Climate signals in stable carbon and hydrogen isotopes of lignin methoxy groups from southern German beech trees

. Stable hydrogen and carbon isotope ratios of wood lignin methoxy groups ( δ 13 C LM and δ 2 H LM values) have been shown to be reliable proxies of past temperature variations. Previous studies showed that δ 2 H LM values even work in temperate environments where classical tree-ring width and maximum latewood density measurements are less skilful. Here, we analyse annually resolved δ 13 C LM values from 1916-2015 of four beech trees ( Fagus sylvatica ) from a temperate site near Hohenpeißenberg in southern Germany and compare these data with regional to continental scale climate observations. Initial 15 δ 13 C LM values were corrected for the Suess effect (a decrease of δ 13 C in atmospheric CO 2 ) and physiological tree responses to increasing atmospheric CO 2 concentrations considering a range of published discrimination factors. The calibration of δ 13 C LM chronologies against instrumental data reveals highest correlations with regional summer (r = 0.68) and mean annual temperatures (r = 0.66), as well as previous-year September to current-year August temperatures (r = 0.61), all calculated from 1916-2015 and reaching p < 0.001. Additional calibration trials using detrended δ 13 C LM values and climate data, to constrain 20 effects of autocorrelation on significance levels, returned r summer = 0.46 (p < 0.001), r annual = 0.25 (p < 0.05) and r prev.Sep-Aug = 0.18 (p > 0.05). The new δ 13 C LM chronologies were finally compared with previously produced δ 2 H LM values of the same trees to evaluate the additional gain of assessing past climate variability using a dual-isotope approach. Compared to δ 13 C LM , δ 2 H LM values correlates substantially stronger with large-scale temperatures averaged over western Europe (r prev.Sep-Aug = 0.69), whereas only weak and mainly insignificant correlations are obtained between precipitation and both isotope chronologies 25 ( δ 13 C LM and δ 2 H LM values). Our results indicate great potential of using δ 13 C LM values from temperate environments as a proxy for local temperatures, and in combination with δ 2 H LM values, to assess regional to sub-continental scale temperature patterns.

