Permafrost Biases Climate Signals in δ¹⁸O_tree-ring Series from a Sub-Alpine Tree Stand in Val Bever/Switzerland

Abstract

During recent decades, stable oxygen isotopes derived from tree-ring cellulose (δ¹⁸O_TRC) have been frequently utilised as the baseline for palaeoclimatic reconstructions. In this context, numerous studies take advantage of the high sensitivity of trees close to their ecological distribution limit (high elevation or high latitudes). However, this increases the chance that indirect climatic forces such as cold ground induced by permafrost can distort the climate-proxy relationship. In this study, a tree stand of sub-alpine larch trees (Larix decidua Mill.) located in an inner alpine dry valley (Val Bever), Switzerland, was analysed for its δ¹⁸O_TRC variations during the last 180 years. A total of eight L. decidua trees were analysed on an individual base, half of which are located on verified sporadic permafrost lenses approximately 500 m below the expected lower limit of discontinuous permafrost. The derived isotope time series are strongly dependent on variations in summer temperature, precipitation and large-scale circulation patterns (geopotential height fields). The results demonstrate that trees growing outside of the permafrost distribution provide a significantly stronger and more consistent climate-proxy relationship over time than permafrost-affected tree stands. The climate sensitivity of permafrost-affected trees is analogical to the permafrost-free tree stands (positive and negative correlations with temperature and precipitation, respectively) but attenuated partly leading to a complete loss of significance. In particular, decadal summer temperature variations are well reflected in δ¹⁸O_TRC from permafrost-free sites (r = 0.62, p < 0.01), while permafrost-affected sites demonstrate a full lack of this dependency (r = 0.30, p > 0.05). Since both tree stands are located just a few meters away from one another and are subject to the same climatic influences, discrepancies in the isotope time series can only be attributed to variations in the trees’ source water that constraints the climatic fingerprints on δ¹⁸O_TRC. If the two individual time series are merged to one local mean chronology, the climatic sensitivity reflects an intermediate between the permafrost-free and –affected δ¹⁸O_TRC time series. It can be deduced, that a significant loss of information on past climate variations arises by simply averaging both tree stands without prior knowledge of differing subsurface conditions.

1. Introduction

Mountains are recognised as important indicators displaying the fortified impact of climate and environmental change. They exhibit dynamics in physical and biological systems that are more directly identifiable than in other geographical regions or ecosystems [1,2,3]. For high elevation sites, Pepin and Seidel [4] reported for the second half of the 20th century a median surface warming of +0.13 °C/decade. This warming is even more pronounced during the most recent decades, clearly indicating a more enhanced and further accelerating increase in temperature [5]. In a study focusing on elevation-dependent warming, Pepin et al. [6] stated that the warming in mountains is verifiably amplified with elevation. As a result, high-mountain environments experience more rapid temperature changes compared to ecosystems at lower elevations. In addition to the most apparent responses of glacier systems towards these changes, permafrost as one key component of the terrestrial system in continental high mountain environments shows an extremely high sensitivity to climate change. For example, the large-scale thawing of alpine permafrost already is—and will be in future—one key component in adjusting regional hydrological cycles in high mountain ecosystems [7].

Profound knowledge about pre-industrial baselines of (hydro)climate is an essential precondition for decision-makers and climatologists (i) to set observed recent climate trends from instrumental datasets or climate models into a long-term context, (ii) to validate findings on climate change derived by forward modeling with (past) natural climate variability and (iii) to de-couple anthropogenic effects from natural (climate) variability. Since the sparsity of available instrumental records from remote high-mountain areas limits the robustness of statements and model outputs on climate change, climate proxy datasets are a key component in evaluating information on past climate variability. During recent decades, the use of stable oxygen isotope ratios (¹⁸O/¹⁶O) stored in tree-ring cellulose (δ¹⁸O_TRC) underlined its capability for serving as a highly valuable paleoclimatic proxy. Numerous studies across the globe revealed that δ¹⁸O_TRC is a reliable recorder of temperature [8,9,10,11] and various hydroclimate variables such as precipitation, relative humidity, cloud cover or drought indices [12,13,14,15]. Moreover, extreme events impacting the regional hydrology as cyclone activity, flooding or river discharge can be reconstructed by δ¹⁸O_TRC [16,17,18,19]. Short-term (weekly/seasonal) to long-term (interannual/decadal) variations in large-scale circulation patterns are also stored in δ¹⁸O_TRC-series, e.g., due to varying moisture sources [20,21,22,23,24,25], which in turn strongly bias/overprint local climate conditions. Several parameters mentioned above may contribute in a similar proportion to the annual δ¹⁸O_TRC signal at a specific sampling site, further exacerbated by feasible changing stabilities of the climate-proxy relationships [12,26]. Ultimately, the complex sensitivity of δ¹⁸O_TRC hampers the identification and disentanglement of a single causative climate parameter [27]. Yet, the advantage of investigating δ¹⁸O from organic tree tissues is its ability to interlink different natural archives via the water cycle [21].

