The Main region (hereafter referred to as MR) is located in Germany and gets its name from the Main River. The Main River originates in the Fichtel Mountains (northeast Bavaria, Germany) and after approximately 500 km empties into the Rhine River. The Main River is the fourth largest tributary of the Rhine and runs from east to west, which is rare for central Europe. From the MR, total-ring-width (TRW) series were used to construct a composite oak TRW chronology covering the period from AD 1 to 2015. To achieve this, an extensive set of 1405 tree-ring series from this distinct geographical region is available. Such data are generally collected and stored in commercial and university dendro labs all over the world and used to develop robust TRW chronologies with sufficient replication. At the University of Hohenheim, for example, a tree-ring archive of ancient pine (Preboreal pine chronology, PPC) and oak samples (Holocene oak chronology, HOC) spanning the entire Holocene (∼12500 years; Friedrich et al., 2004) exists, which serves as a suitable archive for studies of past climate variability. The HOC consists of oak samples from quaternary deposits, archeological wood findings, construction timbers and living trees primarily sampled from southern Germany. Together, these form a unique, annually resolved archive, providing an excellent opportunity to study paleoclimate (for details about the PPC and HOC, we refer to Friedrich et al., 2004).

The composite TRW chronology contains tree-ring series from living trees, construction timbers (historical) and alluvial deposits (subfossil). TRW series from historical (Büntgen et al., 2010; Cooper et al., 2013; Wilson and Elling, 2004; Wilson et al., 2005) and archaeological/subfossil material (Kreuzwieser et al., 2004; Land et al., 2015; Schönbein et al., 2015; Pechtl and Land, 2019) have already been used successfully in dendroclimatological studies to reconstruct past hydroclimate variability. All TRW series used for this study originated from a well-defined geographical region between 49.3–50.8∘ N and 8.2–12.2∘ E (Fig. 1).

For calibration, living TRW series were used from locations along the middle and upper Main valley. The specific growth sites of construction timbers used in historical buildings are unknown, but provenance is assumed to be of local origin. The historical TRW samples are spread over the entire MR and were obtained from the tree-ring laboratory of the University of Hohenheim and the Hofmann tree-ring laboratory. The TRW series from subfossil oak trunks were derived from quaternary deposits of the upper Main River and the lower Regnitz River.

The entire data set of the original TRW series used for this study is available for public access (see Küppers et al., 2018).

Daily precipitation sum (DPS) records from three long-term meteorological stations (Fig. 1) were used to assess the hydroclimate response of the TRW chronology. The meteorological stations Bamberg (49.88∘ N, 10.92∘ E; 240 m a.s.l.), Bergtheim (49.90∘ N, 10.07∘ E; 270 m a.s.l.) and Frankfurt am Main (a.M.) (50.05∘ N, 08.60∘ E; 112 m a.s.l.) are located within the MR. The stations were selected due to their long-term DPS records dating back to 1870 (Frankfurt am Main), 1879 (Bamberg) and 1899 (Bergtheim), as well nearly complete status of these records. The data were provided and verified by Royal Netherlands Meteorological Institute (KNMI) Climate Explorer (http://climexp.knmi.nl, last access: 8 June 2017, Klein Tank et al., 2002) and from the Deutscher Wetterdienst (DWD). The DPS record from the Bamberg station did not contain any missing data during the entire period, whereas the DPS records of the Frankfurt am Main and Bergtheim stations had a gap between March and August 1945.

Precipitation sums of each day from all three stations from AD 1900 to 2015 were averaged to obtain a single DPS record representative of the MR. Between March and August 1945, only the Bamberg DPS record was used. This MR DPS was used for climate-growth analysis.

During AD 1901–2000, the mean precipitation sum between 26 February and 6 July (spring to midsummer) was calculated, serving as a reference period. This interval was chosen based on the results of the climate-growth analysis result during calibration.

For this study, precisely dated TRW series from oaks were assembled to construct a two millennia composite chronology (AD 1–2015) representing the seasonal radial growth variability within the MR.

