Reconstructing Hydroclimatic Variability (1657 AD) Using Tree-Ring Time Series and Observed and Gridded Precipitation Data in Central Greece

Abstract

This study evaluated the long-term hydroclimatic trend through a reconstruction procedure of precipitation variability in central Greece (1657–2020), using eight tree-ring chronologies (Pinus sp. and Abies sp.). Through the combination of gridded climate datasets with tree-ring width (TRW) and earlywood width (EWW) chronologies, we created three precipitation reconstructions, (1) April–August (AMJJA) and (2) May–June (MJ) using TRW and (3) EWW chronologies, utilizing both measured and gridded precipitation data. Chronologies were standardized using ARSTAN, while principal component analysis (PCA) was used for the development of the reconstructions. Verification and calibration of the derived time series (split-period tests, RE > 0, R = 0.62–0.67) confirmed a strong reconstruction that explained 15%–45% of the variability in precipitation. The results revealed strong growth–precipitation relationships throughout spring–summer (AMJJA/MJ). Multi-decadal variability is captured by TRW chronologies, while higher-frequency signals are reflected by EWW. Significant time intervals (19.6-, 12.5-, and 2.2-year cycles) were found by spectral analysis, indicating climatic impacts on tree-ring chronologies. Extremely wet (e.g., 1885, 1913) and dry (e.g., 1894–1895) episodes were confirmed against regional paleoclimate data and were consistent among previous reconstructions (72%–92% agreement). Despite the fact that sample depth reduced after 1978, the EPS was constantly higher than the threshold (EPS > 0.85 post-1746), showing the reliability of the reconstruction. This study expanded the hydroclimatic record of the southeast Mediterranean and highlighted that tree-ring chronologies are reliable variables to predict the historical precipitation.

1. Introduction

Our understanding of historical causes for climatic variability and their effects on human existence can be improved by dendrochronology studies of the links between climate and tree development [1]. Several dendrochronology studies have been implemented dealing with conifer tree species grown in the Balkan Peninsula and Eastern Mediterranean [2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]. Most of these studies deal with the precipitation reconstruction, except for the following studies, which deal with summer sunshine reconstruction [2], streamflow reconstruction [12], standardized precipitation index (SPI) [8,26], Palmer drought severity index (PDSI) [22,25], and standardizes precipitation and evapotranspiration index (SPEI) [23]. The tree-ring width (TRW) was the primary proxy parameter used by the cited performed research; however, in few instances, the maximum density (MXD) [10,27] or the MXD and latewood width (LWW) were also employed [11].

Existing tree-ring chronology studies were based either on a single tree-ring site [3,11,17,23] or on several tree-ring sites, spatially distributed over a wider area [2,6,8,20]. The reconstruction of the climate conditions based on tree-ring proxy data or the investigation of the climate impact on the tree-ring growth is based either on gridded meteorological parameters (i.e., temperature, precipitation, standardized precipitation index (SPI), Palmer drought severity index (PDSI), etc.) provided by several available datasets or on single ground-based weather station data in close proximity with the tree-ring site. Gridded data mainly covers variances of precipitation in large scale, since they are in relation to the spatial interpolation between multiple rain-gauges data in the reference cell center of a broader area [28], whereas single weather station data represent more accurately the precipitation of a specified location. Microclimate conditions, which might be a crucial aspect of the climatic data at a weather station, are not reflected in the mean climatic record of the gridded dataset throughout the study area; rather, it is indicative of the regional climatic conditions according to [29], who have already discussed the characteristics of regional climatic average values. The climate record of the closest single weather station may not share as much variation with tree-ring data as a regionally averaged climate record [30]. Despite several other parameters that influence the reliability of the reconstruction development (tree orientation and type, site elevation, de-trending and cross-dating procedures, etc.), an appropriate combination of chronologies (dense proxy network of many various archives) spatially averaged over a larger area which lead to an increased sample size may reduce several difficulties [6] and finally may lead to an increasing accuracy and robust quality estimation of the analysis [31,32]. This is based on the fact that there is often adequate monthly climatic variance in common between sites of medium or high elevation [27] over a wider area with almost similar characteristics (i.e., Eastern Mediterranean/Balkan Peninsula, Central Europe, Western Europe, etc.).

The aim of the present study is multifaceted. This study combined existing and newer chronologies to develop one of the most recent reconstructions of central Greece (end-year 2020), dated back to 1657 AD. The three developed tree-ring reconstructions cover an extended (April till August) or a shorter (May till June) monthly period and are based on different proxies (TRW or EWW) and gridded data. Additionally, we utilized the recent dataset of the 20th Century Reanalysis Project 20CRv3 [33], which is utilized for the first time in tree-ring studies in the Eastern Mediterranean area. The comparisons between different proxy data (tree-ring width or earlywood width) of the same neighboring area within a common monthly period were also analyzed. Earlywood width (EWW) is introduced for the first time as proxy data in paleoenvironmental studies within the Eastern Mediterranean, as well.

