Climate and atmospheric deposition effects on forest water-use efficiency and nitrogen availability across Britain

Abstract

Rising atmospheric CO₂ (c_a) has been shown to increase forest carbon uptake. Yet, whether the c_a-fertilization effect on forests is modulated by changes in sulphur (S_dep) and nitrogen (N_dep) deposition and how N_dep affects ecosystem N availability remains unclear. We explored spatial and temporal (over 30-years) changes in tree-ring δ¹³C-derived intrinsic water-use efficiency (iWUE), δ¹⁸O and δ¹⁵N for four species in twelve forests across climate and atmospheric deposition gradients in Britain. The increase in iWUE was not uniform across sites and species-specific underlying physiological mechanisms reflected the interactions between climate and atmospheric drivers (oak and Scots pine), but also an age effect (Sitka spruce). Most species showed no significant trends for tree-ring δ¹⁵N, suggesting no changes in N availability. Increase in iWUE was mostly associated with increase in temperature and decrease in moisture conditions across the South–North gradient and over 30-years. However, when excluding Sitka spruce (to account for age or stand development effects), variations in iWUE were significantly associated with changes in c_a and S_dep. Our data suggest that overall climate had the prevailing effect on changes in iWUE across the investigated sites. Whereas, detection of N_dep, S_dep and c_a signals was partially confounded by structural changes during stand development.

Introduction

The ability of forests to mitigate climate change depends on how well they cope and adapt to the rapid forecasted changes in atmospheric conditions, including pollutant emissions from anthropogenic activities , namely CO₂, sulphur (S) and reactive nitrogen (N). While sulphur emissions have been successfully regulated in Europe and North America¹, N compounds, particularly ammonia and nitrous oxide, continue at relatively high levels compared to the pre-industrial period^2,3,4. These changes in N and S atmospheric concentrations have occurred together with increasing atmospheric CO₂ concentration (c_a), which recently exceeded 410 ppm⁵, with a relative change of 46% compared to pre-industrial levels. Atmospheric N and S compounds are deposited from the atmosphere onto terrestrial ecosystems and together with the increasing c_a can have an effect on some of the processes underpinning forest carbon, water and nutrient cycling.

Increases in c_a affect leaf gas exchange by increasing photosynthesis (A) and reducing stomatal conductance (g_s)⁶, thus raising intrinsic water use efficiency of trees (iWUE = A/g_s). Higher iWUE has been commonly reported for conifer compared to broadleaf species⁷. Moreover, the increase in iWUE derived from carbon isotope composition (δ¹³C) in tree rings⁸ has generally been associated with an active response of trees to increasing c_a, whereby intercellular CO₂ (c_i) increased in a proportional way to c_a, resulting in a constant c_i/c_a ratio, due to proportional regulation of A and g_s⁹. However, changes in N (N_dep) and S (S_dep) deposition can also influence carbon–water relations. High levels of S_dep in the 1980s have been hypothesised to have led to stomata closure¹⁰, thus current S_dep reductions might promote an increase in g_s¹¹, counteracting the CO₂-induced water-saving effect of stomatal closure, particularly under non-limiting moisture conditions. Conversely, there is considerable evidence that increased N_dep has a fertilization effect on A, thereby contributing to enhancing a c_a or climate driven increase in tree iWUE^12,13 and forest C-sinks^14,15 in N-limited forests.

Atmospheric N represents an additional input of N for trees, particularly for N-limited forest ecosystems in the Northern hemisphere. An increase in N_dep, however, has been associated with an acceleration of N cycling, with increase in N losses from ecosystems (through nitrate leaching and denitrification), when N saturation is reached¹⁶. Stable nitrogen isotope composition in plant materials (δ¹⁵N) has been used as a proxy of changes in ecosystem N availability¹⁷. An increase in tree δ¹⁵N has been observed in studies along N_dep gradients^18,19 and soil N manipulation experiments^20,21. By contrast, a decrease in tree δ¹⁵N is expected in the case of a reduction of soil N availability (e.g., under a reduction of N_dep²² or in case of tree canopy retention of N_dep²³), or when N supply is insufficient to meet the demand caused by CO₂ fertilization effects on A²⁴.

Combining multiple isotopes in tree rings across climate and N_dep gradients gives a powerful tool to advance our understanding of spatial and temporal patterns of iWUE and its main drivers. In particular, stable oxygen isotope composition (δ¹⁸O) provides physiological information in addition to that derived from δ¹³C, as its variation in plant material depends on the δ¹⁸O of the source water and that of the leaf water, the latter affected by transpiration and g_s²⁵. Moreover, including δ¹⁵N allows the detection of changes in ecosystem N availability due to changes in the N input from the atmosphere. Studies assessing the effect of N_dep and its interactions with other pollutants and climatic variables by using a multiple isotope approach are predominantly site²⁶ or species-specific²². Analyses at regional scales, and involving different tree species, are paramount to achieving a better understanding of forest functioning in response to global changes.

We measured δ¹³C_w, δ¹⁸O_w and δ¹⁵N_w in tree rings from 1980 to 2010 for four of the most common species in Britain, i.e., Scots pine (Pinus sylvestris L.), Sitka spruce (Picea sitchensis Bong. Carr.), pedunculate oak (Quercus robur L.) and European beech (Fagus sylvatica L.) in 12 managed forests. Sites were selected along a gradient of climate and N_dep, while ensuring that within species soil type, forest structure, stand age and management remained similar (Table 1, Fig. 1). Specific goals were to: 1—document the temporal changes and spatial differences in the main climate parameters and N_dep and S_dep along these South–North gradient in Britain; 2—explore the temporal trends in iWUE, c_i/c_a and oxygen isotope discrimination, Δ¹⁸O_w (i.e., difference between δ¹⁸O_w and precipitation δ¹⁸O, see “Methods”) at the 12 forests and assess the possible species-specific physiological mechanisms (changes in A and/or g_s) underlying variations in iWUE; 3—evaluate whether sites receiving high N_dep experience increase in ecosystem N availability and N saturation (using δ¹⁵N_w as a proxy); 4—elucidate the drivers of spatial and temporal changes in the isotope-derived physiological and ecological processes.

Table 1 Description of the forest stands included in the study.

