1 Introduction
As global warming intensifies, understanding regional patterns of winter temperature variability has become a key priority in climate science. Winter temperature trends differ substantially by region (
Jones and Moberg, 2003;
van Oldenborgh et al., 2009,
Matthes et al., 2015;
Rahmstorf et al., 2015;
Overland et al., 2016;
Turner et al., 2016;
Liyew et al., 2024). Winter minimum temperatures (WMT) in particular are subject to complex, region-specific influences that make them challenging to predict (
Easterling et al., 2000).
Tree-ring data offers distinct advantages for paleoclimatic reconstruction, including long-term coverage, annual to seasonal high resolution, and high continuity. These features make them a valuable proxy for studying historical climate variability (
Fritts, 1976;
Briffa et al., 1990;
Cook et al., 2004;
Shao et al., 2010). Due to the lack of observational data, long-term variations in winter temperature remain poorly understood. In recent years, considerable progress has been made using tree-ring records to reconstruct winter temperature patterns or examine their links to tree radial growth across many regions, including north-western Europe (
D’Arrigo et al., 1993), Romania (
Popa and Cheval, 2007;
Sidor et al., 2015), Turkey (
Heinrich et al., 2013), Poland (
Opała and Mendecki, 2014;
Pritzkow et al., 2016), Hungary (
Misi and Nafradi, 2016), the south-western United States and Canada (
Fritts et al., 1979), Mongolia (
Jacoby et al., 1996;
Hauck et al., 2016), East Asia (
Wang et al., 2023), and the Qinghai-Xizang Plateau (
Gou et al., 2007;
Huang et al., 2019;
Gaire et al., 2020). These studies provide key support for understanding past climate dynamics and predicting future trends. They reveal that some regions experienced notable winter warming during the 20th century (
Gou et al., 2007;
Hochman et al., 2016;
Shah et al., 2019;
Li et al., 2021;
Wang et al., 2023), while others showed no clear warming trend (
Heinrich et al., 2013;
Opała and Mendecki, 2014;
Kvaratskhelia and Gavashelishvili, 2025). This variability underscores the strong regional heterogeneity of winter climate responses and suggests that distinct climatic drivers are at play in different areas. However, high-resolution reconstructions of winter temperature based on tree-ring data remain limited in the arid south-western regions of Central Asia, including the area near the Caspian Sea.
The Caspian Sea region, located at the junction of the Eurasian continent, experiences strong seasonal and spatial climatic variability (
Arpe et al., 2000) and is recognized as a key area for studying climate change. Recent observations indicate that the southern Caspian region—primarily northern Iran and Azerbaijan—has experienced rising temperatures and decreasing precipitation (
Kazemzadeh and Malekian, 2018;
Fathian et al., 2020). These climatic changes pose growing challenges for ecosystems and human systems, emphasizing the need for targeting climate research and adaptation strategies in this region.
Significant progress has been made in dendroclimatological research in the Caspian Sea region, with many scholars reconstructing historical temperature and moisture variability (
Bayramzadeh et al., 2018;
Foroozan et al., 2020;
Opała-Owczarek et al., 2021;
Arsalani et al., 2022;
Kvaratskhelia and Gavashelishvili, 2025). However, studies in the south-western Caspian region remain relatively limited. There is a clear need for additional data and reconstructions to improve spatial coverage and inform regional climate models. This study addressed that gap by reconstructing WMT variability in the south-western Caspian region using tree-ring width indices. It evaluates the spatial representativeness of the reconstruction and explores linkages with broader atmospheric circulation patterns, such as the North Atlantic Oscillation (NAO) and El Niño–Southern Oscillation (ENSO). The goal is to enhance our understanding of regional climate variability, improve the spatial coverage of dendroclimatic records, and support future climate predictions and early-warning efforts. Additionally, this research contributes to identifying key teleconnections and feedback mechanisms among large-scale climate systems.
2 Material and methods
2.1 Tree-ring data and chronology
The sampling site is located in the south-western Caspian region, in southern Azerbaijan near the Lerik area (38.71°N, 48.53°E) (
Seyfullayev et al., 2021), within the Hyrcanian forest close to the village of Hamarat (Fig. 1, Wang et al., 2025b). The sampled species is Common Yew (
Taxus baccata L.), growing on a south-west-facing slope at elevations between 1500 and 1600 m. The soil at the site is stony and strongly leached. Sampling took place during the summers of 2016 and 2017, yielding a total of 23 cores from 23 visually healthy trees (available at the NOAA Paleoclimatology website).