these tree rings. Thus growth specific parameters, such as ring width, maximum latewood density, and the isotopic composition of bulk wood, cellulose, or lignin have proven to be climate sensitive (Daux et al., 2011;Esper, 2000;Esper et al., 2015;Hafner et al., 2011;Konter et al., 2014;Kress et al., 2010;McCarroll and Loader, 2004;Reynolds-Henne et al., 2007;Treydte et al., 2001;Wang et al., 2011). In addition  and Keppler et al. (2007) showed that stable hydrogen and carbon 35 isotope ratios of wood lignin methoxy groups (δ 2 HLM and δ 13 CLM values) have great potential to be applied for paleoclimate reconstructions. For a more detailed overview of its applications in paleoclimate research we refer to previous studies by Gori et al., 2013;Greule et al., 2021;Hepp et al., 2017;Lee et al., 2019;Van Raden et al., 2013;Wang et al., 2020). The isotopic ratios of these specific chemical moieties (-OCH3 groups) of lignin remain unchanged throughout the lifetime of a tree and thus reflect the methoxy isotopic composition at the time of its biochemical formation (Greule et al., 40 2008;Keppler et al., 2007). While the traditional methods for analyzing stable isotope ratios of a-cellulose or nitrate cellulose require time-consuming preparation, δ 13 CLM and δ 2 HLM values can readily be measured as iodomethane (CH3I) upon treatment with hydroiodic acid, providing a fast and straightforward preparation (Greule et al., 2008). Furthermore, the removal of water from bulk wood samples is not necessary as it does not affect the isotope analysis (Greule et al., 2008(Greule et al., , 2009. It is also advantageous for isotope analysis that methoxy groups are highly abundant in tree rings as wood contains around 25-30 % of 45 lignin and the proportion of methoxy groups in lignin (on a carbon basis) may reach 20 % (Keppler et al., 2007). Therefore, small sample amounts of only 1 mg and 2.5 mg of bulk wood (milled or in pieces) are required for reliable measurements of δ 13 CLM and δ 2 HLM values (Greule et al., 2008(Greule et al., , 2009). Finally, the analytical procedure for measuring δ 2 HLM values was considerably improved by the availability of new reference materials that are in full accordance with the requirements of normalizing stable isotope measurements . 50 Most previous methoxy based research have applied δ 2 HLM values for climate studies (Anhäuser et al., 2017a(Anhäuser et al., , 2017bGreule et al., 2021;Keppler et al., 2007;Riechelmann et al., 2017;Wang et al., 2020). In general, the hydrogen isotopic composition of trees is controlled by its source water, and hence the stable isotopes composition of local precipitation (Sternberg, 2009;Tang et al., 2000). Therefore, the temperature dominated signal in δ 2 Hprecipitation (Dansgaard, 1964) is reflected in δ 2 HLM values as has been demonstrated for mid-latitude sites (Anhäuser et al., 2017a;Greule et al., 2021). A highly 55 significant correlation was documented between δ 2 HLM values and mean annual temperatures (MAT), whereas 'shifted' MAT (defined as previous September to recent August) showed the highest coefficients with r = 0.56 . Wang et al. (2020) found significant correlations between δ 2 HLM values and April-August temperatures (r = 0.58 to 0.7) for two coniferous species (Larix gmelinii, larch and Pinus sylvestris var. mongolica, pine) from a permafrost forest in northeastern China. Even higher correlations were reported between beech trees from a low elevation site in southern Germany and west 60 European large-scale temperature changes at r = 0.72 .
Although there exist some studies (Gori et al., 2013;Mischel et al., 2015;Riechelmann et al., 2016;Wang et al., 2020), less attention has been given to the climate sensitivity of δ 13 CLM values. The carbon of each annual tree ring has its origin in the atmospheric CO2. Thus the carbon isotope composition in trees mainly consists of the isotopic values of atmospheric CO2 (δ 13 Catmos), the concentration of atmospheric CO2, and the diffusion and fractionation of δ 13 C through stomatal pores (-4.4 mUr) and carbon fixation via the photosynthetic enzyme Rubisco (-27 mUr) (Francey and Farquhar, 1982) (Please note, that we follow the suggestion by Brand and Coplen (2012) and express isotope δ-values in milli-Urey [mUr] (after H.C. Urey, 1948) instead of per mil [‰]). The carbon isotopic composition in trees can be expressed as the deviation between δ 13 Catmos and the isotopic discrimination of 13 C during carbon diffusion and fixation by plants (Eq. 1) (Keeling et al., 2017): (1) 70 Where a expresses the fractionation by diffusion through the stomata and b describes the fractionation to carboxylation, while ci reflects the inner leaf CO2 concentration and ca ambient air CO2 concentrations (Francey and Farquhar, 1982). The last two terms of Eq. 1 represent the mesophyll and photorespiration effects with am the fractionation by dissolution and diffusion from the intercellular air spaces to the sites of carboxylation in the chloroplasts, A the leaf-level gross photosynthesis, gi the mesophyll conductance, f the discrimination due to photorespiration, and the CO2 compensation point in the absence of day 75 respiration (Cernusak et al., 2013;Farquhar et al., 1982;Keeling et al., 2017;Seibt et al., 2008). These effects are normally neglected but are necessary for understanding the discrimination changes of 13 C due to increasing CO2 concentration (Seibt et al., 2008). The terms of mesophyll and photorespiration are both negative and their absolute magnitude decrease with increasing ca, resulting in increasing discrimination with rising CO2 concentration (Keeling et al., 2017).
Since the fractionations due to stomatal conductance and Rubisco (a and b) are considered constant and the terms of mesophyll 80 and photorespiration are normally negligible, the discrimination of 13 C is mainly controlled by the ci/ca ratio. If ci increases, stomatal control is higher than the rate of photosynthesis. The dominance of stomatal control and photosynthesis rate thereby depends on various environmental factors, including temperature, air humidity, precipitation amount, and seasonality (McCarroll and Loader, 2004). Furthermore, the isotopic fractionation between the photosynthate and cellulose or lignin is assumed to be small (Francey and Farquhar, 1982). Consequently, environmental factors that influence the ci/ca ratio are also 85 considered to additionally control δ 13 CLM values.
The few studies that applied δ 13 CLM values of tree rings already demonstrated a relationship with climate parameters. Wang et al. (2020) and Riechelmann et al. (2016) documented highly significant correlations between δ 13 CLM values and mean summer temperatures in high elevation environments. Wang et al. (2020) observed highest correlations with April to August temperatures (r = 0.64) and Riechelmann et al. (2016) with June to August temperatures (r = 0.66). In addition, Gori et al. 90 (2013) report correlations with spring and annual mean temperatures, and Mischel et al. (2015) with August maximum temperatures. However, in all previous studies, non-significant correlations (p > 0.05) were reported with precipitation.
In this study, we evaluate the applicability of δ 13 CLM values of trees as a paleoclimate proxy in temperate low elevation environments. Therefore, we measured annually resolved δ 13 CLM values of four Fagus sylvatica L. trees in southern Germany at Hohenpeißenberg and analyzed the climate sensitivity and non-climatic response (to atmospheric CO2 changes) of δ 13 CLM 95 values. Furthermore, to evaluate the potential of reconstructing past climate variability using dual-isotope approach, we revisit the δ 2 HLM values of the same trees provided by Anhäuser et al. (2020). However, these previous data were corrected according to new constraints regarding analytical issues of the isotope measurements of methoxy groups (Greule et al., 2021). Finally, https://doi.org/10.5194/cp-2021-135 Preprint. Discussion started: 20 October 2021 c Author(s) 2021. CC BY 4.0 License. the dual isotope methoxy measurements of Hohenpeißenberg tree rings were used to critically evaluate their potential as a proxy for regional to sub-continental scale temperature patterns. 100 2 Material and Methods