Figure 1

Sites included in the study. Map showing forested sites along the precipitation gradient in Britain (panel A) and described in the Table 1. Size of the points reflects the levels of N_dep, which are reported in the panel B for each forest stand and species, together with the S_dep (panel C). Each point is the average (± standard deviation) of annual values (kg N ha⁻¹ year⁻¹ and kg S ha⁻¹ year⁻¹) across 1995–2010 for most of the forest stands, with the exception of Rogate (data from only 2010) and Shobdon and Covert Wood (modelled data, Ref. “Methods”). Note that for Goyt, we considered atmospheric deposition data collected at Ladybower, as the two sites are only 30 km apart. Black solid and dashed lines in the panel B indicate the low and high boundaries for the UK critical load for nitrogen deposition.

Results

Spatio-temporal changes in climate and atmospheric deposition across Britain

Across the investigated sites and during the studied years (1980–2010), mean temperature during the growing season significantly increased by approx. 0.02–0.06 °C year⁻¹. Whereas, vapour pressure deficit (VPD) did not show a significant trend, except for an increase at the South-eastern most site of Covert Wood (Table 2). The standardised precipitation-evapotranspiration index (SPEI) relative to August, with three months time-scale (SPEI8_3) increased at seven of the sites, but not at Alice Holt, Rogate, Covert Wood, Goyt and Ladybower (Table 2).

Table 2 Temporal changes in the main environmental parameters at the investigated sites.

Principal components analysis (PCA) using across sites long-term averages of environmental variables (PCA_spatial, hereinafter referred to as PCA_s) and within sites annual values (PCA_annual, hereinafter referred to as PCA_a) were conducted to reduce redundancy in the predictors along the climatic gradient. In the case of PCA_s, more than 90% of the variance was explained with two principal components (PC_s). PC_s1 was a general description of site climatic conditions, with highly significant effects from all variables (temperature, VPD and precipitation for the growing season and for the entire year). Strongest (negative) correlations with PC_s1 were with mean and maximum growing season temperature, suggesting that this axis primarily represented a South–North gradient. The PC_s2 correlated most strongly with VPD differences among sites, likely representing the East–West gradient. For the PCA_a, four PCs (PC_a) were required to explain 78% of the total variance. PC_a1 correlated significantly with temporal variability in VPD (negative) and precipitation (positive), PC_a2 with temporal variability in temperature (negative correlations), PC_a3 with growing season (negative) and annual (positive) SPEI indices and VPD (negative), while PC_a4 represented primarily temporal variability in precipitation and VPD (Fig. S1).

Over the 30 year period of the study, the mean annual c_a increased by 51 ppm (from 339 to 390 ppm, slope = 1.7 ± 3.9 ppm year⁻¹)²⁷. Wet N_dep and wet S_dep decreased between 1995 and 2010 at all sites but Savernake and Tummel for S_dep, and Goyt/Ladybower and Tummel for N_dep (Table 2). Despite the general decreasing trends in wet S_dep and N_dep, half of the investigated sites are above the UK’s critical load for N to woodlands of 10–12 kg N ha⁻¹ year⁻¹;²⁸ (Fig. 1B).

Species-specific changes in iWUE

Relative changes in iWUE (i.e., value at 2010 minus value at 1980 divided by value at 1980) increased for Scots pine, oak and beech by 17.8 (± 5.6) %, 22.5 (± 11.3) % and 15.3 (± 7.0) %, respectively at all the sites. For Sitka spruce we found instead that iWUE decreased by 22.5 (± 9.9) %, with especially strong reductions in the youngest stand at Goyt that was planted in 1981, when tree-ring δ¹³C time series began (Table S2).

Trends in iWUE were not consistent across the investigated species. We found that iWUE increased for Scots pine and oak at all the sites (4 and 2 sites, respectively) over the recent 30 years (Fig. 2A, Table S2). Three of the four beech sites (Covert Wood, Shobdon and Thetford) showed no significant changes in iWUE, while a significant increasing trend in iWUE was observed for beech trees at Alice Holt. In contrast to the other species, the two Sitka spruce stands at Goyt and Tummel showed a reduction in iWUE, particularly for the youngest stand at Goyt (Fig. 2, Table S2).

Figure 2

Trends over time in the main physiological and ecological parameters as obtained from the measured tree-ring stable isotopes at the 12 forest stands across Britain. Changes in iWUE (Panel A), c_i/c_a ratio (Panel B), Δ¹⁸O_w (difference between δ¹⁸O_w and δ¹⁸O of precipitation, panel C) and δ¹⁵N_w (Panel D) for four species over the period 1980 to 2010. Each point represents the parameters derived from—or the actual—isotope ratios measured on wood materials pooled from 10 trees for a given year; i.e., δ¹³C_w—derived iWUE and c_i/c_a, δ¹⁸O_w—derived Δ¹⁸O_w and measured δ¹⁵N_w. Variability among the 10 trees per species and site was assessed for the year 2007 and data are reported in the Table S5. Regression lines are given for each site grouped by Ndep levels (i.e., low Ndep, light blue; high Ndep, black).

We tested whether the observed trends could be partially explained by changes in stand parameters i.e., diameter at breast height (DBH), mean height, basal area (BA) for six stands for which data was available (no data were available for the youngest stand at Goyt). Overall we did not find significant relationships between iWUE and stand-related parameters, with the exception of the Scots pine at Ladybower, where iWUE was positively correlated to changes in BA (Supplementary text S1).

Species-specific trends in c_i/c_a ratio and Δ¹⁸O_w

Similarly to what has been reported already for iWUE, changes in the c_i/c_a ratio were not consistent across the species and sites. For both beech and Scots pine at three sites, the parameter did not significantly change over the 30 years (Fig. 2B). The c_i/c_a ratio increased for the Sitka spruce at Goyt (slope = 0.008, p < 0.001) and Tummel (slope = 0.003, p < 0.001), the oak at Savernake (slope = 0.0007, p < 0.05), while it decreased for the oak at Alice Holt (slope = − 0.002, p < 0.01) and the Scots pine at Rannoch (slope = − 0.0009, p < 0.05).