The individual TRW series were primarily obtained from the Hohenheim tree-ring archive containing series from subfossil (Quercus sp.), historical (Q. sp.) and living (Q. robur, Q. petraea) trees, as well as from the Hofmann tree-ring laboratory (historical series, Q. sp.) to establish a well-replicated TRW chronology. Figure 2 shows the growth period, the mean segment length (MSL) as well as the running inter-series correlation (RBAR) and the expressed population signal (EPS) of the TRW data set (Wigley et al., 1984).

For the study, a set of 1405 precisely dated tree-ring series (Küppers et al., 2018) consisting of 618 subfossil, 586 historical and 201 living trees was available (Fig. 2). The subfossil oaks cover the period until AD 1184, and some individual series go back to 321 before the Common Era. Historical tree-ring series range from AD 797 to 1897 and the living series span the period from AD 1380 to 2015. High fluctuations in the replication throughout the study period can be observed, particularly in the subfossil material, indicating a strong river dynamic which led to low/high deposition frequencies. A sufficient overlap between subfossil and historical as well as between historical and living TRW series was ensured.

The raw TRW series were detrended by fitting a fixed 100-year cubic smoothing spline (Cook and Peters, 1981) to each individual TRW series. A dimensionless index was obtained by calculating the ratio of the raw and predicted values. The individual detrended TRW series were averaged using the robust bi-weight mean method to develop the standard TRW chronology. Due to fluctuations in the replication, variance stabilization was performed by applying the RBAR-weighted method (Frank et al., 2007; Briffa and Jones, 1990). The above-mentioned standardization procedure was conducted using the ARSTAN software (Cook and Krusic, 2005).

EPS and RBAR are key statistics in dendroclimatology for assessing the representativeness of a TRW chronology. EPS and RBAR statistics were calculated as a quality criterion of the coherence of the TRW chronology. Running EPS and RBAR were calculated during 100-year periods with a 50-year overlap. An EPS value of > 0.85 was defined as acceptable for a noise-free chronology. Even when concerns regarding the misinterpretation of the EPS threshold have recently been raised (Buras, 2017), we will use the above-mentioned EPS threshold to ensure that our study is comparable to others.

Tree-ring climate response was assessed using (a) a classical split-period calibration/verification (AD 1900–1957/1958–2015) and (b) a full period calibration approach (AD 1900–2015). The correlation coefficient (r) was calculated for calibration and the coefficient of efficiency (CE) for verification (Cook et al., 1994) to assess the reconstruction quality. A CE greater than zero was assumed to indicate a robust reconstruction.

To assess the temporal stability of the relationship between ring widths and DPS, we applied the bootstrapped transfer function stability test (BTFS; Buras et al., 2017). BTFS bootstraps model parameter ratios between calibration and verification periods of equal length (here, 58 years) for intercept, slope and explained variance over 1000 iterations and tests whether the obtained sample differs significantly from one which would indicate instability of the given parameter. Thus, p values below 0.05 indicate unstable transfer function parameters.

Calibration was conducted between the TRW chronology of the living oak trees and the DPS record. Correlation analysis was performed using a MATLAB^® (MathWorks, 1994–2008) script (Schönbein, 2011) which aggregated DPS data for each year, altering the length of the data interval (from 31 to 361 d in steps of 10 d) and the start date (between 1 January and 15 December). Statistical significance was attained for α=5 %. For details about the running the script, we refer to Schönbein (2011) and Land et al. (2017).

The developed linear climate-growth model was applied to the composite TRW chronology to reconstruct two millennia of hydroclimate variability for the MR. The statistical metric root mean square error (RMSE) was used to measure the climate model performance.

The developed reconstruction of the MR hydroclimate was compared to other hydroclimate reconstructions available for this region. Pauling et al. (2006) reconstructed the seasonal (spring, summer) precipitation from natural proxies (tree-ring chronologies, ice cores, corals and a speleothem) for European land areas from AD 1500 to 1900 (hereafter referred to as P06sp, P06su). Cook et al. (2015) (hereafter referred to as C15) released a tree-ring-based reconstruction of summer droughts and pluvials (calibrated to scPDSI) over Europe, and Büntgen et al. (2011) (hereafter referred to as B11) reconstructed April–June precipitation sums over central Europe, the latest two studies covering the past two millennia. The mentioned reconstruction series were obtained from the National Oceanic and Atmospheric Administration (http://www.noaa.gov, last access: 8 June 2017) database (grid 49.75∘ N, 10.25∘ E, center).