Study site
The study site is located close to the Hohenpeißenberg municipality in southern Germany, where tree samples were collected from the northeastern slope of the Hoher Peißenberg mountain (47° 48′ N, 11° 01′ E; altitude ~800 m). For a detailed map of the sampling site, we would like to refer the reader to the study by Anhäuser et al. (2020). The region is characterized by a 105 strong temperature increase, particularly since the 1980s, and insignificant precipitation trends (Fig. 1a). Annual mean temperatures range from 6.22 °C (1940) to 9.73 °C (2018) and precipitation totals from 788 mm (1943) to 1316 mm (1966).
The seasonal climate is characterized by a distinct precipitation peak in summer including 138 mm (period 1961-1990) in July ( Fig. 1b). For the determination of δ 13 CLM values the modified Zeisel method was used (Greule et al., 2009;Keppler et al., 2004Keppler et al., , 2007. 115 The method is based on the reaction between methyl ethers or esters and hydriodic acid (HI) to form iodomethane (Zeisel, https://doi.org/10.5194/cp-2021-135 Preprint. Discussion started: 20 October 2021 c Author(s) 2021. CC BY 4.0 License. 1885). In a 1.5 ml crimp glass vial, 250 µl HI (57 wt% aqueous solution, Acros (Thermo Fisher Scientific)) were added to the 1-10 mg annually dissected tree rings. The vials were sealed with crimp caps and heated for 30 minutes at 130 °C. Samples were then equilibrated at room temperature (22 °C) for at least 30 minutes before an aliquot of headspace was injected into the gas chromatographcombustionisotope ratio mass spectrometry (GC-C-IRMS) analytical system. 10-90 µl of the headspace 120 were injected via an autosampler (A200S, CTC Analytics, Zwingen, Switzerland) with a split injection of 10:1 to the HP 6890 N gas chromatograph (Agilent, Santa Clara, USA). The gas chromatograph was fitted with a DB-5MS, Agilent J&W capillary column (length 30 m, internal diameter 0.25 mm, film thickness 0.5 µm) with an initial oven temperature of 50 °C for 2.9 minutes, ramp at 50 °C per minute to 110 °C. Helium was used as carrier gas at a constant flow rate of 1.8 ml min -1 . Using an oxidation reactor (ceramic tube (Al2O3), length 320 mm, internal diameter 0.5 mm) with Cu, Ni, and Pt wires inside (activated 125 by oxygen) and a reaction temperature of 960 °C, CH3I is oxidized to CO2. Before the CO2 flows through a GC Combustion III Interface (ThermoQuest Finnigan) into the isotope ratio mass spectrometer (253 Plus 10 kV IRMS, Thermo Fisher Scientific), the accrue water was removed through a semipermeable membrane (NAFION ® ). A tank of high purity carbon dioxide (carbon dioxide 4.5, Messer Griesheim, Frankfurt, Germany) was used as the monitoring gas. For all values, the delta (δ) notation is used, employing the term Urey (Ur, after H. C Urey, 1948) as the isotope delta value unit (Brand and Coplen, 130 2012). Hence, 1 mUr equates to 1 ‰.
The δ 13 CLM values were normalized considering a two-point calibration and two reference materials, potassium methyl sulfate (HUBG 2) and beech wood (HUBG 4) described by Greule et al. ( , 2020. δ 13 CLM values of HUBG 2 and HUBG 4 were calibrated against international isotope reference material (V-PDB) with an isotopic value of δ 13 CVPDB = +1.60 ± 0.12 mUr for HUBG2  and δ 13 CVPDB= -30.17 ± 0.13 mUr for HUBG4 . Before and after every sixth 135 measurement, HUBG 2 and HUBG 4 were measured alternately. The tree rings of the two cores of F1 were measured as triplicates (n=3). Differences between the triplicates were always less than 1 mUr with an average deviation of 0.08 mUr. The maximum differences between two individual cores of the same tree ranged from 1.54 for F1 to 3.26 mUr for F2. Since each tree is represented by the average of two cores, the variances between the triplicate measurements of F1 are marginal compared to the much larger differences of the two cores of the same tree. Therefore, to drastically reduce analytical costs further tree 140 rings from F2-F4 were analyzed by single measurements.