2. Materials and Methods

2.1. Tree-Ring Data

Most of the tree-ring sites are located in Greece and they are evenly distributed (Figure 1) across the mainland. In the current research, eight dendrochronology time series of pines (Pinus sp.) and fir (Abies sp.) species, ranging from 1657 to 2020 (

Table 1. The tree-ring time series chronologies utilized in the current study, based on the ITRDB databank * (https://www.ncei.noaa.gov/products/paleoclimatology/tree-ring, accessed on 14 October 2024), except the Pertouli forest chronology. The truncated period was used in the study.

No	Latitude	Longitude	Altitude (m)	Site	Species	Availability	Proxy Variable	Original Period	Truncated Period *	Chronology doi Number
1	37.08	22.33	1.106	Sparta (Langada)	Pinus nigra subsp. nigra	Standard, ARSTAN, residual	TRW, EWW	1825–1980	1833–1978	https://doi.org/10.25921/jpm4-p943
2	37.63	22.28	1.322	Menalon Oros	Abies cephalonica Loudon	Standard, ARSTAN, residual	TRW, EWW	1832–1980	1833–1978	https://doi.org/10.25921/gpe0-0t88
3	38.72	21.67	1407	Panetolikon-Oros	Abies borisii regis Mattf.	Standard, ARSTAN, residual	TRW, EWW	1812–1980	1812–1978	https://doi.org/10.25921/tf06-v979
4	41.77	23.38	2.100	Vihren National Park	Pinus leucodermis H. Christ.	Standard, ARSTAN, residual	TRW, EWW	1721–1980	1722–1978	https://doi.org/10.25921/fech-b460
5	36.92	22.95	1.400	Taygetos Forest	Pinus nigra subsp. nigra	Standard	TRW	1657–1999	1657–1999	https://doi.org/10.25921/v2v4-r670
6	40.3	20.9	1500	Scotida Forest (1)	Abies cephalonica Loudon	Standard	TRW	1676–1978	1676–1978	https://doi.org/10.25921/mdh4-e656
7	40.3	20.9	1500	Scotida Forest (2)	Pinus nigra subsp. nigra	Standard	TRW	1751–2003	1751–1999	https://doi.org/10.25921/zqwn-3r45
8	39.54	21.45	1200–1500	Pertouli Forest	Abies borisii-regis Mattf.	Standard, ARSTAN, residual	TRW, EWW	1833–2020	1833–2020	https://doi.org/10.3390/f13060879

, Figure 2 and Figure 3), were employed to create the regional reconstructions of central Greece (Figure 1). The dendrochronology site elevation ranged between 1.100 and 2.100 m. All dendrochronology time series, except for the Pertouli forest [23], were downloaded from the National Oceanic and Atmospheric Administration (NOAA) databank (https://www.ncei.noaa.gov/products/paleoclimatology/tree-ring, accessed on 14 October 2024).

The Pertouli forest chronology (valid between 1833 and 2020) is the most recent among the others that were used in the present study. The eight available chronologies were truncated appropriately to formulate the common principal components for the development of the reconstructions (

Table 2. Analyzing the current reconstructions by splitting the calibration–verification period. Lines with italics correspond to periods which finally were excluded from the procedure (not accepted by the relevant criteria).