The set of original single-TRW series included in the respective reconstructions by the mentioned authors is, to the best of our knowledge, not explicitly stated in their work or accessible. This circumstance made a comparison between the respective records extremely difficult. We were therefore unable to check for full independence between the different data sets, meaning that the data set used here might be not fully independent from others, particularly from Middle Ages to the beginning of the modern era. Nevertheless, for the periods in which subfossil and living TRW series have been used, a full independence of the MR exists until C15, B11 and P06. We assume that the historical TRW series MR data set guarantee an independence of at least 55 % over the reconstruction period from AD ∼1150 to 1700. At best, the majority (> 80 %) of the MR historical TRW series have not been used by C15, B11 or P06.

It should be mentioned that this is a suboptimal situation for this study which does not allow us to ensure the necessary independence between the data sets but is unfortunately unavoidable in light of the data availability. However, the MR TRW data set used here gives the unique possibility to study the hydroclimate dynamic in a geographically limited area over two millennia.

To expose the common power and relative phase in time–frequency space between the aforementioned and the MR reconstruction, cross-wavelet transform (XWT) was evaluated and wavelet coherence (WTC) was measured by using a MATLAB^® script from Grinsted et al. (2004). The XWT finds regions with high common power and WTC shows where the compared series covary in time–frequency space (even when common power is low). For all calculations, the Morlet wavelet was chosen, providing a good balance between time and frequency localization (Grinsted et al., 2004). Additionally, a 51-year running correlation between the MR reconstruction and P06sp, P06su, C15 and B11 was performed. This relatively short window length allows us to study abrupt temporal changes in the behavior of the aforementioned reconstructions.

The number of TRW series per year fluctuates over time. The lowest replication can be found around AD 280/AD 850, while the highest is in the first decade AD (Fig. 2a). Mean segment length (MSL) is longer for the subfossil and living TRW series and shorter for the historical series but always exceeds 110 years (Fig. 2b). Thus, the temporal fluctuation in MSL limits the extent to which low-frequency signals can be extracted from the TRW data set, like in many other studies dealing with historical oak tree-ring series (see also, e.g., Cooper et al., 2013; Wilson et al., 2013). It is likely that the changes in replication and MSL to some extent do bias the developed reconstruction series, even when EPS is > 0.85. Thus, the developed TRW composite chronology might be biased in the periods AD ∼300 and ∼800. EPS is always above the threshold of 0.85 during the entire time span (mean EPS of 0.96). The lowest EPS was observed in the transition period of AD 830, where subfossil and historical TRW series overlap, corresponding with low replication. The RBAR is 0.32 (Fig. 2c), showing low values in the first half of the 14th and at the end of the 17th century where the TRW chronology consists of primarily historical TRW series. The decrease in RBAR may lead to a bias in the reconstruction. This is accompanied by increased construction activity in the Main region at the beginning of the 14th century (see the high cutting activity marked by TRW series with a similar end of their growth period in Fig. 2a). In regards to the unknown growth location, and thus the correct origin of the historical trees, the low RBAR could be an artifact of timber trading from outside the study area. The exact oak species (Q. robur, Q. petraea) used for the reconstruction is also unclear. At the end of 17th century, a TRW set of very old living oak trees are added to the historical data set and may lower RBAR. In periods where only subfossil and living TRW series were used, RBAR increases and is considerably higher, e.g., prior to AD 500 (subfossil) as well as after AD 1830 (living). Further worth mentioning is that the subfossil TRW series seem to be much more homogenous in regards to their inter-series growth pattern than the historical series. The relatively small region of deposited subfossil trees (Fig. 1) could provide an explanation and may hold a more accurate and local climate signal than the historical TRW subset.

Figure 3 illustrates the calibration and verification process implemented for the development of the model that was used for further hydroclimate reconstruction.

The TRW chronology reveals a significant relationship (p<0.01) to DPS from 26 February to 6 July (spring to midsummer season) during the full (r=0.49) and split (r=0.58) calibration period. For sensitivity test results of the TRW chronology with total precipitation sum for “classical” monthly resolved seasons, see Table A1 in the Appendix. A CE of 0.23 accounts for a robust reconstruction. This indicates that the developed climate-growth model is suitable for reconstructing regional hydroclimate variability.