Correction of δ 13 CLM values for non-climatic environmental factors
Due to anthropogenic burning of fossil fuels, the atmospheric CO2 concentration is steadily increasing. Since fossil CO2 has a lighter carbon isotopic composition than the atmosphere, the δ 13 C values in atmospheric CO2 (δ 13 Catmos) show a prominent downwards trend. This so-called "Suess effect" (Keeling, 1979) describes a decrease in δ 13 Catmos value from -6.41 mUr in 1850 145 to a current value of -8.6 mUr in 2020. Consequently, leaf internal CO2 (ci) is already depleted in 13 C and even more 12 C can be assimilated in leaf sugars, yielding to more negative δ 13 CLM values. As this decline is a non-climate effect, the carbon isotopic composition of tree rings needs to be corrected by adding the differences for each year between the δ 13 Catmos and the pre-industrial (-6.41 mUr) to the measured δ 13 and Loader, 2004). Here, the δ 13 Catmos series was obtained from McCarroll and Loader (2004) and the Mauna Loa Observatory 150 (Keeling et al., 2005, https://scrippsco2.ucsd.edu/data/atmospheric_co2/mlo.html). Furthermore, the leaf internal 13 C discrimination increases with rising CO2 concentration, as the absolute magnitude of mesophyll and photorespiration decreases (Keeling et al., 2017). It is important to note that there is no pre-defined way to correct δ 13 C values of trees due to increasing CO2 concentrations. However, in our study the Suess effect corrected δ 13 CLM_S values were multiplied considering a correction factor per increasing CO2 parts per million by volume (ppmv) compared to the pre-industrial CO2 concentration. We used the 155 between tree species and locations and is itself a current and important field of study. We additionally detrended the δ 13 CLM data using 30-year cubic smoothing splines to emphasize high-frequency variations and evaluate the effects of autocorrelation in our analyses. The resulting δ 13 CLM_high-frequency data were compared with (30-year spline) detrended temperature data to ensure that significant correlations are not simply related to warming trend prevailing over the last 60 years (Fig. 1a).

Correction of δ 2 HLM values considering new reference material 165
The δ 2 HLM values of trees from Hohenpeißenberg presented by Anhäuser et al. (2020) were normalized using two CH3I reference standards. As CH3I is a different material compared to wood and the two CH3I reference standards did also not cover the entire range of the δ 2 HLM values of the samples (-295 to -224 mUr), this study applied the newly available reference material investigated and recommended by . Thus, previously measured δ 2 HLM values were corrected using the suggested equation of Greule et al. (2021) (Eq. 2). Accordingly, the corrected data shifts the previous δ 2 HLM series 170 to more positive values and the differences between previous and corrected δ 2 HLM series become larger with decreasing δ 2 HLM values.
Please note, that the previous δ 2 HLM chronology of Anhäuser et al. (2020) included nine cores. F1 was represented by three cores and F2-F4 by two cores. To have the same number of replicates of all four trees (n=2) one core from tree F1 was removed. 175

Climate data and statistics
The sensitivity of δ 13 CLM and δ 2 HLM chronologies to climate was assessed by comparisons the mean δ 13 CLM and δ 2 HLM anomalies as deviations from the 1961-1990 mean with monthly resolved temperatures and precipitation data from a nearby grid point at 47.75°N and 11.25°E as well as large-scale gridded temperatures using the latest CRU TS version 4.04 via the KNMI climate explorer (Harris et al., 2020;Trouet and Jan van Oldenborgh, 2013). For correlations with detrended spline. Correlations among the δ 13 CLM and δ 2 HLM tree-ring series and between climate parameters and isotope chronologies were calculated over the period 1916 to 2015 using Bravais Person (r). Temporal changes between proxies and climate parameters were assessed using 31-year running correlations. Here, p values < 0.05 were considered significant, and p < 0.001 highly significant. The reconstructions skills of δ 13 CLM and δ 13 CLM_high-frequency chronologies were assessed using Durbin-185 Watson statistics (DW) testing lag-1 autocorrelation in the linear model residuals, the reduction of error (RE), and coefficient of efficiency (CE) statistics. Any positive values of RE and CE are indicative of adequate skills in the reconstructions (Briffa et al., 1988;Cook et al., 1994). All data analyses, statistics, and graphs were calculated and plotted using the software Arstan, OriginPro 2021, and R.