Trends in Δ¹⁸O_w were not significant, except for the decrease for beech in Shobdon (slope = − 0.03 ± 0.013‰ year⁻¹, p < 0.05) and for Scots pine at Rannoch (slope = − 0.021 ± 0.009‰ year⁻¹, p < 0.05) and the increase for Scots pine at Ladybower (slope = 0.026 ± 0.007‰ year⁻¹, p < 0.01) (Fig. 2C). Changes in Δ¹⁸O_w were not correlated with stand-related parameters, except for Rannoch Scots pine, where we observed a significant negative relationship between Δ¹⁸O_w and mean tree height and BA (Supplementary test S1).

Spatial and temporal changes in δ¹⁵N_w at different levels of N_dep

Differences in tree ring δ¹⁵N_w values between sites with N_dep below (low N_dep) and above (high N_dep) the critical loads (Fig. 1B) did not show a distinctive pattern. In the case of beech, the δ¹⁵N_w values were more positive (on average 1.57‰) at the high N_dep than at the low N_dep sites (Supplementary Table S3). However, we found the opposite for the oak, Scots pine and Sitka spruce, with trees at higher N_dep showing more negative δ¹⁵N_w values (Supplementary Table S3).

No significant trends in δ¹⁵N_w were observed (Fig. 2D), except for oak stands at Alice Holt (slope = 0.03‰ year⁻¹, p < 0.05), beech stands at Thetford (slope = 0.04‰ year⁻¹, p < 0.01) and Sitka spruce stand at Tummel (slope = 0.08‰ year⁻¹, p < 0.01). A significant positive relationship between δ¹⁵N_w and wood %N was observed for 5 sites (Fig. S2), including two beech, one oak and two Scots pine stands, but no relationship was observed in the case of the two Sitka spruce stands.

Contribution of climate and anthropogenic factors on physiological and ecological processes

One of our goals was to elucidate the relative contribution of climate and atmospheric drivers on changes in physiological and ecological parameters obtained from tree-ring stable isotopes. Temperature increase and precipitation decrease (the dominant parameters in PC_s1) enhanced iWUE along the N-S gradient (Fig. 3A–C, Table 3). Whereas temporal increase in iWUE was mostly associated with changes in parameters related to moisture supply and demand conditions (VPD, SPEI and Precipitation, i.e., the dominant variables in PC_a1 and PC_a3) (Fig. 3B–D, Table 3). However, the high correlations among individual variables in many axes both in the PCA_s and the PCA_a precluded the possibility of drawing definitive conclusions about individual climatic drivers. Nonetheless, neither the increase in c_a nor the changes in N_dep and S_dep were significant predictors in the model (Table 3). The inclusion of stand-related parameters (age and soil type) did not improve the statistical model fit (Supplementary text S2).

Figure 3

Spatial and temporal changes in iWUE vs. climate. Relationship between iWUE (mean through 1980–2010 for each tree species) and PC_s1 (panels A–C) and between year-by-year iWUE vs. PCA_a1 (panels B–D) across the investigated forest stands as obtained from the linear mixed effect model analyses. For the sake of clarity, we included the results of PCA analyses (panels A and B), to identify the climate factors that mostly affect the changes in iWUE as shown in the panels C and D. Slope and standard error values are reported in the Table 4. We used different symbols and colours for regression lines for each species, but slope presented in Table 4 refer to all observations together. The four investigated species were indicated with different colours (see legend in the panel C for details), while each site was indicated with a different symbol (see main legend for details). In panels A and B, ‘a’ and ‘grs’ indicate mean annual and growing season T, P and VPD, while 'maxgrs' refers to mean of maximum T. As for the SPEI, 8 and 12 indicate values for the month of August with a time-scale of one (8_1), two (8_2) and three (8_3) months and for the month of December.

Table 3 Relationship between physiological and ecological parameters and environmental factors.

Given that the iWUE trend observed for the Sitka spruce stands could be partially related to the age effect, with particular reference to the youngest one at Goyt, we repeated the analyses excluding Sitka spruce. Results confirmed no relationship between N_dep and iWUE (Table 3). We found, however, a negative relationship between both sS_dep and aS_dep and iWUE (Table 3) and the expected significant relationship between c_a and iWUE (Table 3). Finally, when performing the analysis by level of N_dep (i.e., high and low N_dep for sites above and below the critical loads, respectively) and by including or not the Sitka spruce stands, the positive relationship between iWUE and c_a turned out to be significant, with the exception of the analysis for the high N_dep sites and when the Sitka spruce stand at Goyt was included (Supplementary Table S4).

Differences in both spatial and temporal changes in climatic conditions significantly affected Δ¹⁸O_w. Three axes from the PCA_a analyses (PC_a1, PC_a2 and PC_a3) were significant, suggesting a combined effect of temperature and moisture variability on changes in Δ¹⁸O_w (Table 4). When considering the subset of sites where annual atmospheric deposition data were available, we did find a significant and negative relationship between Δ¹⁸O_w and sS_dep and aS_dep. This, however, was not the case when Sitka spruce stands were removed from the analyses, to account for possible age effects on changes in Δ¹⁸O_w (Table 4, Supplementary text S5).

Table 4 Relationship between physiological and ecological parameters and environmental factors.

When considering all sites and species together, at high and low N_dep sites, a clear pattern for δ¹⁵N_w in tree rings was not seen and no differences were observed when grouping species by plant functional type (Table 4). Changes in δ¹⁵N_w were explained primarily by climate factors, particularly temperature and precipitation, with a marginal (p = 0.1) effect of differences across sites in N_dep (Table 4). Soil type and stand age did not appear significant predictors for δ¹⁵N_w in the statistical model (Supplementary text S3).

The combined influences of temporal changes and spatial differences in climatic and atmospheric deposition variables across all three isotopes were examined with the help of a mixed-effect path model (Fig. 4A). Consistent with the earlier analysis using linear mixed models, the first axis of PC_s significantly affected all three isotopes, negatively for iWUE and Δ¹⁸O_w and positively for δ¹⁵N_w. This is consistent with the first axis of PC_s being related to a South–North gradient (i.e., positively with precipitation and negatively with temperature and VPD). Conversely, the first and third axis of PC_a only affected iWUE (both effects negative) and Δ¹⁸O_w (negative effect for PC_a1 and positive effect for PC_a3). This is also consistent with PC_a1 being positively related to precipitation and SPEI and negatively related to VPD and temperature. N_dep was only marginally (p < 0.1) important for δ¹⁵N_w. A significant positive effect of Δ¹⁸O_w on iWUE was also found, while no relationship was observed between iWUE and δ¹⁵N_w (Fig. 4A). Similar results were found when excluding Sitka spruce stands from the analyses for both Δ¹⁸O_w and δ¹⁵N_w, in addition to the significant positive effect of c_a on iWUE and (Fig. 4B). When including in the analyses only sites where annual atmospheric deposition data were available, we also found a negative effect of sS_dep and aS_dep on iWUE (Fig. S5).