Reconstruction	Period	Number of PC	Calibration	Verification	R	F	RE	Regression Line	R2	R (Probability)	F	Acceptance
AMJJA (20th Reanalysis)	1929–2020	1	1929–1953	1954–1978	0.48	6.8	0.22	256.01 × PC1 − 24.48487	0.31	0.56 (2.5 × 10−5)	21.8	OK
			1954–1978	1929–1953	0.65	17.1	0.39
	1929–1999	3	1929–1953	1954–1978	0.55	10.2	0.32	226.88 + 22.47 × PC1 − 21.13 × PC3	0.36	0.60 (3.5 × 10−6)	27.47	OK
			1954–1978	1929–1953	0.42	4.9	0.22
	1833–1978	5	1929–1953	1954–1978	0.68	20.12	0.35	233.0 + 28.93 × PC1 + 9.07 × PC2	0.45	0.67 (1.2 × 10−7)	35.6	OK
			1954–1978	1929–1953	0.61	13.22	0.39
	1812–1978	4	1929–1953	1954–1978	0.65	16.7	0.25	238.57 + 33.27 × PC1 − 11.47 × PC3	0.39	0.62 (1.9 × 10−6)	29.44	OK
			1954–1978	1929–1953	0.49	7.3	0.29
	1751–1978	3	1929–1953	1954–1978	0.63	14.7	0.33	228.97 + 27.79 × PC1 − 23.63 × PC3	0.35	0.59 (7.7 × 10−6)	25.14	OK
			1954–1978	1929–1953	0.50	7.52	0.16
	1676–1978	2	1929–1953	1954–1978	0.58	11.36	0.46	230.075 + 37.720 × PC1	0.35	0.59 (7.5 × 10−6)	25.22	OK
			1954–1978	1929–1953	0.50	7.5	0.20
	1657–1978	1	1929–1953	1954–1978	0.59	12.4	0.32	213.03 × PC1 + 16.14284	0.24	0.49 (3.5 × 10−4)	14.84	OK
			1954–1978	1929–1953	0.33	2.7	0.12
MJ (GPCC)	1929–2020	1	1929–1953	1954–1978	0.61	13.9	0.1	171.97 × PC1 − 95.78	0.35	0.59 (7 × 10−6)	25.5	OK
			1954–1978	1929–1953	0.63	15.4	0.29
	1929–1999	3	1929–1953	1954–1978	0.52	8.51	−0.19	1.4157 × PC1 − 31.29	0.25	0.50 (2 × 10−4)	16.0	Not OK
			1954–1978	1929–1953	0.65	17.05	0.31
	1833–1978	5	1929–1953	1954–1978	0.61	13.9	0.1	71.22 + 14.53 × PC1 + 9.41 × CPC2	0.42	0.65 (3 × 10−7)	35.5	OK
			1954–1978	1929–1953	0.64	33.6	0.28
	1812–1978	5	1929–1953	1954–1978	0.54	9.4	0.05	71.97 + 18.56 × PC1 + 8.36 × PC2	0.39	0.62 (1.4 × 10−6)	30.3	OK
			1954–1978	1929–1953	0.60	12.7	0.29
	1751–1978	4	1929–1953	1954–1978	0.53	8.7	0.19	74.83 + 17.60 × PC1 − 19.01 × PC4	0.39	0.62 (1.7 × 10−6)	29.7	OK
			1954–1978	1929–1953	0.59	12.3	0.30
	1676–1978	2	1929–1953	1954–1978	0.56	10.7	0.11	74.73 + 23.39 × PC1 − 14.68 × PC2	0.39	0.62 (5.5 × 10−7)	29.8	OK
			1954–1978	1929–1953	0.63	15.4	0.38
	1657–1978	2	1929–1953	1954–1978	0.29	2.1	0.05	105.37 × PC1 − 30.8	0.15	0.38 (7 × 10−3)	8.00	OK
			1954–1978	1929–1953	0.57	10.7	0.38
MJ (CRU)	1929–2020	1	1929–1953	1954–1978	0.61	13.4	0.27	254 × PC1 − 164.56	0.39	0.62 (7 × 10−7)	29.0	OK
			1954–1978	1929–1953	0.67	18.5	0.40
	1833–1978	5	1929–1953	1954–1978	0.48	6.8	0.09	0.9991 × PC1 + 0.00721	0.39	0.62 (2 × 10−6)	29.3	OK
			1954–1978	1929–1953	0.40	4.2	0.13
	1812–1978	2	1929–1953	1954–1978	0.16	0.62	−0.31	-	0.13	0.36 (0.01)	7.05	Not OK
			1954–1978	1929–1953	0.30	2.2	0.07
	1722–1978	1	1929–1953	1954–1978	0.06	0.07	−0.32	-	0.1	0.28 (0.05)	4.18	Not OK
			1954–1978	1929–1953	0.44	5.5	0.034

, Section 3.2). The longest period of tree-ring time-series chronology (Taygetos forest) was valid between 1657 and 1999. A statistical evaluation has been conducted using the COFECHA program [34,35] to validate the cross-dating. ARSTAN soft was used to standardize the chronologies [36]. To eliminate the non-climatic trends caused by age, size, and stand dynamics, a double de-trending procedure was followed applying firstly a negative exponential smoothing or linear regression and then splines with a frequency response of 50% at a wavelength of two-thirds of each series’ length [37]. After de-trending and indexing (standardizing) the tree-ring time series, the program ARSTAN applies a bi-weight robust estimation of the mean value function to eliminate the impact of endogenous stand disturbances and create chronologies. Three time-series versions are generated by the ARSTAN program: the standard, the ARSTAN, and the residual. In the current research, the tree-ring width (TRW) and earlywood width (EWW) chronologies derived from the ARSTAN standard and residual method were chosen as the main parameters which were affected by the climatic variable. The highest correlation coefficients had appeared between the TRW and EWW chronologies with the gridded climatic data (See the following section). More information concerning the procedure of the creation of the dendrochronologies can be accessed on the site of NOAA databank.

The running value of the express population signal (EPS) value (Equation (1), Figure 4) is calculated with a 25-year overlapping period, using a 50-year window, and the formula is as follows:

$$EPS={\displaystyle \frac{t\overline{r_b}}{t\overline{r_b}+\left(1-\overline{r_b}\right)}}$$

(1)

where t is the total number of the tree-series averaged (one core per tree) and $r_b$ the mean between series correlations [12].