In trees, the isotopic composition of atmospheric (source) water is strongly modified by various fractionation processes occurring at leaf level [27,28,29,30,31,32,33]. This includes processes during the isotopic exchanges between leaf, xylem and the atmosphere [31,34,35] and subsequent biochemical isotopic exchanges and fractionations occurring during cellulose synthesis [36,37,38,39]. However, an examination on the isotopic composition of the trees’ source water is still challenging, as it represents a mixed signal of precipitation and soil water. General circulation models can estimate the cascading isotopic fractionation processes encountering between the source (e.g., ocean) and the sink (precipitation at sampling site) [40]. Yet, their spatial resolution is limited and lacks in sufficient representatively of the complex topography in high-mountain systems. During the percolation of precipitation into the soil, further site-specific fractionation processes occur alternating soil properties [41,42,43]. Additionally, different (seasonal) water sources such as meltwater can further mask the initial isotopic composition of precipitation. Still, the quantification of these phenomena is rarely considered in paleoclimatic studies. Especially the influence of permafrost on δ¹⁸O_TRC is often neglected. In particular, the interpretation of climate-proxy relationships inferred from δ¹⁸O_TRC time series from high-latitude boreal ecosystems being prone to permafrost influences feature many difficulties due to the complexity in disentangling regional hydrology [24,44,45,46].

Despite of the large extent of continuous permafrost occurrence in the high latitudes of the Northern Hemisphere [47,48], only few studies address the interaction of continuous permafrost and the dynamics of the active layer on boreal δ¹⁸O_TRC [45,46,49,50,51,52,53,54,55]. In the mid-latitudes, the distribution of permafrost is usually limited to high-elevated mountain sites, commonly above the tree line [7,56,57]. Cold temperatures and a comparatively small amount of annual precipitation favor the regional existence of sporadic permafrost below timberline. However, such sporadic permafrost occurrences are rarely discovered since permafrost is not visible per se, making such locations very scarce.

Interestingly, the potential influence of permafrost on δ¹⁸O_TRC variations and hereof-derived climate reconstructions is still a desideratum of paleoclimatic studies based in high mountain areas. From the conceptual background of the signal transfer of δ¹⁸O from the atmosphere (source) to the tree ring cellulose (sink) it is evident, that the occurrence and seasonal dynamics of permafrost and its coupled effects on soil moisture must impact the strength of the source water signal in δ¹⁸O_TRC. Apparently, sub-alpine sampling sites within the mid-latitudes are highly represented in existing δ¹⁸O_TRC studies [58] but to date the interaction of mid-latitudinal (mountain) permafrost with tree-ring δ¹⁸O_TRC has not been investigated in detail. Therefore, the investigation on verified small-scale permafrost occurrences coupled to subalpine tree stands offers a novel and unique possibility to disentangle possible effects of mountain permafrost on δ¹⁸O_TRC.

Within this study, we compared δ¹⁸O_TRC variations from a tree stand of sub-alpine European larch trees (Larix decidua Mill.) in the Val Bever/Switzerland (Figure 1) where trees are verifiably growing on and outside shallow sporadic low-elevated permafrost lenses [59,60]. We thoroughly analysed the climate-proxy relationship for both settings to investigate the potential effects of permafrost on the signal strength in the respective time series. Most likely, the trees’ source water is significantly altered by seasonal effects of freezing and thawing of the active layer or even melting processes of the permafrost lenses, leading to an attenuated or distorted climate-proxy signal.

2. Materials and Methods

2.1. Study Site

The west-east oriented, trough-shaped valley of the river Beverin (Val Bever) is located in the Upper Engadin, Switzerland. Within the Swiss Alps, it serves as a tributary for the Inn River system (Figure 1 and Figure 2). Due to the high elevations of the Bever and Inn valleys in our study area (1700–1800 m asl), the mean annual temperature recorded by the Samedan meteorological climate station is comparably low (1.2 °C, Figure 1B). A pronounced annual temperature amplitude of approximately 20 °C characterises a typical continental climate with cold winters and warm summers. Adjacent mountain ranges exceeding 3000 m asl mainly block the influx of moist air masses and typically promote highly radiative weather conditions. Accordingly, the mean annual total precipitation accounts for less than 700 mm characterising local climate conditions as a typical inner alpine dry valley climate [65,66]. Due to the low amount of precipitation, a relatively thin closed snow cover is formed in winter. Combined with the feature that the region is prone to pronounced cold inversions during winter, this results in a low isolating effect of the surface against cold winter temperatures [67,68].

The dominant local tree species are the drought-tolerant European larch (Larix decidua Mill.) and Swiss stone pine (Pinus cembra L.) [62]. Mixed forests of larch and stone pine cover the valley slopes up to 2300 m asl, with a steady decrease of larch from the valley bottom upwards (Figure 2B). As a result of intense former pastoralism and human disturbances, southern exposed slopes are mainly forested with larch forests [62,69]. Blocky granitic deposits in the zones of highest morphodynamic activity (rock falls, debris flows or avalanche tracks; Figure 2A,B) are completely treeless or sparsely vegetated with pioneer tree species (Alnus viridis (Chaix) DC., Pinus mugo Turra).

Figure 2. (A) RGB composite image of the Val Bever (Sentinel-2, 5 July 2020) and Alpine Permafrost Index Map [70]. (B) Location of sampled trees of this study and location of permafrost in the sample plot. (C) Minimal annual solar radiation for the investigation area derived from the Solar Analyst tool for ArcView GIS [71,72].

Figure 2. (A) RGB composite image of the Val Bever (Sentinel-2, 5 July 2020) and Alpine Permafrost Index Map [70]. (B) Location of sampled trees of this study and location of permafrost in the sample plot. (C) Minimal annual solar radiation for the investigation area derived from the Solar Analyst tool for ArcView GIS [71,72].

