Concurrently, soil acidity was significantly inversely related to foliar Ca concentrations, and increasing soil acidity was associated with increasing diWUE/dca (Fig. 2b).

Fig. 2: iWUE response to an increase in ambient CO2 (diWUE/dca) related to ecosystem calcium availability.
figure 2

Simple linear regression with confidence intervals (α = 0.95) between soil available Ca concentration and diWUE/dca (iWUE response to an increase in ambient CO2, (a) and between soil total exchangeable acidity (TEA) and foliar Ca concentration (b). Regression coefficients and significance levels in black refer to the combined dataset. Different colors recognize individual species-specific regression results across sites in (a), and the point color in the b refers to the diWUE/dca values. Gray lines indicate standard errors of mean values.

Full size image

Discussion

Our data revealed a complex interaction of rising atmospheric CO2 concentration, climate, and acidic atmospheric deposition on iWUE, consistent with other studies32,37,38,43,44. Our analysis emphasizes the pivotal role of sulfur deposition in driving non-linear trends in iWUE of central European temperate forests and provides only limited supporting evidence of the stimulation effect of nitrogen deposition on iWUE35,36,45. The non-linear temporal increase of iWUE reflected the strong influence of acidic air pollution on stomatal conductance, especially in both conifers. Increases in precipitation acidity likely decreased gs18, as tree growth was also suppressed, leading to increased iWUE during the “acid rain” period, as has been consistently shown in areas affected by acidic air pollution worldwide29,37,38,44,46,47. Given the high sensitivity of Silver fir iWUE48 and stem growth49,50 to acidic air pollution, dendrochronological reconstructions may suffer from substantial species-dependent loss of climate sensitivity51,52,53 even in areas far from industrial pollution sources. Although our study was not designed to fully resolve the effect of climatic factors on tree physiology, on warmer sites with lower precipitation, the mean iWUE was higher, but over time, iWUE responded positively to PDSI, suggesting conditions favoring higher productivity (A) in wetter years. Furthermore, the design of our study includes mesic regions where annual potential evapotranspiration does not exceed annual precipitation. Thus, the detection of iWUE-climate interaction may be confounded by other external and endogenous influences. The combination of tree species and site conditions underscores the intricate interplay between environmental drivers on tree physiology. Specific leaf structure and biochemistry underpinned the intrinsic differences in iWUE among angiosperm and gymnosperm species, being significantly higher in Norway spruce and Silver fir than in European beech54,55.

The most intriguing result connects the soil acid-base chemistry with the ratio of changes in iWUE in response to increasing ambient CO2 (diWUE/dca). As acidic air pollution has declined since the late 1980s, iWUE has been adjusting to new levels in compliance with current ecosystem Ca availability. The highest diWUE/dca persisted at locations with high soil acidity and low soil Ca availability. Considering that calcium nutrition has improved in conifers and remains unchanged in beech during recent years56, the observed deacceleration of increasing iWUE in recent decades43 may be partly related to increased Ca availability for tree uptake.

Since the concentration of Ca in the soil solution is controlled by soil chemistry and diffusion along concentration gradients, it is crucial to recognize the non-linearity of the effects of soil acidity on Ca availability6. Ca mobilization is reduced at low soil pH (<4.0), and Al becomes the dominant cation in the soil solution. Concurrently, Ca availability for tree uptake decreases because Al interferes with Ca uptake and root growth25. We hypothesize that Ca soil depletion and Al mobilization due to soil acidification pose significant stress on trees, which upregulates tree water-use efficiency. It has been shown that GABA (γ-aminobutyric acid) is produced under stress conditions23, including acidic conditions24. Our study suggests that the role of the Ca availability, which is constrained by high Al and hydrogen concentrations in acidic soils, appears to be that of fine-tuning stomatal aperture, i.e., capable of adjusting iWUE responses under increasing atmospheric CO2 concentration.

As Ca transport through tissues has been shown to follow apoplastic pathways, tissue Ca supply is often closely linked to transpiration15. It is, therefore, likely that higher leaf hydraulic conductance in angiosperms enables higher stomatal conductance and, thus, lower iWUE compared to gymnosperms57. In the Fernow Experimental Forest (FEF) acidification experiment, Lanning et al.41 observed an increase in ecosystem evapotranspiration following 25 years of ammonium sulfate addition, which partly contradicts our observations. Authors attributed the increase in vegetation water use to Ca scarcity following soil acidification. Simultaneously, increased aboveground C storage of treated catchment has also been observed, dominating the ecosystem response to long-term N addition58. Next, soil pH in our catchments was substantially lower (pHH2O = 3.69 ± 0.17) compared to FEF (pH = 4.02–4.12)59. Thus acidity mediated feedback on Ca availability and iWUE might be more pronounced in forests under this study.