Figure 4

Direct and indirect effects on tree-ring stable isotopes and derived physiological parameters. Result from the structural equation modelling analysis, which included all sites (panel A) and all sites excluding the two Sitka spruce stands (panel B). Details of equations are reported in the Table S7. Continuous arrows indicate relationships significant at the p ≤ 0.05 or greater (depending on number of stars, i.e., *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001). Dashed arrows (and the ‘¥’ symbol) indicate relationships significant at p ≤ 0.10. Thickness of lines reflects level of significance. Notice that degrees of freedom vary between site-based and time-based analyses. Black and blue arrows indicate negative and positive relationships, respectively. Double-headed arrow indicates correlated errors between variables. PC_s1 indicates the first component from the PCA_s (using long-term average climate variables across sites) and PC_a1, PC_2 and PC_a3 the first three principal components from the PCA_a (using annual climate parameters). Number next to the paths indicate standardized path coefficients. Numbers below each isotope-related parameters indicate marginal and conditional R², respectively. They are not always the same as those reported in the Tables 3 and 4 due to slightly different equations considered in the SEM analyses. In particular, PFT was not included as fixed factor in any model, while Δ¹⁸O_w was included as fixed factor in the model for iWUE (Ref. Table S7 for all details).

Discussion

Climate and atmospheric deposition changes across the investigated sites

The first goal of this study was to document the changes in climate and atmospheric deposition across the investigated sites over the period 1980–2010, so as to better assess their effects on iWUE and other isotope-related parameters. Trees at most of the investigated sites experienced increasingly wetter and warmer growing conditions, particularly in Scotland, corroborating general trends for Britain²⁹. PC analyses showed that temperature and precipitation are the main climate factors describing the South–North gradient, while parameters related moisture conditions (SPEI, VPD and precipitation) were more relevant in long-term climate variations.

While c_a has continued to increase by 14% from 2010 to 1980, differences were observed in magnitude and directionality of changes in atmospheric deposition. A significant reduction of total wet S_dep and N_dep (NH₄-N + NO₃-N) was also observed at most sites^1,30,31, with steeper slopes for S_dep than N_dep. Decreasing S_dep is consistent with national scale reports, whereby emissions and total S_dep have greatly declined over the last few decades⁴. However, a significant decline in total N_dep (including oxidised and reduced form) has not yet been reported nationally despite reduction of N emissions of both nitrogen oxides and ammonia⁴. Taken all together, these results suggest that directionality and magnitude of changes in N_dep and S_dep across the environmental gradient may modulate tree species response to c_a, as we discuss in the following paragraphs.

Tree species differed in iWUE trends and underlying mechanisms

Our second goal was to assess trends in iWUE and elucidate potential physiological mechanisms underlying changes in iWUE. Three of the four investigated tree species showed that iWUE increased over 30-years, corroborating global tree-ring δ¹³C based analyses^12,13,32. Across sites, the relationship between iWUE and Δ¹⁸O_w (Fig. 4) suggests that overall there is a tight coupling between A and g_s across the investigated species. However, the within-sites relationships were significant only at 5 of the 12 forest stands (Table S2), suggesting physiological strategies differ amongst—and likely also within—tree species.

Scots pine and oak showed an increase in iWUE, but the underlying mechanisms were likely different. Scots pine at southernmost sites (Rogate and Thetford) showed no changes in transpiration losses (no significant trend for Δ¹⁸O_w) and a proportional regulation of A and g_s, which contributed to a constant c_i/c_a. Scots pine has a very conservative strategy regarding water use, due to its tight stomatal control under moisture limitation³³. Whereas, Δ¹⁸O_w for Scots pine at the northernmost sites (Ladybower and Rannoch) significantly changed, though in the opposite direction. The increase in Δ¹⁸O_w for Scots pine at Ladybower would indicate a reduction in g_s, which is supported by the positive relationship between iWUE and Δ¹⁸O_w (Table S2). The site showed the highest mean S_dep values among all sites. Even though trend in S_dep has significantly decreased at the site, we cannot exclude a possible legacy (negative) effect of deposition on g_s²³. In the case of Rannoch, the physiological signal recorded in the reduction in Δ¹⁸O_w could be partially confounded by tree stand development (as a negative relationship was found between Δ¹⁸O_w and tree height and BA) or the significant reduction in S_dep, which would have a positive effect on g_s¹⁰.

In contrast, the increase in iWUE and the significant changes in the c_i/c_a ratio (though in opposite direction) for the two oak stands was mostly related to a more dynamic adjustment of leaf gas exchanges to environmental conditions, with likely reductions in g_s and lower A, leading to a lower increase in iWUE in the case of Savernake (where both c_i/c_a and Δ¹⁸O_w increased) compared to Alice Holt (no changes in g_s, as suggested by no significant trend in Δ¹⁸O_w, while c_i/c_a decreased). This result indicated that S_dep is still negatively affecting oak trees at Savernake—one of the few sites where S_dep has not significantly declined over the last decades. Consistent with our results, a previous study found that exposure to high SO₂ pollution significantly reduced g_s leading to an increase in iWUE for oak trees in southern England¹⁰.

The reduction in iWUE in the fast-growing³⁴ Sitka spruce trees, particularly the youngest stand at Goyt), most likely reflect age or management history effects. Height growth during tree development may affect δ¹³C and hence iWUE estimates via changes in light availability³⁵ and hydraulic conductance³⁶, but also through increasing LAI as canopy develops and contributes to respired CO₂³⁷. Note, however, that the trend in iWUE at Goyt cannot be explained by increasing hydraulic constraints with tree height³⁶, which would be in the opposite direction (i.e., an increase in iWUE in taller trees). It is most likely that the observed trends may reflect common historical upland conifer establishment practices, e.g., fertilisation on tree establishment until approximately 10 years of age and thinning between 15 and 20 years. This may have contributed to reduce tree competitions for light and soil water, thus accelerating growth but reducing iWUE in the early stage of development until canopy closure³⁸.