At lower altitudes (~1800 m asl) on the north-facing slope, geophysical soundings and ground surface measurements confirmed the occurrence of thin isolated, sporadic permafrost lenses at shallow depth down to ~20 m [59,60,73,74,75,76,77] (Figure 2B). Compared to permafrost sites in the periglacial zone (e.g., [78,79,80]), the active layer thickness is relatively shallow (1–3 m) and ground temperatures are just below zero degrees (−0.2 °C at 8 m depth). This subalpine permafrost occurrence is approximately 500 m below the expected lower limit of discontinuous permafrost, frequently represented by rock glaciers at 2300 m asl in the Upper Engadin [57,77,81,82]. Permafrost formation outside the expected distribution limits of mountain permafrost is attributed to mutually reinforcing effects caused by micro-climatology, specific topography as well as local surface- and subsurface conditions. The blocky and well-drained talus material enables seasonal micro-circulation systems (chimney effects) and reduces thermal conductance leading to supercooled subsurface conditions [77,81,83]. At our study site, solar radiation potentially reaching the soil surface is hampered by a mainly northern exposition (Figure 2C) and forest cover. Additionally, the poorly developed mineral soils are covered by an isolating organic layer up to 30 cm [59].

2.2. Data Generation and Statistical Evaluation of δ¹⁸O_TRC Time Series

Within this study, a total of 88 L. decidua individuals were sampled at breast height using a five-diameter increment corer (two cores from each tree individual) during various field campaigns between 2006 and 2012, ahead of a complete forest clearance in 2014 [84,85]. Since the tree-ring is not yet fully developed in summer 2012, the last completely formed tree-ring analysed for its δ¹⁸O_TRC is dated to 2011. All sampled cores were cross-dated statistically using the “dplR” package [86] implemented within the open-source statistical language R [87]. Afterwards the individual tree-ring width series were merged into a site chronology. The quality of our site chronology was verified against a reference chronology from the southerly exposed slope in the Val Bever which is located outside the permafrost distribution. In addition, we validated it against available regional tree-ring chronologies from the International Tree-Ring Data Bank (ITRDB). Out of the overall sample pool at our site, in total eight trees were selected for isotope analysis, whereof four cores each represent trees growing on permafrost (PF) and non-permafrost (nPF) locations, respectively (Figure 2B). The cores for isotope analysis were selected based on the following criteria: (i) only old/mature trees were chosen for the isotope analyses and only the last 180 years were regarded to minimise possible tree-ring δ¹⁸O juvenile effects [88,89]. The mean tree age of the eight chosen samples is 223 years (1788–2011), (ii) chosen individuals reveal high inter-series correlation coefficients and Gleichläufigkeit (GLK; sign test), (iii) only trees without missing rings were considered and (iv) trees showing growth depressions or narrow tree-rings were disregarded to obtain sufficient material for α-cellulose extraction and subsequent isotopic analysis. For the establishment of the final δ¹⁸O_TRC series, trees were analysed individually (no data pooling) to minimise potential loss of information on inter-tree variability [90]. Consequently, the individual years were cut off with a razor blade in compliance with the tree-ring width chronology. Subsequently, α-cellulose was extracted following standard laboratory methods [20,91]. During isotope measurements, internal and international laboratory standards were periodically interposed, resulting in an overall analytical precision for the δ¹⁸O_TRC measurements of <0.2‰. The notation for the δ¹⁸O_TRC is specified based on the international standard Vienna Standard Mean Ocean Water (VSMOW). In addition to the individual examination of the permafrost-affected time series (PF-δ¹⁸O_TRC) and non-permafrost affected tree stands (nPF-δ¹⁸O_TRC), a merged site isotope chronology (MEAN-δ¹⁸O_TRC) was developed, which includes all trees irrespective of the subsoil conditions. The statistical reliability of the three resulting δ¹⁸O_TRC series (PF-δ¹⁸O_TRC, nPF-δ¹⁸O_TRC, MEAN-δ¹⁸O_TRC) were tested using standard dendrochronological methods implemented within the “dplR” package [92] for the open-source statistical language R [87]. The mean inter-tree correlation coefficient (RBar [93] and inter-series correlation coefficient (IC) express the averaged correlation value out of each tree pair and the correlation of each tree with the resulting chronology, respectively. As statistical test to express parallel variation of the individual trees’ time series, the Gleichläufigkeit (GLK) was calculated. Since L. decidua is a deciduous tree species prone to possible carry-over effects of the previous growing season [94,95], we additionally tested the respective chronologies for significant first order autocorrelations (AC1). Subsequently, the final chronology uncertainties were estimated by applying a bootstrapped 95% confidence interval [96,97], based on 1000 replications. As Larix decidua Mill. is prone to cyclic defoliation caused by insect outbreaks (larch budmoth) we cross-validated our δ¹⁸O_TRC time series against a dataset of known insect outbreaks in the Engadin [98]. However, no influence could be detected which is in accordance to findings by Kress et al. [99] who successfully demonstrated that insect-defoliation is not affecting δ¹⁸O_TRC variations in European larch.