2.2. Climatic Data

The reconstruction of the precipitation time series is depended on gridded monthly CRU TS 4.07 (Climate Research Unit, East Anglia, UK [28]), GPCC v2020 (Global Precipitation Climatology Centre [38,39], and 20CRv3 weather data (20th century reanalysis V3 [33,40]). Gridded data also provides a more regional signal than ground-based weather station data [12]. Three different reconstructions were developed, based on each one of the referred monthly gridded data. These data were averaged within the area between latitude 39–40° and longitude 21–23°, situated at the central part of mainland Greece. This area is chosen because of the increased correlation between proxy data and monthly gridded precipitation sets. Chronologies used in the present study are almost evenly distributed away from the area under consideration so as this area may become a representative geometric center of the tree-ring sites. It must be noted that Pearson’s R is the main criteria to choose the reconstruction proxy and gridded data. Our iterative procedure included several tryouts between proxy data (TRW or EWW), periods (i.e., AMJJA, MJ, etc.) and gridded precipitation data, in order to improve the Pearson R.

A common period between 1928 and 1978 is used to examine the growth climate response patterns between the time series and the conditions of the environment (precipitation). Previous studies [3,6,7,8,26,41] revealed that the period near the mid-19th century till recent years was accurate and reliable enough, because of the increased available rain gauge data which could be utilized in the interpolation procedure. Especially for the case of 20CRv3 data, an increased accuracy is introduced after the year 1930 owing to the availability of 5% more observations per assimilation cycle [33]. All data used in the present study were downloaded from the Royal Netherlands Meteorological Institute (KNMI) climate explorer portal (https://climexp.knmi.nl/, accessed 22 November 2024).

2.3. Statistical Analysis

A split calibration–verification procedure was employed to test the reliability of the precipitation reconstructions (Table 2 [42]). The period from 1928 to 1953 was used for calibration and the period from 1954 to 1978 for verification. This process was then reversed. Principal components analysis (PCA) was used in the current study. The PCA method’s primary benefit is its ability to enhance the correlation coefficient between the reference period of the climate variables and the final time series by appropriately combining the available chronologies. A nested regression procedure [43] was used to optimize the reconstruction length. This procedure involves a combined synthesis of the available nests (Table 2). After adjusting each one nest to the more recent and better-replicated nest (referred to as the calibration period 1928–1978) [12] using multiple linear regression analysis, the nested reconstructions were spliced together. The nest that had the strongest relationship with the meteorological factors—which is frequently the one closest to the reference period (1928–1978) nest—is incorporated together with the total of the best-replicated nests in the final reconstructions. This spliced procedure was implemented twice, one for the nests before the reference period and another for the nests after the reference period (the more recent nests). When a nest fails to satisfy the splitting calibration–verification criteria, as shown below (i.e., RE > 0), the number of contributing nests was lowered from the maximum allowed.

The statistics used to test the reliability of the reconstruction models included the reduction of error (RE), the Pearson correlation coefficient (R), the coefficient of determination (R2), and the F-test [37,42]. To validate the results of the analysis, the RE statistic was used. The RE statistic’s theoretical bounds go from negative infinity to a maximum of one, which denotes complete agreement. Any positive RE value suggests that there is some valuable information in the reconstruction, which is a very rigorous statistic [44].

3. Results

3.1. Climate and Tree-Ring Data Relationship

In the south-east Mediterranean environment, drought stress during extremely warm times typically results in low growth rates; the most common cause is water deficit observed throughout the growing season [45,46,47]. Restoring water availability results in a restoration of cambium zone activity [48]. In Mediterranean locations, the growing season typically begins in late April and lasts until September [49,50,51,52,53]. Consequently, among several other parameters, the monthly precipitation during spring and summer may be some of the major driving factors of the conifer tree growth rate in a medium- to high-elevation Mediterranean forests.

Figure 5 represents the Pearson correlation coefficient between the standard TRW (Figure 5a,b) and residual EWW (Figure 5c) chronologies with different gridded monthly precipitation data (20CRv3, GPCC v2020, and CRU TS v4.07), during the period from 1928 to 1978 within the reference area (lat. 39–40°, long. 21–23°, Figure 1). From May of the prior year to December of the current year, the Pearson correlation coefficients were generated.

In Figure 5a, it is observed that TRW (std) chronologies are strongly positive with the April to August (AMJJA) 20CRv3 precipitation data. The same influence continues to be weaker and almost non-significant during October and November. An almost negative and weak correlation is observed in precipitation during January to March (JFM), September, and December, respectively. This result indicates that the precipitation during the late spring and summer months increases the tree-ring width, while the January to March precipitation decreases the TRW.

In the case of Figure 5b (TRW std), observations relative to the AMJJA period are different. The number of the TRW chronologies with strong positive correlation with the complete (AMJJA) GPCC v2020 precipitation data period is reduced, resulting only to significant influence in May and June. An almost negative and weak correlation is observed during January to March (JFM), whereas during August to December the precipitation influence on chronologies is mostly weaker with sign changes.