Calcium-rich tree species modulate the soil environment so that Ca-rich litterfall promotes higher soil pH, exchangeable Ca, soil base saturation, and forest floor turnover12,60. Thus, distinct calcium cycling among trees may explain the modulation of the relationship between carbon and water fluxes and contribute to community assembly processes. As such, widespread co-occurrence of beech-spruce-fir in temperate Europe61 may arise partly from distinct species-dependent Ca biogeochemistry affecting physiological processes. Given the recent growth and competition advances of beech and fir over spruce62 enhanced water loss through transpiration can be expected63. Future research into the integrated role of Ca in water transport processes at tree and ecosystem levels is needed. Furthermore, S deposition in European and North American forests has changed drastically in the past 50 years, implying that pollution may have significantly influenced tree physiology during this period. It also raises the possibility that some observed iWUE changes were not caused by climate, at least in areas with acidified soils. We argue that a more holistic approach to the impacts of multiple environmental drivers is needed to correctly interpret existing records and accurately predict future changes in forest ecosystem functioning and productivity in response to global climate and environmental change.

Methods

Study sites and environmental parameters

Five mixed forest sites were selected to represent common temperate forest types across a wide range of nutrient availability conditions in Europe. At each site, mature European beech (Fagus sylvatica L.), Norway spruce (Picea abies L.), and Silver fir (Abies alba Mill.) were sampled to represent a gradient in biological properties (angiosperm vs. gymnosperms) and subsequent nutritional requirements (broadleaf deciduous vs. needle leaf evergreens)12,64,65. The forest habitats represent remnants of mixed temperate forest, with uneven-aged structure and non-intensive management. The average age was for beech 110 ± 34 years, spruce 129 ± 28 years, and fir 111 ± 33 years. We further selected individual sites alongside the former acid pollution gradient, with a prerequisite to maintaining a narrow range in elevation and in mean annual air temperature (Supplementary Table 1). Mean growing season precipitation totals range from 320 to 490 mm, with annual precipitation totals between 650 and 1390 mm. Concurrently, annual potential evapotranspiration varied between 505 and 611 mm and is lower than annual precipitation. All sites are underlined by acid-sensitive bedrock (granite, gneiss, mica-schist, sandstone) with a mean CaO rock concentration of 1.0 ± 0.7%. However, in the Beskydy site, CaO concentrations in a mixture of claystone and sandstone (flysch) usually vary between <0.5% and 10% (source: Czech Geological Survey Lithogeochemic database), which creates heterogenetic soil Ca distribution.

Determination of carbon isotope ratio in tree rings

Wood cores were collected from five trees for each species at each site in 2018. One core per tree was extracted using a Pressler borer (Haglof Company Group, Sweden) at breast height (1.3 m). All cores were sampled parallel to the slope to avoid wood compression. All samples were measured using a VIAS TimeTable device with a measuring length of 78 cm and resolution <0.01 mm (©SCIEM, Vienna, Austria). The obtained tree-ring width (TRW) series were visually synchronized, statistically cross-dated, and additionally corrected for missing and false rings using PAST4 (SCIEM, Vienna, Austria) and COFECHA66. To remove non-climatic, age-related growth trends and other non-climatic factors (e.g., competition) from the raw TRW series, we applied cubic smoothing splines with 50% frequency cutoff at 100 years using ARSTAN software67. We used this method to preserve inter-annual to multi-decadal growth variations. TRW indices were calculated as residuals between the measured TRW and the corresponding fitted values after applying an adaptive power transformation to minimize end-effect problems. The indexed stand species-specific chronologies were calculated using bi-weight robust means and used for further analysis.