2.3. Evaluation of Climate-Proxy Relationships

For evaluating climate-proxy relationships, monthly climate data from the two neighboring meteorological stations in Samedan and Segl-Maria were used (Federal Office of Meteorology and Climatology MeteoSwiss, Figure 1). The long-term climate station data was quality checked and homogenised by a proven method developed by MeteoSwiss [100,101]. To explore the climate-proxy relationships between the δ¹⁸O_TRC chronologies and standard climate parameters (precipitation, temperature), Pearson’s correlation coefficients were calculated for the common period from 1864 to 2011. In addition, data on cloud cover fraction available for Segl-Maria (since 1864) and for Samedan (since 1980) meteorological stations were used. To investigate the temporal stability of the obtained climate-proxy relationships, we performed moving window correlation analyses applying a 30-year moving window with an offset of one year. To study possible impacts of drought on our isotope series, the Standardized Precipitation-Evaporation Index (SPEI; [102]) was calculated with the R-package “SPEI” [103] using data from the Samedan climate station.

Possible influences of large-scale circulation patterns on our isotope time series were tested by calculating spatial correlations with gridded climate datasets derived from the Climate Research Unit (CRU v 4.04, [104]). Since the CRU data is lacking more complex parameters such as geopotential height fields or incoming solar radiation, we additionally used respective datasets from the Twentieth Century Reanalysis project (20CRv3 [105]). To ensure statistical comparability between the CRU and 20CRv3 data, we truncated both datasets to a common investigation period ranging from 1901 to 2011.

For the investigation and validation of a possible signal transfer of source water/atmospheric precipitation δ¹⁸O into the trees’ tissue, monthly precipitation data from the nearest GNIP (Global Network of Isotopes in Precipitation) station in Pontresina (Figure 1) were used (https://www.iaea.org/services/networks/gnip, accessed on 28 June 2021). Since the common observational period between the Pontresina GNIP station and our δ¹⁸O_TRC time series is with only 17 years relatively short, the nearest available long-term recording GNIP station in Locarno (46.17 N, 8.79 E, recording since 1973) was additionally considered.

3. Results

3.1. Time Series Characteristics of the Val Bever δ¹⁸O_TRC Site Chronology

The resulting eight individual L. decidua δ¹⁸O_TRC time series at the sample site demonstrate a high level of coherence as indicated by the parameters sign test (Gleichläufigkeit: GLK = 70%) and inter-series correlation coefficient (IC = 0.54;

Table 1. Time series statistics of the individual δ¹⁸O_TRC chronologies. Abbreviations: GLK (Gleichläufigkeit), RBar (inter-tree correlation coefficient), IC (Inter-series correlation coefficient), AC1 (first order autocorrelation), SNR (signal-to-noise ratio) and SD (standard deviation).

δ18OTRC	Period	GLK	RBar	IC	AC1	SNR	SD
MEAN-δ18OTRC	1830–2011	0.70	0.41	0.54	0.26	4.90	0.90
PF-δ18OTRC	1830–2011	0.61	0.37	0.36	0.22	1.78	0.93
nPF-δ18OTRC	1830–2011	0.69	0.47	0.53	0.24	2.55	1.02

). This is further underlined by the calculated bootstrapped range of uncertainty for the resulting site δ¹⁸O_TRC chronology for the Val Bever (hereafter referred to as MEAN-δ¹⁸O_TRC; Figure 3A). Year-to-year variations are highly pronounced for both the individual isotope series (standard deviation SD = 1.39‰) and for the MEAN-δ¹⁸O_TRC chronology (SD = 0.9). The MEAN-δ¹⁸O_TRC time series indicates a significant long-term trend with 0.18‰ per half century towards more positive δ¹⁸O values (r = 0.27, p < 0.01; Figure 3A). In order to emphasise short-term variations of the chronology, we calculated the inter-tree correlation coefficient (RBar) and the expressed population signal (EPS) within a representative window length of 20 years (Figure 3A).

Before 1860, a low sample replication expectably causes low RBar and EPS values. Since 1890, the EPS exceeds—with the exception of a relatively short period between 1880 and 1890—the accepted threshold of 0.85 [106]. Hence, our site chronology confirms a reliable quality for the common period to perform sound climate-proxy analyses with nearby local meteorological stations (Figure 3B).

3.2. Evaluation of Climate-Proxy Relationships with the δ¹⁸O_TRC Site Chronology

The evaluation on the influence of temperature and precipitation on the MEAN- δ¹⁸O_TRC site chronology during the common period 1864–2011 shows for both climate stations quite similar results (Figure 3B). For the current growing season, mainly summer temperatures reveal significant positive influences, with the highest correlations in August at Segl-Maria (r_Aug = 0.22, p < 0.01). Correlations between the MEAN-δ¹⁸O_TRC and precipitation are almost opposite, featuring the most significant negative correlation during summer (July) for Samedan station (r_Jul = −0.25, p < 0.01). Closely associated are results between MEAN-δ¹⁸O_TRC and cloud cover data from the Segl-Maria climate station. Highly significant negative correlations occur for the peak summer season from June to August (r_JJA = −0.38, p < 0.01). Even though the first order autocorrelation of the MEAN-δ¹⁸O_TRC series is rather low (AC1 = 0.26), significant correlations are recognisable with climate conditions during the previous year. For temperature, significant correlations at both stations are apparent during previous years’ October and December (e.g., Samedan: r_pOct = 0.22, p < 0.01; Segl-Maria: r_pDec = 0.22, p < 0.01). For precipitation, the analyses with previous years’ data reveal a significant impact of previous September rainfall (r_pSep = −0.17, p < 0.05).