Almost the same observations can be made in the influence between EWW (res) and CRU TS v4.07 gridded data (Figure 5c) in the study area, with the most notable difference being the strong positive correlation with October precipitation. Although the different precipitation datasets used for the calculation of the EWW and TRW growth influences (Figure 5b,c), there is a quite common notable environmental performance of both proxy values.

No remarkable correlations are observed during the previous year’s months, except for some single significant positive/negative correlations.

Finally, Figure 5a–c show that the higher correlations between the time series and the climatic data are the TRW (std) AMJJA period for the 20CRv3 data and the TRW (std) and EWW (res) MJ periods for the GPCC v2020 and CRU TS 4.07, as well.

3.2. The Followed Process for Reconstructing the Tre-Ring Chronologies

Table 1 and Figure 2 and Figure 3, show the eight (standard) and five (residual) chronologies that were used to reconstruct the monthly precipitation of the AMJJA (based on 20CR data) and MJ (based on GPCC data and CRU data), applying principal component analysis (PCA) [42]. After the implementation of several combinations and tests, the chronologies of Chalkidiki and Menalo were removed from the creation of the TRW AMJJA (20CR data, Figure 5a) reconstruction whereas the chronology of Vihren was excluded from the reconstructions of TRW AMJJA (20CR data) and TRW MJ (GPCC data, Figure 5b). The overall variation described by each period’s final reconstructions (Table 2) ranges from a minimum of 15% (MJ TRW, period 1657–1978) to a maximum of 45% (AMJJA reconstruction period 1833–1978) [12,43].

Table 3. The resulting reconstructions and the obtained reconstructions from the literature were compared using Pearson correlation coefficients. Bold characters correspond to significant values (p < 0.05 or 95% c.l).

Reconstructions of the Present Study	Griggs et al. (2007) (MJ) [6]	Pauling et al., 2005 (JJA) [55]	Pauling et al., 2005 (MAM) [55]	Cook et al., 2005 (PDSI) (JJ) [56]	Akkemic and Aras 2005 (AMJJA) [14]	Akkemic Dagdeviren and Aras 2005 (MAM) [13]	Touchan et al., 2005 (MAM) [5]	Touchan et al., 2005 (MAM) Second [5]	D Arrigo and Kullen (2001) [57]	Sakalis (2025) Greece JJA [27]	Sakalis (2025) Trikala JJA [27]	Sakalis (2025) Tripoli AMJJA [27]
GPCC MJJA PRECIPITATION	0.27 (p < 6 × 10−5)	0.24 (p < 3.4 × 10−4)	0.18 (p < 0.0074)	0.51 (p < 1.2 × 10−15)	0.28 (p < 3.6 × 10−5)	0.13 (p < 0.042)	0.25 (p < 6.5 × 10−5)	0.29 (p < 8.2 × 10−6)	0.23 (p < 6.4 × 10−4)	0.17 (p < 0.01)	0.18 (p < 0.005)	0.18 (p < 0.007)
REANALYSIS AMJJA PRECIPITATION	0.25 (p < 1.3 × 10−4)	0.26 (p < 2.4 × 10−5)	0.15 (p < 0.015)	0.5 (p < 1 × 10−20)	0.29 (p < 8 × 10−6)	0.13 (p < 0.04)	0.24 (p < 5 × 10−5)	0.34 (p < 2 × 10−7)	0.26 (p < 6.5 × 10−5)	0.14 (p < 0.03)	0.14 (p < 0.03)	0.14 (p < 0.029)
EARLYWOOD MJ CRU PRECIPITATION	0.30 (p < 2.9 × 10−4)	0.13 (p < 0.084)	0.21 (p < 0.0073)	0.52 (p < 5.1 × 10−13)	0.39 (p < 10 × 10−6)	0.12 (p < 0.11)	0.33 (p < 4.5 × 10−6)	0.40 (p < 1.8 × 10−8)	0.44 (p < 2.1 × 10−8)	0.08 (p < 0.33)	0.09 (p < 0.29)	0.09 (p < 0.45)

shows that the Pearson correlation coefficients R are often increased, significant, and balanced across the calibration and verification periods, and the RE statistic values are positive [7,54]. Given the large and highly significant F values of the sub-periods, the resulting reconstructions were appropriately validated in order to create the final reconstructions. A few exceptional cases, where RE < 0 during the splitting periods (Table 2, 1929–1999 for TRW MJ and 1812–1978 and 1722–1978 for EWW MJ) were considered unacceptable, and the relevant principal component periods were excluded from the reconstruction. As described in the preceding paragraph, the precipitation data for the whole 1928–1978 period were utilized for the final calibration to produce the final reconstructions (Figure 6) using the successful splitting technique.

These three final precipitation reconstructions (Figure 6), which extended from 1657 to 2020 (cases of AMJJA and MJ based on TRW) and from 1834 to 2020 (case of MJ based on EWW), presented mean values of Pearson correlation coefficients between 0.62 and 0.67 (Table 2), which showed an acceptable fit to the gridded observational data that was high enough for a trustworthy and accurate examination of historical climate conditions [2,6,7,11,26,27].