Bulked five-year segments (starting back from the most recent 2014–2018 segment) of 75 precisely dated wood samples were analyzed for their 13C/12C isotopic ratios with a continuous-flow mass spectrometer ISOPRIME100 (Isoprime, UK) interfaced with a Vario PYRO cube Elemental Analyzer (Elementar Analysensysteme, Germany) (Supplementary Data 1). The finely homogenized wood samples (Retsch MM200 mill) were weighed (approximate weight of ~1.0 mg), enclosed in tin capsules, and subsequently combusted at 960 °C. Before each set of measurements, the mass spectrometer’s ion source was centered and tuned, and tested for stability (standard deviation ≤0.04‰ on ten pulses over three consecutive runs) and linearity (≤0.03‰/nA) over the entire range of expected ion currents obtained from the measurements of test samples. The standard deviation was ≤0.06‰ on five consecutive measurements of the same wood sample. The system was calibrated using certified reference materials with known isotopic ratios from the International Atomic Energy Agency (IAEA) and the United States Geological Survey (USGS). The δ13C wood values (in ‰) were calculated as the deviation from the Vienna Pee Dee Belemnite (VPDB) standard, according to δ13Cwood = [(13C/12Csample/13C/12Cstandard) − 1]*1000.

Calculation of tree-ring iWUE

We calculated iWUEwood from δ13Cwood values based on the known relationship between leaf ci/ca and isotopic carbon discrimination (Δ13C)28,32. To account for photorespiration and post-photosynthetic fractionation effects, we first calculated ci based on:

$${{c}}_{{{{{{\rm{i}}}}}}}=\left(\frac{({\Delta }^{13}\!{\,}{{C}}-{a}+{f} * (\frac{{\Gamma }^{\ast }}{{{{{{\rm{pca}}}}}}}))}{{b}-{a}}\right)\ast {{c}}_{{{{{{\rm{a}}}}}}},$$
(2)

where a (4.4‰) is the fractionation associated with CO2 diffusion through stomata68, f (12‰) is the isotopic fractionation associated with photorespiration69, pca is the partial pressure of atmospheric CO2 in pascals and b (28‰) denotes fractionation associated with Rubisco carboxylation69. ca is the atmospheric CO2 concentration in the year of tree-ring formation, and Δ13C was calculated as follows:

$${\Delta }^{13}\,{{{{{\rm{C}}}}}}=\frac{{{{{{{\rm{\delta }}}}}}}^{13}\,{{{{{{\rm{C}}}}}}}_{{{{{{\rm{atm}}}}}}}-{{{{{{\rm{\delta }}}}}}}^{13}\,{{{{{{\rm{C}}}}}}}_{{{{{{\rm{wood}}}}}}}}{1+\frac{{{{{{{\rm{\delta }}}}}}}^{13}\,{{{{{{\rm{C}}}}}}}_{{{{{{\rm{wood}}}}}}}}{1000}}$$
(3)

Where δ13Catm is the carbon isotopic signature of mean atmospheric CO2 in the years of ring formation. The values of ca and δ13Catm were taken from Belmecheri and Lavergne70 and McCarroll and Loader71. Γ* referred to the Rubisco CO2 compensation point in the absence of mitochondrial respiration (in pascals) and was calculated as72:

$${\Gamma }^{* }={\Gamma }_{25}^{* }* \left(\frac{{{P}}_{{{{{{\rm{atm}}}}}}}}{{{P}}_{0}}\right)* {{{{{{\rm{e}}}}}}}^{\left(\frac{\Delta {{H}}_{{{{{{\rm{a}}}}}}}* (\left({T}+273.15\right)-298.15)}{{R}* \left({T}+273.15\right)* 298.15}\right)},$$
(4)

where \({\Gamma }_{25}^{* }\) (4.332 Pa) is the Rubisco CO2 compensation point at 25 °C73, Patm refers to ambient atmospheric pressure at a given elevation (in pascals), P0 (101325 Pa) is the atmospheric pressure at sea level, R (8.3145 J mol−1 K−1) is the universal gas constant, and T (in °C) is the mean air temperature over the growing season (May–August), and ∆Ha is the activation energy (37,830 J mol−1).

Finally, iWUEwood (in µmol CO2 mol−1 H2O) was calculated as follows:

$${{iWUE}}_{{{{{{\rm{wood}}}}}}}=\left({{c}}_{{{{{{\rm{a}}}}}}}-{{c}}_{{{{{{\rm{i}}}}}}}\right)* 0.625$$
(5)

The 0.625 constant accounts for the different diffusivities of H2O and CO228.