Beech was the least responsive species, as no changes in iWUE were observed in general, even though the four selected sites were along a clear precipitation gradient (Fig. 1). However, the wettest site (Shobdon) showed the lowest % increase in iWUE (only 5% compared to 16–19% changes at the remaining three sites), which could partially be explained by the increase in g_s suggested by the reduction in Δ¹⁸O_w. Moreover, when comparing even-aged beech and oak stands at the same site (i.e., Alice Holt), with similar soil and climate conditions, we found that despite the two species showing similar iWUE, beech had a higher Δ¹⁸O_w than oak (Fig. S3). This suggests differences between the two species for g_s, which could partially explain the significantly (p < 0.05) higher slope in beech compared to oak for temporal trends in Δ¹⁸O_w.

Changes in tree iWUE and Δ¹⁸O_w across Britain were associated with climate and atmospheric S deposition

Changes (over time) and differences (across sites) in moisture conditions and temperature along the North–South gradient in Britain were overall more important than atmospheric deposition in explaining variations in iWUE. Increase in temperature, both moving North–South (PC_s1) across Britain and over time (PC_a2), significantly increased iWUE. This could be explained by a positive effect of temperature on A, but also a reduction in g_s, likely associated with the increase in VPD, the major component of PC_a3. Indeed, the latter was significantly and positively correlated with Δ¹⁸O_w. This finding is in agreement with results for two broadleaf species in the Northeastern US²⁶, whereby changes in iWUE were predominantly associated with soil moisture conditions and no effects of either N_dep or S_dep were observed.

The detection of atmospheric deposition and c_a signals in our study, however, was possibly confounded by likely age effects and/or associated tree structural and functional changes (as we discussed above). Interestingly, when Sitka spruce stands were excluded from the analyses, the increase in iWUE was associated with rising c_a, but also with the reduction in S_dep, while there was no relationship with N_dep. This is quite important, as S_dep and c_a may have a similar effect on g_s, i.e., an increase in S_dep and c_a leads to a reduction in g_s¹⁰. However, changes in S_dep and c_a do not always follow the same direction. Indeed we reported a reduction in S_dep at most of the sites, under a general increase in c_a. The negative relationship between Δ¹⁸O_w and S_dep suggests that the alleviation of the negative effects from S_dep on g_s²² may be stronger than the c_a effect leading to stomatal closure (which does not find indication in our data). These results on one hand suggest that variations in S_dep outweigh changes in N_dep and c_a. S_dep may influence tree water-use directly, by affecting g_s²² or indirectly, by leading to soil acidification, loss of calcium, which play a significant role in controlling g_s³⁹. On the other hand, they indicate that it is essential to account for changes caused by stand development and management history, if the goal is to disentangle climatic and anthropogenic drivers of change in iWUE.

No changes in ecosystem N availability due to N_dep

Given that some of the investigated forest stands have reached the critical load in terms of N_dep (Ref. Fig. 1B), we expected to find significant changes in ecosystem N due to possible N losses from the forest ecosystems. Yet, spatial and temporal variations in δ¹⁵N_w values did not indicate differences in ecosystem N dynamics between high and low N_dep levels. If this was the case, we should have observed an increase in tree-ring δ¹⁵N_w over time and moving from low to high N_dep sites, as a consequence of high nitrification and loss (through denitrification and/or leaching) of the ¹⁵ N-depleted NO₃, leaving trees with ¹⁵ N-enriched NO₃⁴⁰. This was only observed for the beech at Thetford, which is already confirmed as an N saturated site by detailed gradient studies⁴¹. Repeated soil analysis suggests N accumulation in soil organic layers under the conifer species, including the Tummel Sitka spruce site over the last 15 years, but no signs of N saturation⁴¹. However, a significant positive relationship has been observed between N_dep and NO₃ leaching across some conifer sites in this study, so further N input to conifer forests could cause significantly higher NO₃ leaching⁴¹.

Our tree-ring δ¹⁵N_w data suggest that N availability has not changed for most of the forest stands over the investigated 30-years period, which is in contrast to declining δ¹⁵N trends reported both in the USA^22,42 and globally⁴³, whereby studies suggest on going oligotrophication. However, we cannot exclude that N recycling processes within trees could also have contributed to the observed lack of a trend in δ¹⁵N_w⁴⁴, particularly in the case where no relationship was found between δ¹⁵N_w and wood %N (Fig. S1).

The strong climate signal explaining spatial variations in the tree-ring δ¹⁵N values is consistent with a global analysis⁴⁵, which identified temperature and precipitation as the main climatic drivers of changes in the foliar δ¹⁵N_w across different forest ecosystems. An indirect effect of wet N_dep-via climate differences- on tree-rings δ¹⁵N_w cannot be fully excluded. Indeed, the significant negative relationship between atmospheric deposition and PC_s1, partially indicating that wet N_dep is strongly related to the amount of precipitation. The predominantly westerly airflow across the UK brings less polluted air from the Atlantic. Orographic cloud formation in the more mountainous regions of the NW however leads to a substantially higher rainfall and consequently higher N_dep⁴⁶.

Our results also indicate that pinpointing variations in δ¹⁵N_w caused by gradual changes in ambient N_dep is more challenging than in N manipulation experiments, where abrupt and high N doses over relatively short periods of time predominate. Moreover, the lack of information regarding both temporal variability in dry N_dep and changes in the N isotopic signatures for NH₄ and NO₃ might increase the uncertainties in the detection of changes in atmospheric N_dep by using δ¹⁵N in tree rings⁴⁷. Monitoring changes in the isotopic composition of N-specific compounds in rainfall over time can greatly improve our ability to use δ¹⁵N in tree rings to detect changes in N input from N_dep.

Conclusions

Long-term forest monitoring systems, such as the Level II—ICP forests programme, provide unique near-natural systems for assessing the effect of climate change on ecophysiological responses of different tree-species at a regional scale and elucidating interactions among environmental forcing factors and forest ecosystem response. Our results showed that the increase in iWUE was not uniform across sites and that species-specific underlying physiological mechanisms were likely affected by the interactions between climate and atmospheric drivers (for oak and Scots pine), but also an tree structural changes during stand development (for Sitka spruce).