Nevertheless, as can be seen from Fig. 3, the TRW chronology does not track extremely low (e.g., 1903, 1976, 1991, 1993, 2003, 2015) or high (e.g., 1965, 2007) precipitation rates adequately. Thus, the model underestimates the true sum of precipitation in the spring to midsummer season. Similar results were published by other authors (Cooper et al., 2013; Wilson et al., 2013). Based on the work of McCarroll et al. (2015), we simply defined a 10 % threshold of years in which the precipitation sum was low/high during the full calibration period (116 years, 10 % equivalent to 11.6 years each) as extremes. After ranking the precipitation data, we checked how many years of the TRW data capture these extremes. Only 3 years lie beyond the lowest (statistically not significant) but 4 years beyond the highest threshold (statistically significant, p<0.05). Even though a close statistical relationship between DPS and TRW is evident, there is spurious correlation regarding extreme values.

The applied BTFS indicated stability of transfer function parameters (slope, intercept, r2) over time (p≥0.25 with regard to the null hypothesis of perfectly stable model parameters). A highly significant sign test (p<0.001) revealed true collinearity between TRW and DPS over the calibration period. Model residuals were normally distributed and did not express significant autocorrelations. A moving window correlation between TRW and DPS revealed significant temporal correlations over the calibration; however, these varied between 0.33 (p<0.05) and 0.60 (p<0.001).

Both classic calibration/verification as well as BTFS confirm the sensitivity of TRW to 26 February–6 July precipitation totals.

The developed climate-growth model was applied to the composite chronology to reconstruct spring to midsummer (26 February–6 July) precipitation variability for the MR from AD 2015 back to AD 1. The seasonally resolved reconstruction series over two millennia is shown in Fig. 4. The reconstruction maintains high seasonal- as well as decadal-scale fluctuations.

The developed reconstruction shows high variability on an annual to decadal timescale with repeated phases of below-/above-average rainfall. According to our findings, 9 years of less than 25 % of mean precipitation sum (with regard to AD 1901–2000) could be detected (Table 1). The years AD 338 and 337 are marked by severe low pluvials in the MR. In the decade from AD 334 to 344, the MR was exposed to a conspicuous drought far below rainfall average, and thus it can be described as an extraordinarily dry period. In the 12th century (around AD 1165 and 1177) and in the last decade of the 14th century (around AD 1395), three additional periods of below average rainfall occurred. Heavy single-season rainfall exceeding the mean precipitation sum by a minimum of 25 % (with regard to AD 1901–2000) appeared in 26 individual years during the past two millennia (Table 1). The years AD 357 and 985 appear as the wettest seasons with high deviations from the reference period. Periods of unusually high rainfall lasting 5 years or longer appear in the mid-6th, mid-11th and at the beginning of the 14th century. The period from AD 1125 to 1138 is characterized by consistently high pluvials exceeding the average seasonal precipitation sum for almost one and a half decades. A detailed list of dry and wet decades is provided in Table A2.

It can be summarized that the Main region suffered from severe long-lasting droughts and pluvials over the past two millennia, especially during the mid-4th century.

To evaluate the connection between the MR reconstruction and C15, B11 and P06sp/su, a running correlation (Fig. A1 in Appendix) as well as wavelet analyses between MR and C15/B11 were performed (Fig. 5).

The 51-year running correlation accounts for temporal variation between the reconstructions. The correlation between MR and C15/B11 is high from the 12th until the 18th century. Before the 12th and from the 18th century onwards, the agreement declined to sporadic cases. Around AD 300 and 400 as well as during the 9th century, the MR shows only weak connections to C15 and B11. In the mid-19th century, MR and B11 have no statistical relationship. When the MR is compared to P06sp and P06su, a relationship is only evident during short time periods. Especially during the end of the 17th to the mid-18th century as well as over the 20th century, an agreement to P06sp/su is not detectable. In general, MR and P06sp/su are only spuriously related.