Sample length (Figure 2 and Figure 3) reaches a maximum at 1977 (131 for AMJJA TRW, 159 for MJ TRW, and 91 for MJ EWW) and a common minimum value of 24 at 2019. The sharp decrease in samples after 1978 can be attributed to the fact that only Pertouli, Taygetos, and Scotida (2) chronologies were extended till recent years, 2020 and 1999, respectively.

According to Figure 4, after 1746 (or 1815 for MJ EWW), the running expressed population signal (EPS) is above the critical value of 0.85 for all the reconstructions. Also, a slightly lower than 0.83 temporary reduction in the running EPS is observed between 1840 and 1860, owing to the introduction of several new chronologies with a low number of samples and inter-series correlation. All of the reconstructions may be classified as interpretable based on this empirical criterion. The crucial 0.85 cut-off number, however, is purely arbitrary and ought to be utilized just as an estimate [43,56].

Figure 7 presents the reconstructed (MJ or AMJJA) series and the relevant precipitation data during the calibration period 1928–1978 (gridded instrumental data). It is clear that the series have a similar variance. The length of the series is the primary distinction between the instrumental (gridded) and reconstructed precipitation series throughout the calibration period. The instrumental (gridded) series have increased length compared to the reconstructed series, which is more prominent in the case of MJ TRW and EWW (Figure 7b,c).

3.3. Verification with Other Existing Reconstructions

Table 3 presents the correlation coefficient between the reconstructions of our research with the existing reconstructions of precipitation based on the north Aegean [6], precipitation based on Central Europe [58], PDSI values in the summer period [55], precipitation in spring and summer [14], and the Balkan Peninsula [8,14,27,56]. Although the existing reconstruction may refer to quite different areas, variables, and monthly span periods, significant correlations appear. Based on these results, reconstructions of the present study correlated significantly (p < 0.01) during the common yearly period with the majority of the existing reconstructions referred to above. Only the MJ EWW reconstruction presents weak correlations with some of the existing series. The highest correlations are observed with the existing reconstructions of JJ PDSI of [55], spring TRW western Turkey of [8], and AMJJA TRW south-central Turkey of [14].

The low (but already significant) correlation coefficients between the present TRW reconstructions and the MXD reconstructions of [27] are noticeable. Although there are several common tree-ring chronologies (Menalon, Panetoliko, Vihren, and Sparti) used by both reconstructions, the relevant correlation coefficients were expected to be higher than the existing values. However, a moving average filter (5, 10, and 20 year) between both the reconstruction of the present study and the reconstruction of [27] finally resulted in a significant increase in the relevant correlation coefficients (Table S1, Figure S1, Supplementary Material). This led to a conclusion of a common and robust reconstruction performance on low-frequency precipitation variations (long-term mean precipitation variations).

When filtering only the reconstructions of [27] with a 5-year moving average (an actually low pass filter, Figure S1, Supplementary Material), the correlation coefficient between the reconstructions increased (it almost double). Reversely, filtering only the reconstructions of the present study with a 5-year moving average does not lead to a substantial increase in the correlation coefficient. Also, implementation of a 10-year moving average alternately does not differ gradually in any case, whereas a 20-year moving average removes the correlation. This final observation may present a different environmental signal between the residual version of MXD reconstructions of [27] and the TRW (present study) in the 0–0.2 frequency domain (high-frequency precipitation variations between 1 and 5 years), which may possibly be triggered by the different nature of the MXD and STD chronologies and the reference precipitation data (20CR vs. CRU) used during the calibration procedures. As indicated also by [12], residual chronologies may contain a high-frequency environmental signal. It must be noted that the low-frequency signal (corresponding to a low amplitude) of the present reconstruction is obvious in Figure 7 as well.

Finally, the consistency of the current investigation’s reconstructions of the climatic signal is confirmed by the previously noted substantial correlations between the reconstructions of the current study and the reconstructions of nearby locations. The KNMI climate explorer was used to download the existing reconstructions (http://climexp.knmi.nl, accessed on 12 October 2024) and the ITRDB database was also used https://www.ncei.noaa.gov/pub/data/paleo/treering/reconstructions/turkey, accessed on 22 November 2024).

3.4. Reconstructions and Regional Precipitation Variability

Although reconstructions of the present study differ in monthly span (AMJJA vs. MJ) and in gridded precipitation data used for their calibration (20CRV3, GPCCv2020, and CRU TS4.07), there exist significant correlation between them (

Table 4. Pearson correlation coefficients between the calculated reconstructions of the current research.

	AMJJA TRW	MJ TRW	MJ EWW
AMJJA TRW	1	0.93	0.67
MJ TRW		1	0.70
MJ EWW			1

). EWW reconstruction reveals slightly lower correlation with the other two. Although the two relevant MJ reconstructions were based on different proxy data chronologies and quite different tree-ring sites and calibration gridded data, the value of R (0.70) between them is high enough to ensure the consistency of the developing procedure and their common environmental signal.