3.3. General Characteristics of the Individual δ¹⁸O_TRC Chronologies

Figure 4 illustrates the δ¹⁸O_TRC time series differentiated into L. decidua trees growing on permafrost (hereafter: PF-δ¹⁸O_TRC, Figure 4B in blue) and outside permafrost lenses (hereafter: nPF-δ¹⁸O_TRC, Figure 4B in red). To ensure a better comparability of possible differences between the two series, ordinate axes are shifted against each other with an offset of 2.5‰. For both settings, the respective tree individuals show high internal synchronicity as indicated by the parameters bootstrapped chronology uncertainties, sign test (GLK) and inter-series correlation (IC) (Figure 4B; Table 1). However, GLK, RBar and IC consistently reveal a stronger common signal for those trees growing outside the permafrost lenses (nPF-δ¹⁸O_TRC). This finding is further corroborated by a weaker signal-to-noise ratio (SNR, Table 1) of the PF-δ¹⁸O_TRC time series, indicating an attenuated common signal between the trees. Year-to-year variations of both series are highly pronounced with an average variation of 1‰. In selected years, even maximum values of more than 5‰ can occur. Consequently, the nPF-δ¹⁸O_TRC time series exhibits a relatively high standard deviation (SD~1‰, Table 1).

Overall, the two δ¹⁸O_TRC -series are significantly correlated (r = 0.76, p < 0.01). However, their absolute annual deviations display pronounced discrepancies of up to several per mille for distinct years (e.g., 1849 or 1999) (Figure 4C). Both chronologies indicate comparably low first order autocorrelations (AC1 = 0.22 and 0.24). The non-permafrost tree individuals reveal a significant positive long-term trend of 0.22‰ (r = 0.23, p < 0.01) per half a century. In contrast, such a trend cannot be detected for trees growing on permafrost. The respective PF-δ¹⁸O_TRC time series only exhibits a slightly positive trend of 0.13‰ per half a century (p > 0.05).

3.4. Climatic Imprints Within δ¹⁸O_TRC Time Series in Permafrost and Non-Permafrost Terrains

Imprints of precipitation, temperature and cloud cover derived from the Segl-Maria meteorological station on our δ¹⁸O_TRC time series are illustrated in Figure 5. An equivalent figure displaying the results with the meteorological station of Samedan is provided as Supplementary Figure S1. It is recognisable that trees located on ice lenses (PF-δ¹⁸O_TRC) tend to show weaker correlation patterns than the permafrost-free tree stands (nPF-δ¹⁸O_TRC) or than the MEAN-δ¹⁸O_TRC time series. In particular, temperature and precipitation conditions during the summer months, which demonstrably revealed having a strong influence on the MEAN δ¹⁸O_TRC series (Figure 3B), are in comparison verifiably dampened (Figure 5A,B). Results of 30-year moving correlations with the PF-δ¹⁸O_TRC further corroborate this with notably weak correlations for temperature and precipitation (p > 0.05, Figure 5).

Contrasting results are obtained applying moving correlation analysis with the permafrost-free time series (nPF-δ¹⁸O_TRC). In general, highly significant correlation patterns are apparent (Figure 5A,B). Results on the climate sensitivity of the MEAN-δ¹⁸O_TRC illustrate its intermediate position between permafrost-free and permafrost-affected tree stands. Except for the overall robust and homogeneous correlation patterns observed with cloud cover, for which the MEAN-δ¹⁸O_TRC is slightly enhanced, the climate imprint of temperature and precipitation on the MEAN-δ¹⁸O_TRC is consistently less prominent as for the permafrost-free terrain (nPF-δ¹⁸O_TRC). In particular, the summer precipitation signals (July and JJAS) are significantly attenuated within the MEAN-δ¹⁸O_TRC. Most reasonably, the less significant correlation values of the permafrost-affected trees highly bias the stronger climate signal reflected by the permafrost-free tree stand. This implies that a loss in climate signal strength can arise from simply averaging both tree stands without prior knowledge of differing subsurface conditions.

Furthermore, the 30-year moving correlation patterns with climate station data (1864–2011) indicate an instable climate-proxy relationship for all three δ¹⁸O_TRC time series. For summer temperatures (Figure 5B), significant correlations (p < 0.05) are absent within the first decades of the time series until approximately 1880 (note that the abscissas axis in Figure 5 and Figure S1 represents the mid-year of the 30 years moving window). Roughly centered at 1930 and 1945 correlation values with temperatures also did not reach the level of significance. However, the most striking feature is the almost total loss of the climate-proxy relationships since 1975. Simultaneously, precipitation no longer reveals any significant influence on variations in δ¹⁸O_TRC during the recent four decades. An identical pattern for all three series of this study can also be found for the analyses with the Samedan climate station (Figure S1).

Additional calculations with Standardized Precipitation-Evaporation Index data (SPEI) show neither an increase in the level of significance nor in the correlation coefficients (Figure S2) compared to standard climate station data (Figure 5 and Figure S1). For SPEI-indices, highest correlations were obtained for the permafrost-free individuals during the summer season from July to August (nPF-SPEI_JA; r = −0.3, p < 0.01) showing a comparably low sensitivity to droughts which is rather expectable for the northern exposition of our study site. Nevertheless, the results with SPEI further corroborate the stated results that permafrost-unaffected trees consistently provide better imprints of climate signals compared to permafrost-affected individuals.