Spatial and temporal changes in temperature and moisture conditions overrode the effect of atmospheric deposition and c_a on changes in iWUE for the investigated forests in Britain. This is remarkable since such an increase is widely predicted to occur in response to increasing c_a. Our results suggest that the effect of increasing c_a on temporal changes in iWUE could be over estimated, if concomitant changes in atmospheric deposition or ontogenic effects (such as structural changes during tree development) are not accounted for. The tree-ring δ¹⁵N_w analyses did not provide evidence for changes in N availability caused by changes in N_dep. In particular, sites receiving high N_dep (now considered above the critical loads) did not show evidence of N saturation, with the exception of the beech site at Thetford. Spatial differences in tree ring δ¹⁵N_w were mostly explained by differences across sites in temperature and precipitation, rather than changes in N_dep.

The multiple-species and regional analysis indicate that climate change may affect the most common native and introduced species in British woodlands. Lower summer rainfall and high temperature and VPD are likely to become more frequent in South-eastern Britain, thus affecting the future site suitability of beech woodland, as the species is more susceptible to drought⁴⁸, while oak and Scots pine could cope better (see e.g., results at Alice Holt). However, even though oak may be physiologically plastic in response to future climate change, widespread oak decline across Britain has been observed related to a number of biotic and abiotic factors, including climate and pollution⁴⁹. Sitka spruce, the major upland planted timber species in the UK, native of coastal Northwestern America, is likely to maintain continued good growth in British northern uplands where water stress is less pronounced. However, the young ages of these stands and intensity and frequency of management interventions make it more difficult to disentangle development from environmental effects on iWUE trends.

Methods

Site and sampling

We selected twelve monoculture tree stands of the most common tree species in Britain, Scots pine (Pinus sylvestris L.), Sitka spruce (Picea sitchensis Bong. Carr.), pedunculate oak (Quercus robur L.) and common beech (Fagus sylvatica L.). The majority of the stands were experimental sites within the Level II- ICP intensive forest monitoring network (http://icp-forests.net/), with the exception of Covert Wood, Shobdon and Goyt. The Goyt site was added as a high N_dep site as a contrast to the low N_dep Sitka spruce site in Scotland (Fig. 1, Table 1, Supplementary Table 1). For each species, forests were selected with similar soil type and age, but with contrasting N_dep, S_dep and climate, particularly rainfall and temperature, as described in Fig. 1, Table 1 and Supplementary Table 1. Stand information (mean tree height, mean diameter at the breast height and basal area) as measured for target years and for some of the forest stands are shown in Fig. S4.

At each ICP forest site, a plot of 0.25 ha was established in 1995 to carry out monitoring³⁰ and a similar protocol was followed at the Goyt and Shobdon sites. Within each plot, 10 trees were selected for the collection of 3 wood cores per tree by using a 5 mm diameter increment borer, which were placed in paper straws for transport. Sampling was carried out between November 2010 and March 2011. Once in the laboratory, samples were dried at 70 °C for 48 h. Of the three wood cores sampled, one was kept for dendrochronology, with the other two used for stable isotope analyses.

Climate and atmospheric deposition data

Climate data (Temperature, T, Vapour Pressure Deficit, VPD, Precipitation, P) were obtained from automated weather stations at the sites and/or the nearest local meteorological stations (data were provided by the British Atmospheric Data Centre). Annual mean (T_a) and mean maximum (T_amax) values for temperature were calculated from monthly mean and maximum air temperature, T, respectively, and annual precipitation (P_a) was calculated as the sum of total monthly precipitations. Annual VPD was calculated from averaging monthly values obtained from mean monthly maximum temperature and minimum monthly relative humidity. For all the parameters, mean values were also calculated over the growing season, i.e., from May to September. We also considered the standardised precipitation-evapo-transpiration index, SPEI, relative to August, with 1 (SPEI8_1), 2 (SPEI8_2) and 3 (SPEI8_3) months time-scale and SPEI relative to December, with 1 and 12 months time-scale, the latter providing year-cumulated soil moisture conditions. SPEI values were obtained from the global database with 0.5 degrees spatial resolution available online at: https://sac.csic.es/spei/.

Yearly wet nitrogen (N_dep) and sulphur deposition (S_dep) were obtained from measured bulk precipitation and throughfall water volumes at the sites and measured elemental concentrations (NO₃⁻, NH₄⁺ and SO^2–₄) as previously described³⁰. For the spatial analyses, we considered mean of annual deposition (sN_dep and sS_dep), obtained as the sum of total (NH₄-N + NO₃-N for N_dep) wet and dry deposition. The latter were estimated as difference between throughfall and bulk precipitation N fluxes³⁰. For Rogate only 1 year (2010) of monitoring was available. For Goyt site, atmospheric deposition data collected at Ladybower were considered, as the two sites are 30 km apart. For two sites (i.e., Shobdon and Covert Wood), which were not part of the regular ICP forest sites, the wet deposition obtained from the UK 5 × 5 km grid N_dep and S_dep dataset was used⁴. The estimate included wet and dry NH_x-N (NH₄, NH₃), NO_y-N (NO₂, NO₃, HNO₃) and S (SO_x = SO₂ and SO₄) deposition, modelled using FRAME with 2005 emissions data⁴. However, only the total wet deposition was included in the analyses here, as we previously reported a good agreement between modelled and measured wet N_dep⁵⁰.

For the temporal analyses, only wet deposition (as calculated from NO₃⁻, NH₄⁺ and SO^2–₄ concentrations in bulk precipitation) was considered (indicated as aN_dep and aS_dep), given the uncertainties associated with the quantification of the dry deposition. For instance, when differences between throughfall and bulk precipitation are < 0 it is assumed atmospheric deposition is retained by tree canopies, but this does not necessarily mean that there is no dry deposition. At Rogate only one-year data were available so we considered annual wet deposition data for Alice Holt, which is within 19 km distance. This was also the case for Goyt and Ladybower, which are 30 km apart. Shobdon and Covert Wood were not included in the analyses where annual deposition data were considered (see earlier in the text and Ref. Table S7).