Figure 5 shows the cross-wavelet transform (XWT) and the wavelet coherence (WTC) between MR and B11 and MR and C15 over two millennia.

The XWT between MR-B11 shows significant common power in the ∼10- to 15-year band from the 12th century onwards and in the ∼20- to 60-year band around the 11th and 12th centuries. The XWT between MR-C15 depicts similar results. Albeit, the ∼23- to 65-year band from AD 1000 to 1300 is more pronounced. Looking at the squared coherence spectrum (WTC) between MR-B11 and MR-C15, the close connections appear on the ∼4-year band and are more pronounced on the 10- to 60-year band, especially after AD 1000. Nevertheless, there is a distinct difference between pre- and post-AD 1000. Coherence between MR-B11 and MR-C15 is also obvious prior to AD 1000, on short as well as on long terms, but not as high as in the second millennium. The right-pointing arrows in the significant regions (thick contours in the right panels of Fig. 5) indicate that the compared reconstructions clearly swing in phase, holding evidence for close time–frequency connections. The previously mentioned weak connection between MR and C15 in some periods (Fig. A1) becomes more obvious from the WTC. From AD 200 to 400, for example, no significant connection on shorter and even on longer timescales can be found, supporting the results from the running correlation analysis and indicating substantial differences between the different reconstruction series to the regional MR. The weaker, or even non-significant, connections between MR-B11 and MR-C15 in the first millennium AD show that there is a particularly low coherence between the regional hydroclimate reconstruction (MR) and B11/C11 when an independent data set (see Sect. 2.5) is used.

Figure 6 shows the 20-year low-pass-filtered reconstructions from MR, B11 and C15 from AD 1 to 2000 (Fig. 6a) as well as for two subperiods with an annual resolution (Fig. 6b, c).

As shown in the previous section, the reconstruction series contain differences and similarities on year-to-year and mid-frequency fluctuations. It is obvious from Fig. 6a that MR, C15 and B11 hold strong discrepancies regarding their fluctuations until the end of the 7th century (see also Fig. 5). For example, B11 shows a much higher fluctuation on a longer timescale between AD 250 and 700 than C15 and MR. However, during the aforementioned dry decade around AD 338 (Fig. 6b), all three reconstructions independently show similar patterns and document a long-lasting drought. On the other hand, the pronounced long-lasting drought in the 6th century as reconstructed by B11 is not seen in MR or C15. Although C15 and B11 show evidence of a severe wet decade around AD 435, the MR does not confirm this result. On the contrary, the MR shows moderate changes on a mid-frequency scale. From the end of the 11th to the mid-18th century, all reconstruction series show highly similar patterns (Figs. 5, 6c; Fig. A1) from year to year and on mid-frequency, with the exception of P06sp/su. They share changes in the amplitude synchronously even in very wet and severely dry years/periods. In particular, from AD 1800 onwards, when full independence of the MR data set is guaranteed, notable differences occur. Running correlation analyses also account for substantial differences between the MR and the other reconstructions over the last two centuries (see also Fig. A1).

The oak TRW chronology from the Main region (MR) in southern Germany reveals a significant (p<0.01) sensitivity to the precipitation sum from 26 February to 6 July (spring to midsummer season). During calibration and verification, it became clear that TRW does not cover the full range of precipitation variability, resulting in a poor correlation between TRW and extremes. Low and high seasonal precipitation sums are not accurately reflected in the applied climate-growth model, which is also mentioned by, e.g., Cooper et al. (2013) and Wilson et al. (2013). Land et al. (2017) pointed out that oak tree-ring series are not sensitive to short, heavy rainfall events during the growing season, which may provide a further explanation for this fact. Another reason could be a loss of photosynthetically active leaf area caused by fungal infestations (e.g., mildew) during such phases of high precipitation. We further note that in the MR, a massive insect attack took place between AD 1954 and 1958 (Steger, 1959, 1960), which reduced growth considerably and certainly influenced the climate-growth model performance. The complexity of oak growth over the course of the 20th century is also well established. Friedrichs et al. (2008) showed that oak growth can increase when a combination of warm and dry conditions occur. This leads to the assumption that the control of oak growth remains complex, at least during isolated periods. Altogether, the applied climate-growth model considerably underestimates the total seasonal precipitation sum, which leads to a loss of explained variance.