Large-scale climate representatively analyses of the permafrost and non-permafrost δ¹⁸O_TRC series with gridded CRU 4.04 climate parameters (precipitation, temperature, cloud cover) are illustrated in Figure 6. An identical figure for the MEAN-δ¹⁸O_TRC series is given in Supplementary Figure S3. The obtained results in Figure 6 further confirm the previously detected finding of an attenuated climate signal in permafrost-affected trees (PF-δ¹⁸O_TRC). Spatial analyses with the PF-δ¹⁸O_TRC series demonstrably reveal significantly weaker spatial correlation patterns for the entire observation period (1901–2011) as well as for three evenly distributed subperiods. The observed signal decrease within the climate-proxy relationships during recent decades (Figure 5 and Figure S1) is also continuously evident for spatial correlations with CRU 4.04 data (period 1973–2011; Figure 6 and Figure S3). Furthermore, the PF-δ¹⁸O_TRC time series reveals a generally weaker signal compared to the nPF-δ¹⁸O_TRC series. The spatial correlation patterns for the MEAN-δ¹⁸O_TRC chronology again represents an intermediate position between the well and weakly correlated time series of nPF-δ¹⁸O_TRC and PF-δ¹⁸O_TRC, respectively (Figure S3).

4. Discussion

The evaluation of climate-proxy relationships between our three individual isotope time series from Val Bever resembles related correlation patterns being observed in studies on tree-ring δ¹⁸O_TRC from other Alpine regions and in different tree species [26,107,108]. In these studies, the strongest climate-proxy correlations were found for the summer season (July to August). The authors postulate that high summer temperatures result in corresponding plant physiological responses to reduce evaporative water loss leading to an increase in δ¹⁸O_TRC. In good accordance with our findings, their sampling locations demonstrate a general higher sensitivity towards variations in temperature than precipitation (Figure 5 and Figure 6, Supplementary Materials Figures S1 and S2; [107]). Results for monthly and seasonal correlation values are also in a comparable range to the observations of these studies. However, annual climate averages/sums (January to December/J-D, Figure 3) do not reveal a significant influence on all δ¹⁸O_TRC series from the Val Bever.

Alpine oxygen chronologies represent well long-term climate variations [107]. Therefore, we assessed the inter-year variability (high-frequency variations) and decadal temperature variability (low-frequency variations) by calculating moving averages of 10 years. Mean annual (Jan–Dec; J-D) and summer temperatures (Jul–Aug; JA), recorded at the Segl-Maria and Samedan meteorological stations, were merged into a regional composite and correlated in a 10-years moving average with the three individual Val Bever isotope series (

Table 2. Correlation coefficients between 10-years moving averages of Val Bever isotope time series and 10-years moving averages of temperatures (as a composite from Segl-Maria and Samedan climate stations 1864–2011). Bold values indicate significant correlations, whereas the 95 and 99% level of significance are labelled by one or two asterisks, respectively.

Temperature	MEAN-δ18OTRC	PF-δ18OTRC	nPF-δ18OTRC
Mean annual	0.41 *	0.16	0.63 **
July-August	0.49 *	0.30	0.62 **

). The most striking results are the missing correlation patterns observed for the permafrost-affected tree stand (PF-δ¹⁸O_TRC). Significant imprints of low-frequency temperature variations are missing when accounting for mean annual temperatures (r_J-D = 0.16, p > 0.05) or summer temperatures (r_JA = 0.30; p > 0.05). Consistent with previous findings (Section 3.4), the permafrost-free tree stand (nPF-δ¹⁸O_TRC) reveals, in contrast, the most robust sensitivity towards decadal temperature variations (r_J-D = 0.63/r_JA = 0.62; p < 0.01). Again, the MEAN-δ¹⁸O_TRC resembles a climate sensitivity intermediate between the permafrost-free and permafrost-affected δ¹⁸O_TRC (Table 2).