Seyfullayev et al. (2021) found that tree-ring width is primarily influenced by winter temperatures, highlighting its dendroclimatological potential, however, no climate reconstruction was performed.
Tree-ring chronologies were developed using the ARSTAN program (
Cook, 1985). A negative exponential curve and cubic smoothing splines with 50% frequency-response cutoffs at wavelengths of 50 and 100 years were each fitted to remove non-climatic growth trends related to tree age, genetic factors, and competition. Based on the standardized series, the standard (STD) and residual (RES) chronologies were subsequently developed. All chronologies cover the period from 1867 to 2016 (Fig. 2). Statistics and the common period (1960–2016) for the chronologies are given in Table 1.
In principle, higher values of M.S., R, SNR, and EPS indicate stronger common signals among individual series and better representation of the growth characteristics of the sampled tree population (
Fritts, 1976). As shown in Table 1, these statistics are relatively high across the six developed chronologies, suggesting that the sampled trees exhibit strong representativeness. Among them, the SPL50RES chronology exhibits the highest values of R, SNR, and EPS, and is therefore selected for further dendrochronological analysis. The starting year of SPL50RES chronology corresponding to a subsample signal strength (SSS) > 0.85 (
Wigley et al., 1984) is 1879.
2.2 Climate data
Climate data were sourced from the Climatic Research Unit (CRU) TS 4.08 data set provided by the Climatic Research Centre of the University of East Anglia, with a spatial resolution of 0.5° × 0.5° (
Harris et al., 2020). Monthly data for total precipitation, mean temperature, mean maximum temperature, and mean minimum temperature were extracted for the grid cell (38.5°–39.0° N, 48.0°–48.5° E) encompassing the sample site. Additional observational data is available from nearby meteorological stations. The Lankoran City Airport station, located approximately 30 km from the site, provides records from 1979 to 2011 (
Seyfullayev et al., 2021), while the Ardebil station, about 55.5 km away, offers data from 1951 to 2005, albeit with some missing months. Considering the coverage of the tree-ring chronology, the CRU climate data used in this study span the period from 1951 to 2016.
Seasonal patterns (Fig. 3) show average annual precipitation of 625.3 mm at the sampling site. Precipitation increases from January to May, peaking in May (70.4 mm), then dropping sharply in July (29.0 mm). After August, precipitation gradually rises again, reaching a second peak in October (77.8 mm), before declining into December (39.7 mm). The region’s mean annual temperature is 10.0°C, with July being the warmest month (21.5°C). This seasonal cycle reveals a mismatch between peak temperatures and moisture availability: the hottest months coincide with the driest conditions. Notably, while winter is the coldest season, it is not the driest – precipitation levels in winter exceed those in mid-summer.
To assess the relationship between tree radial growth and climate, monthly values of precipitation, mean temperature, mean maximum temperature, and mean minimum temperature were examined over an 18-month window from the previous June to the current November.
2.3 Statistical methods
Pearson correlation analysis was used to quantify the relationship between tree-ring width and climatic variables. Scatter plots were used to analyze the linear relationship between the tree-ring index and limiting factors. Following preliminary testing, a linear regression approach was used to develop a transfer function linking the tree-ring width index with climatic variables, enabling the reconstruction of high-resolution historical climate patterns. The model was validated using cross-validation techniques (
Michaelsen, 1987) and split-period calibration/verification analysis (
Meko and Graybill, 1995). Evaluation statistics included the sign test (ST), first-difference sign test (FST), product means test (PMT), the reduction of error (RE) metric, and coefficient of efficiency (CE). Spectral analysis (
Huang, 2000) was conducted to identify the dominant periodicities in the reconstructed climate series. Cross wavelet transform (XWT) and wavelet coherence (WTC) (
Jevrejeva et al., 2003;
Grinsted et al., 2004) were used to examine the multi-scale coupling between the reconstruction series and NAO index (1879–2015; available at Prof. Jianping Li’s homepage website) and the Niño 3.4 sea surface temperature (SST) index (available at the NOAA Physical Sciences Laboratory website), widely accepted as a standard indicator of ENSO (
Trenberth, 1997).