δ 13 CLM values, correction for the Suess effect and physiological response
The δ 13 CLM values of the four tree series range from -32.66 to -26.02 mUr from 1916 to 2015 (Fig. 2a). The δ 13 CLM anomalies (deviations from the 1961-1990 mean) range from -3.18 to 3.34 mUr with a standard deviation σ = -0.05 ± 1.05 mUr (Fig. 2b). Applying additional corrections that account for physiological response due to increasing atmospheric CO2 concentrations, the δ 13 CLM values further increase to more positive values. This effect is particularly visible in the second half of the 20 th century until today and clearly depends on the value that has been used as the correction factor (Fig. 3 pink,

Climate sensitivities of δ 13 CLM chronologies
When comparing δ 13 CLM chronologies with climate data, we find positive correlations with regional temperatures and largely non-significant correlations with precipitation. The coefficients tend to increase when considering δ 13 CLM chronologies that were corrected for the Suess effect (δ 13 CLM_S) and physiological response due to increasing atmospheric CO2 concentrations (δ 13 CLM_T < δ 13 CLM_FE, < δ 13 CLM_RL) ( Fig. 4 and S1). Highest correlation coefficients were found between δ 13 CLM values and 225 summer (JJA) temperatures (δ 13 CLM_S: r = 0.52 to δ 13 CLM_RL: r = 0.68) (p < 0.001; the degrees of freedom were reduced, due to lag-1 autocorrelation), followed by MAT ranging from r = 0.42 to 0.66, and 'shifted' annual temperatures (previous September to August) ranging from r = 0.34 to 0.61. Among the mainly non-significant correlations with precipitation totals, the highest coefficient was identified with the just Suess effect corrected δ 13 CLM_S chronology and summer precipitation

235
Additional assessments of climate signals using 30-year high-pass filtered chronologies revealed that correlations with summer temperatures decrease to r = 0.46 but are still significant at p < 0.001. The correlation coefficient with MAT (Jan-Dec) is lower with r = 0.25 (p < 0.05) and non-significant with shifted MAT (Sepp-Aug) (r = 0.18, p > 0.05).
Since the ideal correction factor was determined by considering summer temperatures as the best fit climate parameter, further relationships with large-scale seasonal temperatures were modelled with δ 13 CLM_RL values (https://climexp.knmi.nl/start.cgi). 240 The patterns extend from southern Europe to middle Scandinavia and cover the UK and western Poland, Slovakia, and Hungary. Here, the highest correlations with r > 0.6 were found between the δ 13 CLM_RL values and summer temperatures in southern Germany, Austria, northern Italy, and northeastern Spain. Followed by somewhat lower correlations of r > 0.5 with summer temperatures in middle Germany, Switzerland, and France and with fall temperatures in northeastern Spain (Fig. 5,   S2). Correlations with winter and spring temperatures are always r < 0.5 (S2). Highest correlation between the 245 δ 13 CLM_high-frequency chronology and large-scale summer temperatures extend from southeastern Germany to middle France and cover Switzerland and northern Italy with r > 0.4.

Running correlations and transfer function
The local summer temperatures were modelled by δ 13 CLM_RL values using a linear regression model following Eq. 3. (4) 255 The regression model residuals ranging between -2.13 to 1.91 and show an increasing trend of 0.022 ± 0.002 over the past 100 years. Between 1916-1963 residuals are mainly negative, whereas since 1964 residuals are mainly positive (Fig. 6a,b). Furthermore, we calculated the Durbin-Watson (DW) statistic of the regression model residuals and reveal a weak DW value of 0.86 which indicates a strong positive autocorrelation (p < 0.001). If two series are autocorrelated, the effective sample size and thus the degrees of freedoms may be reduced. Since the significance of the correlation coefficient depends on the degrees 260 of freedoms, significant correlations may well be non-significant under the consideration of autocorrelation (Wigley et al., 1987).
To eliminate the effect of autocorrelation and to constrain high-frequency variations we employed δ 13 CLM_high-frequency indices for modelling summer temperatures using Eq. 4. The high-frequency chronology reveals a DW value of the regression residuals close to the optimum value of 2 (DW = 2.3) suggesting the robustness of the high-frequency signal. The residuals ranging between -0.04 and 0.07 and show no trend over the 1916-2015 period (Fig. 6c,d). 270 To determine the strength of the relationship between modelled and observed temperatures we calculated the reduction of error (RE) and coefficient of efficiency (CE) statistics after Cook et al. (1994).
When conducting a split calibration/verification on the high-frequency linear model, a rather weak temporal robustness is