Recent studies demonstrated that low-frequency temperature variations stored within δ¹⁸O_TRC series are a valuable indicator of large-scale circulation variabilities [20,21]. Therefore, the significant spatial correlations which extend far beyond the Alpine region (Figure 6 and Supplementary Materials Figure S2) can be regarded as highly reasonable. This is in line with the study by Saurer et al. [107], who observed imprints of decadal large-scale temperature variations on δ¹⁸O_TRC across the Alps for various tree species. Within a subsequent analysis [108], extreme positive (negative) anomalies in the δ¹⁸O_TRC time series from the Alps (>1‰) were successfully linked to positive (negative) anomalies of geopotential height fields (500 mbar) and anticyclonic synoptic weather types during summer (July–August). For δ¹⁸O_TRC, variations in different tree species from several sites across Switzerland Saurer et al. [108] determined highly significant, common supra-regional signals identified as strong influences of large-scale climate dynamics. Our series also reveal significant correlations with their results suggesting a common forcing. For example, our MEAN-δ¹⁸O_TRC is significantly correlated with their Central Alpine and North Alpine δ¹⁸O_TRC chronologies (r = 0.65, p < 0.01 and r = 0.57, p < 0.01, respectively). Therefore, we likewise tested the impact of large-scale atmospheric influences on our oxygen isotope series. Correspondingly, our Val Bever permafrost-free time series (nPF-δ¹⁸O_TRC) exhibit strong and significant (p < 0.01) positive spatial correlations with geopotential height fields (GpH) over broad areas across the temperate and Mediterranean parts of Europe. An additional correlation pattern of opposing sign is located in northeastern Europe (Figure 7A). This dipole-like pattern between the mid- and northern latitudes strongly influences the isotopic composition of precipitation in Central Europe [108]. In contrast, the PF-δ¹⁸O_TRC series reveals no regional representatively direct at the Val Bever sampling location (Figure 7A). The MEAN-δ¹⁸O_TRC correlation pattern with the 850 GpH in turn represents an intermediate between the permafrost-free and permafrost-affected tree-stand (not shown). Anticyclonic pressure patterns during July to August lead to increased cloud dispersal accompanied by increased downward solar radiation reaching the surface (r_JA > 0.5, p < 0.01) for large parts of Central Europe, including the Alpine region (not shown). As expected from the negative relationships between all three Val Bever δ¹⁸O_TRC chronologies and cloud cover or precipitation (Figure 5 and Figure 6), the correlation patterns concerning downward solar radiation are reverse (Figure 7B). Averaging the four nearby grid cells of the 20CRv3 downward solar radiation reveal a mean correlation coefficient of r_JA = 0.43 (p < 0.01) and r_JA = 0.27 (p < 0.01) with the permafrost-free (nPF-δ¹⁸O_TRC) and permafrost-affected (PF-δ¹⁸O_TRC) tree stand, respectively.

An increase in downward solar radiation intensifies the warming of the near-surface air temperature (r_JA = 0.72, p < 0.01 for the four adjacent 20 CRv3 grid cells to the sampling location) and leads to higher leaf temperatures. Both effects will result in an increase in leaf water enrichment and therefore higher δ¹⁸O_TRC values [26,29,108]. Additionally, δ¹⁸O in precipitation is positively correlated with temperatures [109,110]. Due to the small distance between permafrost-free and -affected tree stands, featuring identical exposition and inclination, no differences within the individual climate elements or isotopic composition of perceptible water can be expected. The consistent attenuation or even lack of climate signals observed for the permafrost-affected tree stand (PF-δ¹⁸O_TRC) assumes that for these individuals mainly source water variations constraint the climatic signals in δ¹⁸O_TRC.

A comparison between the individual Val Bever δ¹⁸O_TRC series and δ¹⁸O in precipitation derived from Pontresina GNIP station (1994–2011; Figure S4) further underline this assumption. Trees growing on permafrost-free terrain are highly dependent on the mean isotopic composition of precipitation during June to August (r_JJA = 0.76, p < 0.01) and July to August (r_JA = 0.58, p < 0.05). In contrast, trees growing on permafrost lenses reveal a weak coherence with varying isotopic composition in precipitation at Pontresina (r_JJA = 0.42, r_JA= 0.36, p > 0.05).

A similar, albeit slightly weakened correlation pattern, is evident for the comparison with the more distant but longer recording series of the Locarno GNIP station (1973–2011, Figure S4). Here, the nPF-δ¹⁸O_TRC series reveals a stronger dependency of δ¹⁸O on precipitation (r_JJA = 0.48, p < 0.01) while the influence on PF-δ¹⁸O_TRC series is less significant (r_JJA = 0.40, p < 0.05). As the trees’ source water represents a composite isotopic signal of precipitation and soil water, the latter exerts a significant mitigating impact on the climate-proxy relationships. This can be attributed to the more complex soil hydrology of permafrost-affected sites, e.g., differing seasonal water pools accompanied by temperature-dependent lag times and active layer dynamics. Such impacts have already been observed in the zone of continuous permafrost of the high latitudes, whereby the soil water isotope composition during dry and warm summers was significantly below the predicted level due to the thawing of more depleted soil ice within the permafrost bodies [45,46,49,111]. In contradiction to an overall warming climate and the expected increase in δ¹⁸O_TRC, such impact of a thawing active layer or melting processes of the permafrost lenses can introduce even decreasing trends within δ¹⁸O_TRC time series [45,111] and highly overprint climate signals. This can explain the lack of a significant long-term trend towards more positive δ¹⁸O values in the PF-δ¹⁸O_TRC time series compared to the increasing trend of the nPF-δ¹⁸O_TRC (Figure 4).

Such an increasing trend in δ¹⁸O_TRC might be attributed to an enhanced temperature-driven leaf water enrichment and to less depleted δ¹⁸O in precipitation. The latter is evident at the Locarno and Pontresina GNIP stations, featuring positive δ¹⁸O summer trends (June-August) according to 0.28‰ (p < 0.05) and 0.35‰ (p > 0.05) per decade during the recording periods (1973–2019 and 1994–2019, respectively). In addition to changing circulation systems and, therefore, alternating sites of evaporation and/or moisture sources, the temperature dependency of δ¹⁸O in precipitation is decisive [112,113,114]. The two climate stations Segl-Maria and Samedan and their regional mean reveal marked warming trends in both the annual and the seasonal averages (Figure 8). Relative to the climate normal period (1961–1990), spring and summer temperatures increased by more than 2 °C since the mid-1970s. In conjunction with the strong warming trend during the recent decades, an enhanced signal loss in the climate-proxy relationships is observed for precipitation and temperature at Val Bever (Figure 5; Figure S1 and spatial correlations Figure 6 and Figure S3). This temporal instability in the climate-proxy relationship (loss of significance or changing sign in the correlation coefficient) also has been frequently observed for tree-ring series [115,116,117] as well as for tree-ring isotope series [12,108,118]) and is commonly referred to as proxy divergence. For different sites and species across the Alps, a temporal instability in δ¹⁸O_TRC series can be observed (e.g., [107]) and attributed to changing circulation patterns [12,119]. In addition, varying water accessibility to deeper and more depleted soil water pools during warmer and dryer years is postulated [118]. However, determining the influence of the latter on our isotope series is impossible since the trees are growing on a blocky debris cone that prevents any deep soil profiling.