Foliage and soil sampling and analysis

Similarly to wood core retrieval, we randomly selected five trees of each species for foliage sampling at each location (Supplementary Data 2). All foliage samples were collected in August 2018 by tree climbers from the upper third, sun-exposed canopy. Needle samples (restricted to the current year needles) from spruce, fir, and beech leaves were dried at room temperature, homogenized, and analyzed for total organic C and total N by dry combustion with a CNS elemental analyzer FLASH 2000 (Thermo Scientific, USA). After acid digestion of foliage samples, P concentration was analyzed spectrophotometrically, and Ca concentration by flame atomic absorption spectrophotometry (AAnalyst Perkin Elmer 100, Norwalk, USA).

Soil samples were taken in five replicates for each species at each site at two different soil horizons: forest floor (F + H horizon) and upper mineral soil (A + B horizon). This sampling design resulted in 150 soil samples that were taken within the projection of the tree canopy at a distance of approximately two meters from the tree trunk. Soil samples were air-dried, sieved (mesh size five mm for forest floor and two mm for mineral soil), and analyzed for total organic C and N with a CNS elemental analyzer FALSH 2000. Available P (Pavailable) was determined in Mehlich extract by the molybdate blue method74. Soil pH was determined in deionized water (1:5 w/v). Using flame atomic absorption spectrophotometry, exchangeable Ca was analyzed in 0.1 M BaCl2 extracts (1:10 w/v). TEA (TEA = Al3+ + H+) was determined by titration of 0.1 M BaCl2 extracts with 0.025 M NaOH solution (to pH of 7.8). We refer to all soil chemistry data as the arithmetic mean of forest floor and mineral soil (Supplementary Data 2). Tree foliage and soil samples were taken independently from trees subjected to wood coring.

Air pollution and climatic variables

To infer historical S and N depositions across our sites, we employed a statistical method based on the temporal coherence between the measured precipitation chemistry and the respective Central European emission rates of SO2, NOx, and NH3 emissions44,75. Empirically-based interpolation, taking into account key variables such as altitude, precipitation, and geographical coordinates, enabled the representation of spatial variations in precipitation chemistry (Supplementary Data 1).

Climate data covering the period from the early 1910s to 2018 were derived through interpolation from the three most representative weather stations in the vicinity of each sampling area using locally weighted regressions, including the effect of altitude. All observations of weather variables were tested for outliers and broke through a detailed homogenization sequence, and gaps in missing data were filled76. Resulting climate information, i.e., air temperatures, precipitation, and simplified soil water availability indicator (self-calibrated Palmer Drought Severity Index or PDSI77), were used to derive environmental variables representing climatic conditions at sites over the vegetation season (May-August) (Supplementary Data 1).

Statistical analysis

To calculate trends in iWUE since 1914, we fit an LME model with year as a fixed effect and tree ID as a random factor using nlme package78 in R79. We further identified breakpoints in iWUE, deposition, and TRW chronologies using the segmented R package80. We then used LME models to determine which environmental factors, including atmospheric CO2 concentration70, S deposition, N deposition, PDSI, precipitation, air temperature, and standardized TRW, were most affecting long-term iWUE changes. We separately tested LME models with the species and site as fixed factors to assess the importance of local conditions on resulting iWUE trends. Each LME model accounted for temporal autocorrelation, was fit via maximum likelihood, and tree ID represented a random factor. The final LME model, having the lowest corrected Akaike information criterion, was selected using the MuMln R package81. All continuous environmental variables entering LME analysis were standardized to avoid multicollinearity issues. We used hierarchical partitioning (HP) in consecutive analysis to infer the contribution of each predictor to the total explained variance of a final LME model, both independently and in conjunction with the other predictors using hier.part R package82. A Z-score-based estimate of the “importance” of each predictor is provided by using a randomization test. HP analysis allows the identification of the predictors that explain most variance independently of the others, helping to overcome the problems presented by multicollinearity.

Variation in nutritional and chemical properties of foliage and soils (Supplementary Data 3) was assessed using Principal Component Analysis (PCA). Variables entering the PCA comprised foliage and soil C/N, C/P, and N/P ratios, Ca concentration, soil pH and TEA, and mean temperature and precipitation over the vegetation season. We included calculated mean iWUE and diWUE/dca to the environmental variables. R packages factoextra83 were used to summarize and visualize multivariate data with principal components (dimensions). Complementary to our PCA, we used the R package ggstatsplot84 to calculate and visualize the tree-species effect on iWUE. Non-parametric Kruskal–Wallis one-way ANOVA followed by Dunn pairwise comparison of medians was used for data with non-normal distribution. Finally, simple linear regression analysis was employed to accent relationships among diWUE/dca, foliage, and soil chemistry (Supplementary Data 4).