δ 2 HLM chronology and climate signal
The δ 2 HLM values of the tree cores from Hohenpeißenberg previously presented by Anhäuser et al. (2020) were corrected using 290 the relationship provided by Greule et al. (2021) (see Materials and Methods). The revised data showed maximum and minimum δ 2 HLM values ranging from -274 to -221 mUr around a mean value of -246 ± 9 mUr (1σ standard deviation) and maximum and minimum δ 2 HLM anomalies from -12.3 to 19.4 with a mean value of 1.9 ± 6.4 mUr. Considering the chronologies of the four trees a highly significant inter-series correlation of r = 0.33 (p < 0.001) can be reported, whereby somewhat higher Rbar values are observed between single radii of the same tree ranging from 0.57 to 0.8. Figure 8 (solid black line) shows the 295 mean δ 2 HLM chronology of the four trees over the past 100 years. A linear increasing trend of 0.14 mUr year -1 is observed over the whole period from 1916-2015, but it is also obvious that the trend is more positive (0.38 mUr year -1 ) since 1970.  Highest r values between the corrected δ 2 HLM anomalies and temperature are recorded for the 'shifted' annual temperatures

Comparison of δ 13 CLM and δ 2 HLM chronologies
The δ 2 HLM chronology correlates negatively with the uncorrected δ 13 CLM chronology (r = -0.36), but the coefficients change to positive when comparing with the δ 13 CLM chronology that was corrected for the Suess effect (r = 0.29) and increase substantially with the chronology that was additionally corrected for the physiological response (δ 13 CLM_T: r = 0.52 to 315 δ 13 CLM_RL: r = 0.62). The driver for these changes is the common warming trend now inherent in both isotopic records. The correlations are incoherent over time, however considering the δ 13 CLM_RL chronology, lowest and non-correlations were found between 1940-1953, 1959-1964, and 1975-1981 (Fig. 10). The highest correlations were observed since 1980 when r values gradually increased from a mean of r = 0. 13 during 197513 during -198113 during to r = 0.6 during 199013 during -200013 during . Importantly during the 197513 during -1986 period, δ 2 HLM anomalies decrease while δ 13 CLM_RL anomalies constantly increase. 320 Comparing the potential application of δ 13 CLM and δ 2 HLM series as climate proxies, it can be shown that δ 13 CLM_RL values are more affected by regional temperature changes than δ 2 HLM values. The sum of correlation coefficients shown in Fig. 4 and 9 considering regional temperature, reached r = 7.91 for δ 13 CLM_RL values and only r = 7.12 for δ 2 HLM values. However, comparing δ 2 HLM values with large-scale western European 'shifted' annual temperatures, correlations are substantially higher at r = 0.69. For both isotopic elements, weak and mostly non-significant correlations are recorded with precipitation data.

δ 13 CLM values, corrections for non-climate related trends 330
The average inter-series correlations changes from the uncorrected (r = 0.23), Suess effect corrected (r = 0.16) to the maximum physiological response corrected (r = 0.55) δ 13 CLM values as they are influenced by different trend changes (Riechelmann et al., 2016). The correction procedure of the Suess effect removes the prevailing long-term decrease from δ 13 CLM values, and the correction for the physiological response produces a positive long-term trend.
Our results showed that the incremental application of corrections for non-climate related trends not only increases the inter-335 series correlation among trees but also improves the climate correlations. Correlation coefficients with temperature increase significantly by adding a correction factor that accounts for a strong CO2 response (correlations with δ 13 CLM_T < δ 13 CLM_FE < δ 13 CLM_RL). Hence, considering summer temperature as the target climate parameter, the ideal correction factor for Fagus sylvatica at this site is 0.032 mUr ppmv -1 as introduced by Riechelmann et al. (2016). This factor is also the strongest factor among published and suggests that δ 13 CLM values of trees growing in low elevation environments might contain a strong 340 response to increasing CO2 concentrations. The higher discrimination of 13 C by Fagus Sylvatica might be related to a lower https: //doi.org/10.5194/cp-2021-135 Preprint. Discussion started: 20 October 2021 c Author(s) 2021. CC BY 4.0 License.
water use efficiency of deciduous trees compared to evergreen conifer species (Riechelmann et al., 2016). It has been shown that the studies by Feng and Epstein (1995) and Treydte et al. (2009) mainly used evergreen conifer species and therefore required lower correction factors.