A simple test for addressing the reconstructions’ ability to describe the high-frequency precipitation variability was performed, calculating the SD of the difference between the current and the previous year of each reconstruction. Results showed that the SD of AMJJA TRW reconstruction is almost half the SD of MJ TRW (0.21 vs. 0.41), whereas the SD of MJ EWW is the highest (0.58). It is evident that residual MJ EWW reconstructions may contain higher frequency environmental signals than the others. This is possibly influenced by the use of the residual version of chronologies to obtain the relevant reconstruction [12] as well.

Reconstruction also correlated highly and significantly with the regional (AMJJA or MJ) gridded precipitation data of Greece (period 1928–1978), with coefficients R = 0.66 (AMJJA TRW), R = 0.64 (MJ TRW), and R = 0.65 (MJ EWW), as well. The three reconstructions represent almost the same spatial extent. This notably high regional signal of reconstructions containing chronologies from different climatic zones (i.e., Vihren chronology-alpine climate zone) is not something new [8,9,10,20,26,27].

A periodogram (power spectra, Figure 8) of the reconstructed series showed several significant (p < 0.05) cyclic signals. The most significant of them existed in 19.57, 12.45, and 2.21 years (peaks of diagrams) for the AMJJA TRW, MJ TRW, and MJ EWW reconstructions, respectively. Several common significant cycling signals occurred between 3 and 19 years as well (

Table 5. Significant cycling signals and the relevant probability according to power spectra (Figure 8). Bold characters correspond to the most pronounced period (with the maximum probability) for each reconstruction, whereas values with italics characters correspond to common (or almost common) pronounced periods between the 3 reconstructions.

Reconstruction	Period T	Probability P
Reanalysis AMJJA precipitation
	54.80053	0.01561
	19.57139	0.0001
	15.22232	0.02127
	14.42107	0.02976
	13.69994	0.02289
	12.45454	0.0002
	7.82859	0.04842
	7.6111	0.01683
	6.68293	0.00593
	3.80555	0.04062
GPCC MJ Precipitation
	30.44418	0.00991
	22.83314	0.03905
	19.57139	0.00194
	14.42107	0.03196
	12.45454	0.00131
	7.02563	0.04313
	6.68293	0.00455
	3.91429	0.02663
	3.80555	0.03906
Earlywood MJ CRU Precipitation	18.60015	0.02078
	12.40002	0.03086
	3.95745	0.04443
	2.61972	0.03652
	2.21429	0.00119

, characters with italics). The lower-cycle periodic significant signals of the MJ EWW reconstruction may ensure the ability to capture the high-frequency environmental variation, as is noted above.

The straightforward and widely accepted standard deviation (SD) classification criteria [13,26,57,59,60,61] was applied in order to confirm the dry/wet events of the current study. There were two phases to the process. Initially, the standard deviation (SD) was used to divide the difference between the precipitation of the current year and the average precipitation of the reconstructed chronologies. In the literature, these numbers are frequently referred to as the standard deviation precipitation abnormalities. Values which exceed the 2 SD threshold are denoted as extremely wet events, whereas values between 1 SD and 2 SD are denoted as wet events. Similar values between −1 SD and −2 SD, and values less than −2 SD are denoted as dry or extremely dry events as well.

Table 6. Dry/wet years based on present reconstruction (a simplified version of Table S3). Bold characters denote years with common observation of drought or flood by at least 1 existing reconstruction derived from the literature.

TRW AMJJA (20CRv3)				TRW MJ (GPCCv2020)				EWW MJ (CRUTS 4.07)
WET		DRY		WET		DRY		WET		DRY
1 ≤ SD < 2	SD ≥ 2	−2 < SD < −1	SD ≤ −2	1 ≤ SD < 2	SD ≥ 2	−2 < SD < −1	SD ≤ −2	1 ≤ SD < 2	SD ≥ 2	−2 < SD < −1	SD ≤ −2
1747	1748	1756	1773	1766	1748	1756	1830	1841	1857	1852	1834
1755	1773	1767	1893	1771	1799	1767	1893	1842	1865	1854	1861
1766	1799	1769	1894	1792	1812	1769	1894	1844	1876	1858	1869
1771	1804	1774	1895	1804	1814	1773	1895	1859	1877	1867	1871
1792	1812	1779	1908	1813	1815	1774	1908	1860	1881	1868	1929
1813	1818	1782		1816	1818	1779		1866	1884	1878
1814	1827	1794		1819	1885	1782		1889	1915	1880
1815	1859	1795		1822	1913	1794		1892		1882
1819	1885	1796		1827	1937	1795		1897		1887
1822	1913	1806		1855		1796		1899		1893
1826	1914	1808		1858		1806		1900		1894
1837	1915	1810		1859		1808		1913		1895
1838	1975	1830		1860		1810		1917		1898
1842		1833		1876		1829		1920		1908
1855		1834		1877		1831		1930		1916
1858		1840		1884		1832		1936		1923
1860		1847		1899		1833		1959		1927
1876		1861		1901		1834		1960		1928
1877		1869		1912		1835		1975		1932
1884		1870		1914		1840		1979		1935
1899		1871		1915		1847		1983		1952
1900		1880		1934		1848		1998		1969
1912		1887		1936		1861				1971
1926		1896		1940		1869				2002
1936		1909		1941		1870				2006
1937		1918		1959		1875				2020
1959		1923		1960		1879
1960		1928		1975		1880
1998		1929		1979		1887
2018		1944		1983		1890
		1945		1998		1891
		1949		2005		1896
		1969		2018		1907
		1971				1909
		1988				1918
		2020				1923
						1928
						1929
						1944
						1945
						1969
						1981
						2020