In summary, this study shows that caution is advised when investigating tree-ring δ¹⁸O_TRC from sub-alpine sites in high mountain areas where the influence of permafrost cannot be completely excluded. This refers especially to less radiation-exposed slopes and exposures, where permafrost is expected to occur. Demonstrably, this also includes areas and settings that are located below known permafrost distribution patterns. As shown, the influence of permafrost lenses can be particularly significant for microscale subsoil conditions that it highly biases the climate signal incorporated in δ¹⁸O_TRC at a sample site. Therefore, it is advisable for such areas to consistently analyse trees individually to enable a careful evaluation of related δ¹⁸O_TRC time-series. Such detected offsets in δ¹⁸O_TRC can consequently contribute to the recent discussion on divergence in δ¹⁸O_TRC time series across the globe, since permafrost dynamics can even influence small-scale variations of hydrological conditions.

5. Conclusions

This study represents the first quantitative evaluation in isotope-based dendroclimatology, focusing on the impact of permafrost on δ¹⁸O_TRC series at sub-alpine sites. Especially in high-mountain areas, the occurrence of permafrost and its related influence on soil water status is still a widely neglected factor that can highly influence the δ¹⁸O_TRC in trees. We therefore analysed a total of eight Larix decidua Mill. trees from a subalpine tree stand in the Val Bever/Switzerland. The investigated tree stand is located several hundred meters below the tree line and therefore below the proposed lower limit of recent alpine permafrost occurrence in a northern exposition on a scree cone. Small-scale alpine permafrost lenses have been proven to be present at our sample site based on geophysical soundings. Our sampling approach is based on an evenly distributed sampling of permafrost-free and permafrost-affected tree individuals. For the final δ¹⁸O_TRC analyses, all sampled trees were analysed individually on an annual basis for a period of the last 181 years.

The displayed results indicate a substantial influence of permafrost on δ¹⁸O_TRC variations at our site. A thorough investigation of climate-proxy relationship between permafrost-affected (PF-δ¹⁸O_TRC) and permafrost free (nPF-δ¹⁸O_TRC) tree stands reveal striking differences in their individual responses. The nPF-δ¹⁸O_TRC reveal high positive correlations with summer temperatures, geopotential height fields and associated downward solar radiation. High precipitation and cloud cover during the summer season are vice versa detrimental for increased δ¹⁸O_TRC values. For the permafrost-affected trees (PF-δ¹⁸O_TRC), climate fingerprints of all investigated climate parameters are also present but highly biased. However, in an extremely attenuated pattern, the displayed climate-proxy relationship often does not pass the 95% level of significance. This frequently equates to the (statistical) loss of the common signal between our proxy and the climate parameter targeted for climate reconstruction. Since both investigated tree stands are generally subject to the same climate influences, this observed divergence in the climatic sensitivity can be attributed to varying trees’ source water that constraints a coherent response.

As the occurrence of permafrost was apparently neither expectable nor directly visible, this suggests that the sampled trees in this study are by first sight mainly exposed to the same external climate forcing. However, this may initially lead to the use of common pre-analytical approaches such as averaging of individual δ¹⁸O_TRC series or sample pooling prior cellulose extraction. Especially sample pooling is inevitably applied as an established scientific approach in many δ¹⁸O_TRC studies. In particular, this is often necessary due to various constraints, e.g., during the establishment of multi-centennial/millennial δ¹⁸O_TRC time series from subalpine tree species featuring very narrow tree-rings, which frequently results in insufficient wood material for a subsequent a-cellulose extraction process [14,120,121]. However, this study highlights the importance of pre-screening possible permafrost occurrences at sub-alpine sample sites prior to data averaging or sample pooling. In neglecting such background information in high-mountain ecosystems, a significant loss of climate signal can be imminent if the obtained mean time series includes permafrost- and non-permafrost affected δ¹⁸O_TRC series prone to individually differing trends or offsets. As a result, the verifiably high suitability of δ¹⁸O_TRC as a parameter recording past (hydro-)climate variability may be unnecessarily hampered.

In summary, our studies show that indirect climatically induced changes in hydrology can also peek through into the δ¹⁸O_TRC values when analysing trees from high altitude sites. Therefore, it is advisable for such settings to monitor the individual δ¹⁸O_TRC series for temporary as well as general offsets and to relate these to possible permafrost dynamics. This is particularly important due to the recently observed global permafrost thawing processes across the globe, which demonstrably highly alters regional hydrology and soil water dynamics. However, as our study shows, the climate signal incorporated in the δ¹⁸O_TRC can be significantly overprinted.