The precipitation sensitivity of the MR oaks does, however, agree with findings from other studies conducted with oak tree-ring series in Europe. Büntgen et al. (2011) reported a sensitivity of the seasonal precipitation sum from April to June (central Europe), while Friedrichs et al. (2008) (central-west Germany) and Čufar et al. (2008) (southeast Slovenia) showed this for the June rainfall sum. Cooper et al. (2013) (East Anglian), Wilson et al. (2013) (south-central England) and Karanitsch-Ackerl et al. (2017) (northeast Austria) found a close relationship to March–July precipitation sum and Land (2014), Land et al. (2015, 2017) and Schönbein et al. (2015) (Franconia, Germany) to spring–summer precipitation.

European oak TRW chronologies show an explicit connection to hydroclimate during the growing season, in particular from spring until midsummer.

The developed climate-growth model leaves a high level of unexplained variance, which is most apparent in seasons with far below-/above-average rainfall totals. This leads to an inevitable underestimation of past climate variability and is also reported from other authors (Cooper et al., 2013; Esper et al., 2005; von Storch et al., 2004; Wilson et al., 2013; Bürger et al., 2006; Christiansen and Ljungqvist, 2017). Thus, we can assume that in the MR, the year-to-year dynamic might be much more variable than suggested by the tree rings due to the tree rings' inability to capture years with extreme low/high precipitation. This would mean that during the past two millennia, very low/high pluvials are much more pronounced, and thus the hydroclimate dynamic is stronger than suggested by the presented reconstruction. While the oak tree-ring series used here respond very well to spring–summer rainfall during the calibration/verification period, it is not unlikely that, to some extent, warmer/colder phases during the past two millennia (e.g., the Medieval Climate Anomaly or the Little Ice Age) affect the presented reconstruction. It has been shown by Friedrichs et al. (2008) that oak trees from central-west Germany (which is close to our study region) lose their precipitation sensitivity in the anomalously warm 1940s decade. This decreased response to hydroclimatic conditions in central Germany has been confirmed by Büntgen et al. (2010), whose study revealed that oak TRW sensitivity is greatly reduced to scPDSI in the mid-20th century. A well-established decrease in precipitation sensitivity during the 1940s is also observed in our study (data not shown). With this in mind, it is possible that during extraordinary warm (Medieval Climate Anomaly) or extraordinary cold (Little Ice Age) (Mann et al., 2009) periods, TRW reconstructions may show a certain level of bias. However, it remains unclear to what extent the reconstruction here presented is biased.

On the other hand, a severely reduced growth of oak trees does not necessarily mean a “true” dry season but may instead be due to a combination of moderate rainfall, a warm spring season and/or an insect attack. Oak trees are prone to insect attacks, especially under natural forest dynamic processes within a small region. Capturing non-climate-driven tree-ring fluctuations, e.g., caused by insects, would therefore necessitate an investigation of wood anatomy which, due to the high number of wood samples, would be labor intensive and cost prohibitive. An attempt of such can be found in Land et al. (2015) and Schönbein et al. (2015). Nevertheless, single seasons with pronounced below-average (e.g., AD 338)/above-average (e.g., AD 357) average rainfall totals as well as very dry several-year periods (e.g., around AD 1395) appeared in the MR. Interestingly, Spurk et al. (2002) investigated the depositional frequency of subfossil oaks in the MR, which includes the tree samples used here, and linked them to climatically induced fluctuations. They found a sudden onset of germination at AD 400 in the MR, indicating humid conditions. From AD 400 onwards, our reconstruction gives no evidence for a continuous long-lasting period with above-average rainfall but shows distinct high year-to-year fluctuations between AD 420 and 550. Despite that, changes on a low-frequency scale over centuries to millennia are masked by the standardization procedure.

Nevertheless, due to the changes in the mean segment length and the highly variable sample replication, the standardization procedure applied here (100-year spline) is suitable for preserving high- to mid-frequency fluctuations, while the low-frequency variance from the TRW data set is removed. Thus, no inferences can be made for centennial-long precipitation fluctuations for the study region. When comparing the MR hydroclimate reconstruction to B11 and C15 (both capturing the low-frequency domain, e.g., by using RCS detrending), some differences appear on the low-frequency timescale. These are especially apparent in the first millennium AD, where trees from alluvial deposits are available for reconstruction purposes, and is more pronounced between MR-B11 than between MR-C15.