Stable isotope analyses

Wood cores were subjected to removal of mobile N and extractives with a soxhlet apparatus as described in Guerrieri et al.²³. After the chemical pre-treatment, wood cores were dated and cross-dated from 2010 back to 1980 and then separated with a scalpel as follow: single annual rings from 2010 to 1995 and then groups of 3 annual rings from 1994 back to 1980. We maintained the annual resolution from 1995 onward because this is the period when the UK-ICP forest network was established and atmospheric deposition was monitored.

To minimise the cost of the stable isotope analyses while including 12 sites and 4 different tree species, the wood materials were pooled from 10 trees (2 wood cores per tree) for each given ring or group of rings. However, for one year (i.e., 2007) and at two sites for each species, carbon and oxygen isotope ratios for each of the 10 trees was measured, so as to assess the variability among trees (Supplementary Table S5). For each core per tree species, each ring was cut with a razor blade under a microscope, milled and homogenized in a centrifugal mill, and then pooled by year. Moreover, for δ¹⁵N_w analyses, we only included the sites where long-term atmospheric deposition data were available (i.e., tree species at Shobdon and Covert Wood were excluded).

An amount of 0.4–0.6 mg of extracted wood sample from each given ring (or group of annual rings for the years 1994 back to 1980) was weighed in tin capsules and combusted in the elemental analyzer (NA2500, Carlo Erba) for the determination of δ¹³C_w by VG Prism III Isotope ratio mass spectrometer at the School of Geosciences (University of Edinburgh, UK). For δ¹⁵N_w, 25–30 mg of wood sample was weighed in tin capsules and combusted on a PDZ Europa ANCA-GSL elemental analyzer interfaced to a PDZ Europa 20–20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK). For δ¹⁸O_w, 0.8–1 mg of each sample was weighed in silver capsules and analyzed on a PyroCube (Elementar Analysensysteme GmbH, Hanau, Germany) interfaced to an Isoprime VisION (Isoprime Ltd., Stockport, UK, a unit of Elementar Analysensysteme GmbH, Hanau, Germany). Analyses were carried out at the Stable isotopes facilities of the School of GeoSciences (University of Edinburgh, UK) for δ¹³C_w, and at the Stable Isotope facility of the UC Davis, University of California (USA) for δ¹⁸O_w and δ¹⁵N_w. Stable isotope abundances are expressed as ratios of ¹³C/¹²C, ¹⁵ N/¹⁴ N and ¹⁸O/¹⁶O using δ-notation (in per-mil;‰) relative to international standards (VPDB for δ¹³C_w, atmospheric N₂ for δ¹⁵N_w and VSMOW for δ¹⁸O_w). The standard deviation for internal standards was 0.1‰ for δ¹³C_w (PACS-2), 0.2, 0.3 and 0.4‰ for δ¹⁸O_w (IAEA 600, IAEA 601 and IAEA 602, respectively), and between 0.1 and 0.3‰ for δ¹⁵N_w (USGS-41 Glutamic Acid and peach leaves, respectively).

Calculation of iWUE and Δ¹⁸O_w

The iWUE and the c_i/c_a ratio were derived from measured δ¹³C_w values, and based on the well-established theory linking leaf c_i/c_a with carbon isotope discrimination, Δ¹³C_w,⁵¹ as shown in the equation below:

$$ \Delta^{13} C_{w} = a + \left( {b - a} \right) \frac{{c_{i} }}{{c_{a} }} = \frac{{\left( {\delta^{13} C_{a} - \delta^{13} C_{w} } \right)}}{{\left( { 1 + \frac{{\delta^{13} C_{w} }}{1000}} \right)}} $$

(1)

where δ¹³C_a and δ¹³C_w are the carbon isotope compositions of c_a and wood, a is the isotope fractionation during CO₂ diffusion through stomata (4.4‰) and b is the isotope fractionation during fixation by Rubisco (27‰). Note that Eq. (1) is the “simplified version” of the Farquhar model describing carbon isotope discrimination in plant material, which does not include effects due to mesophyll conductance and photorespiration. We derived c_i from the following equation:

$$ c_{i} = c_{a} \frac{{(\delta^{13} C_{a} - \delta^{13} C_{w} ) - a}}{b - a} $$

(2)

c_a values were obtained from Mauna Loa records²⁷, and δ¹³C_a values were obtained from Mauna Loa records⁵ from 1990 to 2010, while from 1990 back to 1980 we used data published in Ref.⁵². iWUE (μmol CO₂ mol⁻¹ H₂O) was then calculated using the following equation:

$$ iWUE = \frac{A}{{g_{s} }} = \frac{{c_{a} - c_{i} }}{1.6} = \frac{{c_{a} }}{1.6} \left( {\frac{{b - \Delta^{13} C_{w} }}{b - a}} \right) $$

(3)

where 1.6 is the molar diffusivity ratio of CO₂ to H₂O (i.e., g_CO2 = g_H2O/1.6). Note that in the Eqs. (2) and (3) we used average of values measured over growing season months (May–September) for both c_a and δ¹³C_a.

Tree-ring oxygen isotope discrimination, Δ¹⁸O_w, was calculated according to Eq. (4)⁵³:

$$ \Delta^{18 } O_{w} = \frac{{\delta^{18} O_{w} - \delta^{18} O_{s} }}{{1 + \left( {\frac{{\delta^{18} O_{s} }}{1000}} \right)}} $$

(4)

where δ¹⁸O_w is the oxygen isotope composition we measured in each ring, while δ¹⁸O_s is the oxygen isotope composition of the source water, i.e. the soil water, which we assumed to reflect the δ¹⁸O of precipitation (δ¹⁸O_P). We estimated annual values of δ¹⁸O_P at each site as described by Barbour et al.⁵⁴, by considering the following equation:

$$ \delta^{18} O_{P} = 0.52T_{a } - 0.006T_{a}^{2} + 2.42P_{a} - 1.43P_{a}^{2} - 0.046\sqrt E - 13.0 $$

(5)

where T_a, P_a and E are the annual mean air temperature, precipitation (this latter expressed in m) and elevation (m asl), respectively. Mean values of estimated δ¹⁸O_P from Eq. (5) were in line with estimates from the Online Isotopes in Precipitation Calculator (https://wateriso.utah.edu/waterisotopes/pages/data_access/oipc.html) and measured δ¹⁸O_P values at Keyworth (Supplementary Table S6). The modelled δ¹⁸O_P did not show a significant trend at most of the sites, with the exception of Goyt/Ladybower (slope = − 0.03 ± 0.007‰ per year, p < 0.001) and Covert wood (slope = 0.02 ± 0.009‰ per year, p = 0.05).