Climate sensitivity of δ 13 CLM values 345
The greatest climate response was documented between δ 13 CLM values and summer temperatures. Here, the strongest relationship was found with the ideal corrected δ 13 CLM_RL chronology (r = 0.68) (Fig. 4). Since correlations with seasonal largescale temperatures (western European surface temperatures) also showed the highest correlations with summer temperatures in the surrounding area of the study site (Fig. 5), δ 13 CLM_RL values seem to reflect local temperature variations better than largescale fluctuations. 350 The strong temperature and weak precipitation response indicates that δ 13 CLM ratios are predominantly controlled by the photosynthetic rate (McCarroll and Loader, 2004). Similar findings were reported in previous studies applying stable carbon isotopes of tree rings (Gagen et al., 2006;McCarroll and Pawellek, 2001;Riechelmann et al., 2016;Treydte et al., 2009).
However, most of these studies analyzed trees from cold, moist, high latitude or high elevation sites. Here, we now demonstrate that local temperatures also strongly influence the δ 13 CLM ratios of trees growing in mid-latitude low elevation environments 355 and thus, less extreme conditions. Summer temperature as the controlling factor of δ 13 CLM ratios at the Hohenpeißenberg site seems to be intensified in the last 50 years, since moving correlation coefficients between gridded instrumental and modelled records substantially increase after 1966 (Fig. 7). This change in climate sensitivity is not entirely controlled by the increasing temperature trend recorded over recent decades, as similar correlation changes are recorded when using 30-year high-pass filtered tree ring series 360 (δ 13 CLM_high-frequency) and gridded instrumental data. Similar inferences were reported in the study by Treydte et al. (2009). Here, climate correlations were calculated using high-frequency δ 13 CLM series to avoid biases from potentially non-climatic longterm trends.
The weak DW statistic and positive trend in the low-frequency regression model residuals is partly influenced by the value that has been set as the correction factor on δ 13 CLM values. A change in 13 C discrimination in early increasing CO2 365 concentrations might be stronger than the response after an initial adaptation time (Drake et al., 1997;Treydte et al., 2009).
This may lead to lower 13 C discrimination and a decreasing correction factor. Second, anthropogenic warming increases the drought stress of plants growing in mid-latitude sites. To protect from drought stress, plants reduce stomatal conductance, which might lead to an increase in δ 13 CLM values (Büntgen et al., 2021). The modern correction of δ 13 CLM values, especially in the last decade, may overcorrect the δ 13 CLM values and leads to overrated modelled temperatures. It is thus important to note 370 that there are additional uncertainties beyond the linearity of the physiological response to increasing CO2 concentrations.
The correlation coefficients between gridded instrumental and modelled summer temperatures were weak between 1935-1954 and 1939-1965 when using the δ 13 CLM_high-frequency and δ 13 CLM_RL values, respectively. During this early-to-mid 20 th century https://doi.org/10.5194/cp-2021-135 Preprint. Discussion started: 20 October 2021 c Author(s) 2021. CC BY 4.0 License. period the temperature sensitivity seems to be influenced by intra-series inconsistencies. This conclusion is supported by 375 contemporaneous low Rbar values (Fig. 11). Indeed, both chronologies correlate strongly (r = 0.74 for δ 13 CLM values and r = 0.54 for δ 13 CLM_high-frequency indices). In this study, however, we used two cores from just four trees each. Irregularities in only one or two trees, such as nutrient, water, or light availability, might massively influenced our mean chronology.
Additionally, the study side at Hohenpeißenberg is located in an area strongly influenced by human activities. Therefore, soil sealing or tree clearing may have affected the isotopic series and conceal the temperature signal. Saurer et al. (1997) suggested 380 that local effects such as ozone or the availability of water and nutrient could disturb the climate signal. For example, stressors usually increase the δ 13 C values and improvements in growing conditions can be expected to decrease δ 13 C values (Saurer et al., 1997).
Calculating the RE and CE statistics without the period of low inter-series correlation, the RE and CE values substantially increase indicating the potential of temperature reconstructions by δ 13 CLM values if tree ring series are not influenced by non-385 significant inter-series correlations. The temporal robustness could be improved in further studies by using more replicates on sample sites with less human activity.