As mentioned by Spurk et al. (2002), human influence could have had a severe impact on forest structure and forest dynamic since the third millennium before Common Era and may have therefore impacted the dynamic of tree growth, perhaps leading to a bias in the established reconstruction. Therefore, investigating epochs where human impact is low, or at best did not occur, is crucial to get a clear picture of pre-human time and its climate dynamic. However, it should be mentioned that humans already settled in the MR over 7000 years ago (Bickle and Whittle, 2013) and certainly influenced their environment accordingly. Much more effort must be made in the future to fully understand the hydroclimate dynamic on a small scale (e.g., in the Main region) during the entire Holocene. Even when the tree rings used represent a highly resolved natural archive, combining different scientific fields, e.g., as recently done by Pechtl and Land (2019), is very necessary to capture potential human effects on former forests. Another factor plays a crucial role: for southern Germany (and for large parts of central Europe), oak is the only species from which TRW series are available from the present back to the Early/Mid-Holocene. It is therefore of tremendous interest for paleoclimate research to interpret the growth fluctuations more precisely and to enhance climate-growth model performance.

The reconstruction of spring to midsummer precipitation variability for the MR developed here shows high connectivity to C15 (Cook et al., 2015) and B11 (Büntgen et al., 2011) for substantial parts of the past two millennia, which is apparent on different time–frequency domains as well as from continuous in-phase fluctuations. Nevertheless, the results of running correlations and wavelet coherence analyses clearly speak for notable differences during some periods. These differences are particularly obvious around AD 300–400, 750–850 and 1820–1870, accounting for intervals where the MR TRW data set is fully independent from others. The differences can primarily be observed on a mid-frequency scale but are also seen in year-to-year fluctuations. We assume that one reason for this could lie in the differing standardization procedures utilized in chronology construction, while the use of mixed TRW series from various provenance and from different sources (archeological findings) could also play a considerable role (for more details, we refer to the mentioned studies). This underlines the significance of this study, as it presents a hydroclimate reconstruction for a small geographical region in southern Germany, in addition to providing a new time series that can contribute to investigations about past European hydroclimate variability.

The high connectivity between MR–C15 and MR–B11 over ∼700 years (end of 11th to mid-18th century) is outstanding and supports the assumption that duplicates within the TRW data are used in the different studies. The exact level of dependency between the different data sets cannot be stated here and remains undetermined. Thus, we assume that during this period our TRW data set might be not as independent as initially considered, which once again highlights the importance of data transparency.

Nevertheless, the MR hydroclimate reconstruction accounts for an on-site variability.

Developing a hydroclimate reconstruction for central Europe reaching back to the Early/Mid-Holocene requires a large data set with a sufficient number of TRW series. This goal could be achieved with the use of TRW series from the Hohenheim Holocene oak chronology (HOC) (Friedrich et al., 2004), assuming that these subfossil oaks reflect the representative central European hydroclimate. The results of our study support the assumption that the TRW series from the Main region, consisting of a well-replicated TRW data set (for more details, see Friedrich et al., 2004; Leuschner et al., 2002; Spurk et al., 2002), have the potential to reflect the rainfall variability of central Europe, at least for southern Germany, as well as regional hydro-regime aspects. Nevertheless, a 10-millennia hydroclimate reconstruction would require the inclusion of TRW data from other river systems (e.g., upper Danube, upper Rhine) due to changes in the deposition frequency of oak trunks and the resultant variance in replication (Spurk et al., 2002; Friedrich et al., 2004). Such temporal changes of distribution are known from tree-ring archives of Ireland, northern Germany and the Netherlands (Leuschner et al., 2002; Spurk et al., 2002) as well. Combining such TRW data sets lasting for millennia from different regions or countries across Europe is crucial for the development of a well-replicated reconstruction representing the overall European seasonal rainfall variability during substantial parts of the Holocene. Additional regional investigations of hydroclimate variability that cover multiple millennia must also be conducted.