We assumed Δ¹⁸O_w to reflect the leaf water Δ¹⁸O, which is affected by transpiration. Notably, less enriched (in ¹⁸O) water from the soil and more enriched (in ¹⁸O) water at the leaf evaporative sites continuously mix, as a function of transpiration rates and the pathway of water movement through foliar tissues (i.e., Péclet effect)⁵⁵ so that lower leaf Δ¹⁸O results from an increase in transpiration and g_s⁵⁶.

The physiological signal imprinted in the foliage may be dampened in tree rings, due to post-photosynthetic fractionation during translocation of sucrose and synthesis of cellulose in the tree stem⁵⁷. This leads to an offset between foliar and tree-ring δ¹⁸O and also δ¹³C values. However, accounting for the offset when interpreting tree-ring isotopes is still challenging, as it is not clear whether the offset is species-specific, if it is maintained over the long-term and what are the mechanisms driving it^57,58.

Statistical analyses

Linear regression analyses were initially used to explore whether (1) temporal trends existed between tree ring isotopes, iWUE and environmental data at each site (Fig. 2); (2) there was a relationship between iWUE and Δ¹⁸O_w and parameters describing stand development (diameter at the breast height and height, Supplementary Table S2 text S1); (3) changes in iWUE and δ¹⁵N_w were correlated with age and soil type (Supplementary text S2 and S3). Subsequently, we considered models jointly allowing explanation of spatial and temporal variation in isotopic data. To explain temporal trends in isotopic data (iWUE, Δ¹⁸O_w, c_i/c_a, δ¹⁵N_w) jointly across multiple sites, we considered both explanatory variables that varied yearly for each site and mean climatic data averaged over time for each site. Yearly time series and mean climatic data for each site (mean temperature, maximum temperature, precipitation and vapour pressure deficit VPD) were calculated for each year (from 1980 to 2010) and also separately for each growing season. We considered SPEI for the month of August with a time-scale of one, two and three months and then the SPEI for the month of December with a time-scale of one and twelve months. By definition, being centered around zero, SPEI defines yearly anomalies and cannot be used as a site index. We also included both the annual deposition data (i.e., aS_dep and aN_dep, to assess the effect of temporal changes) and the mean over the monitoring time (i.e., sS_dep and sN_dep) to evaluate both the contribution of within-site temporal changes and cross-sites differences on changes in iWUE, c_i/c_a, Δ¹⁸O_w and δ¹⁵N_w. Note that we only have data for the aSdep and aNdep at 10 of the 12 sites. To eliminate auto-correlation between mean site variables and yearly variables for each site, site variables were globally centered, whereas yearly data were group-centered⁵⁹. To eliminate auto-correlation among individual climatic variables of the two groups, we conducted two principal components (PC) analyses, after centering and scaling the variables. The first PC analysis considered across sites long-term averages of environmental variables (PCA_s), the second within sites annual time series (PCA_a). Tables of percentage of variance explained and scree plots were examined to determine how many PC to retain. Linear mixed models (using library nlme in R⁶⁰) were employed for each of the isotope data series to explain temporal and spatial patterns of variation as a function of climatic and atmospheric deposition conditions. We also ran the linear mixed model without the two Sitka spruce stands, in order to account for other source of variations (e.g., age-effect) for the isotope parameters and iWUE (particularly for the youngest stand at the Goyt site). To explore the significance of systematic differences among the 12 (or 10, when Sitka spruce was excluded) sites occupied by the two evergreen and the two deciduous species, a categorical variable with two levels (combining species into two functional groups) was introduced as a fixed factor. Since multiple species were present at some of the sites, an identification factor for each site × species combination was employed as random factor. The initial models included all PC previously identified as potential explanatory variables for both spatial and temporal variation and also forest stand-related (age and soil type) and anthropogenic (atmospheric CO₂, N_dep and S_dep) factors (Ref. Supplementary Table S7 including all the models tested and reference to tables and supplementary text where results are reported). These models were then gradually simplified until the minimal significant model was achieved, i.e., excluding all PC and other factors that were not significant following simultaneous tests for general linear hypothesis testing (package multicomp,⁶¹). A correlation structure of order 1 was included in the model for each site × species combination to allow for the temporal dependency of measurements carried out in subsequent years. In the case of nested models, significance was tested using a chi-square test with one degree of freedom. Quality of fit was assessed using residual distribution plots, qqnorm plots of standardised residuals against quantiles of standard normals for both individual points and for the random effects, and auto-correlation function plots of normalized residuals as a function of measurement lags. Marginal (only fixed factors) and conditional (fixed + random factors) percent of explained variance (R²_m and R²_c, respectively) were calculated using package MuMIn⁶².

Finally, to examine the joint effects of climatic conditions on Δ¹⁸O_w, δ¹⁵N_w and iWUE, a mixed-effect confirmatory path model was employed using package piecewiseSEM⁶³. For each isotope and iWUE, the final model from linear mixed effect model analyses (Tables 3, 4, Supplementary Table S7) was considered, with the only modification of excluding PFT as fixed factor and including also Δ¹⁸O_w as fixed factor in the model for iWUE and the correlated errors between PC_s1 and sNdep. PiecewiseSEM allows the fitting of hierarchical models with random effects on data with a multivariate structure, allowing for the identification of indirect effects and unresolved covariance among endogenous variables. Model goodness-of-fit was assessed using a chi-square test against Fisher C, based on Shipley’s test of direct separation, which tests for the overall significance of missing relationships among unconnected variables, while the significance of any given missing path was evaluated using individual p-values. A combination of non-significant Fisher-C and individual p-values tests implies that the hypothesised relationships among the variables are consistent with the data without missing any significant relationship⁶³. All statistical analyses were carried out inside RStudio 1.0.143 with R 3.4.0⁶⁴.