presents the dry/wet years for each reconstruction according to classification criteria of the SD. Bold characters denoted years with common observations of drought or wetness by existing reconstruction from the literature. Analytical references for verified events by the available studies are presented in Table S3, (Supplementary Material). Some of the available reconstructions from the literature were based on several common tree-ring arrays but different proxy data (MXD-[27]) or common tree-ring arrays and proxy data chronologies (i.e., TRW-[23], TRW-[20]). For the case of TRW reconstructions, there is a balance between the wet and dry events, with the number of wet years being slightly higher than drought years. The same occurred during the extremely wet and extremely dry years. EWW reconstruction presented more drought years than wet years and more extremely wet years than extremely dry years. The years of 1885 and 1913 were by far the years with the most extreme wet threshold during the period under consideration, whereas 1894 and 1895 were the most extreme drought years of the TRW reconstructions. Similarly, 1865, 1884, 1877, and 1861, 1869, 1871 were the most extremely wet and dry years of the EWW reconstructions, respectively. The most extreme dry/wet years referred above were also recognized by at least one reference study, except for 1895 (extreme dry for TRW MJ) and 1871 (extreme dry for EWW MJ).

In general, an overall verification of drought and flood events between reconstructions of the present study and by at least one other reconstruction of neighboring areas derived from the literature (years with bold characters), led to values of 72.5%, 69%, and 64% for the TRW AMJJA, TRW MJ, and EWW MJ, respectively. These values increase gradually in the case of extreme wet and dry events to 92%, 78%, and 90%.

Concerning the driest/wettest periods lasting for 5, 10, and 20 years (Figure 6), we obtain the following:

–

TRW AMJJA: 1892–1896 and 1911–1915 (5 y), 1887–1896 and 1813–1822 (10 y), 1879–1898 and 1809–1828 (20 y).

–

TRW MJ: 1892–1896 and 1812–1816 (5 y), 1887–1896 and 1813–1822 (10 y), 1879–1898 and 1809–1828 (20 y).

–

EWW MJ: 1867–1871 and 1862–1866 (5 y), 1925–1934 and 1857–1866 (10 y), 1927–1946 and 1901–1919 (20 y).

The 10-year dry period observed by TRW reconstructions almost coincides with the relevant period highlighted by [62]. Likewise with the wet 20-year period of EWW reconstruction was also verified by [12].

4. Conclusions

In this work, eight tree-ring chronologies (Pinus sp. and Abies sp.) with a span time of 1657–2020 were used to provide a complete dendroclimatic reconstruction of precipitation variability in central Greece. Reliable reconstructions for AMJJA (April–August) and MJ (May–June) precipitation were constructed using standardized (TRW) and residual (EWW) chronologies and gridded climate data (20CRv3, GPCC v2020, CRU TS 4.07). The reconstructions revealed high- and low-frequency climatic signals, and were consistent with previous paleoclimate studies, explaining 45% of precipitation variance.

The results showed that tree growth is sensitive to spring–summer moisture availability, with TRW chronologies showing significant correlation with both MJ precipitation (GPCC) and AMJJA precipitation (20CRv3). EWW chronology presented more noise, however it showed clear high-frequency signals which may represent the cambial reaction to seasonal water stresses. A total of 72%–92% of extreme years were confirmed, while it was important to validate extreme dry/wet occurrences using previous reconstructions that were supported by independent research, such as the severe droughts of 1894–1895 and the exceptionally wet years of 1885 and 1913.

The main limitations of the study were the short length of instrumental data compared to tree-ring time series, which is a frequent problem in dendroclimatology, and the limited sample depth beyond 1978. The necessity for increased spatial sampling to improve signal strength was shown by the removal of some chronologies because of low RE values.

This work enhanced our knowledge about the variability of the Mediterranean hydroclimate and revealed the utility of tree-rings in reconstructing historical precipitation patterns. The results highlighted the susceptibility to droughts and floods and showed past extreme events and climate cycles in the specific region, which are crucial factors for managing water resources under climate variability conditions.