δ 2 HLM values and climate response
The recently provided correction method by Greule et al. (2021) improves the δ 2 HLM series of Anhäuser et al. (2020) as a regional temperature proxy. We found a correlation of r = 0.58 with local 'shifted' annual temperatures (Fig. 9), which is 390 slightly higher than the observation of Anhäuser et al. (2020). In addition, we confirm the strong correlation of r = 0.69 with western European surface temperatures as shown by Anhäuser et al. (2020). The increasing long-term trend of the δ 2 HLM series clearly reflects the anthropogenic warming trend (slope of 0.14 mUr year -1 from 1916-2015). However, a much stronger https://doi.org/10.5194/cp-2021-135 Preprint. Discussion started: 20 October 2021 c Author(s) 2021. CC BY 4.0 License.
increase of 0.38 mUr year -1 was found during the most recent period from 1970-2015 (Fig. 8). Interestingly, the ratio between the two slopes of 2.7 is similar to the rates of gridded instrumental temperature changes of 2.4 over these two time periods 395 (0.015 °C year -1 from 1916 to 2015, and 0.036 °C year -1 from 1970 to 2015, S3). A possible explanation for the intensified anthropogenic warming trend in the δ 2 HLM chronology could be the fact that recent δ 2 HLM values are additionally controlled by other environmental factors, such as drought. Strong anthropogenic warming may foster a weaker latitudinal temperature gradient and therefore a weaker westerly flow, followed by a decreased net terrestrial mid-latitude precipitation (Routson et al., 2019). Büntgen et al. (2021) found generally wetter conditions in the first half of the twentieth century, followed by a 400 gradual drying trend since the early 1940s. Thus, increasing drought stress for trees growing in the mid-latitudes may have additionally influenced the intensified δ 2 HLM trend over the most recent 45 years. This conclusion can be supported by a comparison of δ 2 HLM values with summer drought data (self-calibrating Palmer drought severity index) as a tendency toward increasing negative correlations is recorded since the 1970s (S4, S5).

Conclusion 405
We measured δ 13 CLM values of eight annually resolved 100-year Fagus sylvatica tree ring series from the Hohenpeißenberg in southern Germany and evaluated their sensitivity to climate variations. The δ 13 CLM values were corrected for the Suess effect and the physiological response due to increasing atmospheric CO2 concentrations using different factors for possible changes in discrimination. The highest temperature sensitivity was recorded with δ 13 CLM values that were corrected for the Suess effect and a correction factor that accounts for a strong CO2 response of 0.032 mUr ppmv -1 (δ 13 CLM_RL) as suggested by Riechelmann 410 et al. (2016). At Hohenpeißenberg, inter-annual to decadal summer temperature variations are significantly reflected in δ 13 CLM values. The highest correlation was observed between JJA temperatures and δ 13 CLM_RL values at r = 0.68 (p < 0.001). Lower but still highly significant correlations were recorded for annual and 'shifted' annual temperatures. To assess the temporal stability of our tree-ring proxy, summer temperatures were modelled by linearly regressing the δ 13 CLM_RL chronology. Highly significant running correlations particularly over the last 50 years, indicating the potential of δ 13 CLM values for reconstructing 415 summer temperatures at annual resolution. The highly significant correlations between gridded instrumental temperatures and δ 13 CLM_high-frequency values confirm the suitability of this proxy to reconstruct high-to-low frequency summer temperatures.
However, our results also indicate that temperature reconstructions based on stable isotope ratios of tree ring lignin methoxy groups are sensitive to low inter-series correlations. These uncertainties were quantified by evaluating moving Rbar values, RE, and CE statistics and can be improved in further studies by increasing the number of replicate tree samples. 420 Additional consideration of δ 2 HLM values from the same trees ; corrected after the suggestion of Greule et al. 2021) demonstrate that δ 2 HLM values predominantly reflect large-scale temperatures since highest correlations were found with western European 'shifted' temperatures (r = 0.69, p < 0.001) and somewhat lower correlations with local 'shifted' temperature variations (r = 0.58, p < 0.001). The results obtained in this study described for the first time a reliable summer temperature proxy derived from δ 13 CLM values 425 in temperate, low elevation environments. We found that δ 13 CLM values better preforms with regional and δ 2 HLM values with large-scale temperatures, indicating that the two proxies could be combined to reconstruct long-term temperature variations at different spatial and temporal scales.

Data availability
We provide the data in heiDATA, which is an institutional repository for research data of the University Heidelberg 430 (https://doi.org/10.11588/data/ZCMVUY).