3 Results
3.1 Correlation between tree-ring width and climate data
Figure 4 presents the correlation analysis between the tree-ring width index and various climatic variables. Overall, the tree-ring width index exhibits a stronger correlation with temperature variables than with precipitation.
Regarding precipitation, significant negative correlations (p < 0.05) are observed for January and February, while significant positive correlations (p < 0.05) occur for May and June. Further analysis using seasonal (spring, summer, autumn, winter) and combined-period totals reveals that the most significant negative correlation is with total precipitation during January–February (p < 0.01). The strongest positive correlation is found with total precipitation during May–June (r = 0.408, n = 66, p < 0.01).
For temperature, the tree-ring width index is positively correlated with monthly mean, maximum, and minimum temperatures, particularly from the previous December through the current March. Positive correlations with mean and maximum temperatures from January to March reach the 0.01 significance level. Similarly, minimum temperatures from the previous December through February also show significant positive correlations at the 0.01 level. Among seasonal and combined-period temperature variables, the tree-ring width index shows the highest correlation with winter minimum temperature (December–February), with a correlation coefficient of r = 0.624 (n = 65, p < 0.001). This strong relationship indicates that the tree-ring width index can serve as a robust proxy for reconstructing historical winter minimum temperature (WMT) variability.
3.2 Development of a transfer function between tree-ring width and WMT
A linear regression model was developed to reconstruct winter minimum temperature (WMT) using the current year’s tree-ring width index (TRW) as the predictor:
This model explains 39.0% of the variance in WMT, with an adjusted R2 of 0.380. The model is statistically significant at p < 0.001, and the F-statistic (40.199) further supports the model’s robustness. Figure 5 shows the scatter plot comparing tree-ring width index and observed WMT, and a comparison of observed and reconstructed WMT.
To assess the model’s reliability, cross-validation and split-period calibration/verification analyses were conducted. The calibration periods were 1952–1983 and 1984–2016, respectively. Results are summarized in Table 2.
The full-period validation indicates excellent agreement between observed and reconstructed WMT. The values of ST and FST are 44 and 51, respectively, both reaching the 0.01 significance level. In the split-period analyses, all FST values exceed the 0.05 threshold, confirming the model’s ability to capture high-frequency variability. The PMT value is also significant (3.996, p < 0.01), supporting the consistency of the reconstruction. The RE and CE values of 0.347 further validate the model’s stability. In summary, the transfer function demonstrates strong statistical support and stable performance, indicating that the tree-ring width index is a reliable proxy for reconstructing historical WMT variability.
3.3 Reconstructed WMT from 1879 to 2016
Using the transfer equation described above, we reconstructed WMT variations from 1879 to 2016 (Fig. 6). The reconstruction shows substantial interannual variability. The five coldest years were 1942, 1925, 1964, 1911, and 1950, while the five warmest years were 1924, 1963, 1881, 1940, and 1901.
The reconstructed mean WMT is –4.68°C, with a standard deviation (δ) of 0.93. Years exceeding the mean by more than +1.0δ are classified as warm periods, while those below mean −1.0δ are considered cold periods. Over the past 138 years, extended warm periods (> 5 years) occurred in 1886–1890, 1914–1918, 1921–1925, 1933–1939, 1961–1965, 1975–1980, and 1997–2005. The 1997–2005 period was the longest, while 1975–1980 was the warmest. Cold periods of more than five years include 1903–1907, 1909–1913, 1926–1932, 1948–1953, 1955–1960, 1969–1974, 1981–1985, 1987–1992, and 2006–2013, with 2006–2013 and 1926–1932 being the first and second longest, respectively. The curve exhibits distinct variability in winter minimum temperatures in the study area since 1879.
Power spectral analysis revealed significant periodicities in WMT variations at ~5.41 years (
p < 0.05), 4.18 years (
p < 0.05), 5.75 years (
p < 0.10), 4.38 years (
p < 0.10), 7.67 years (
p < 0.15), 7.08 years (
p < 0.15), 4.0 years (
p < 0.15), and 2.49 years (
p < 0.15). The identified cycles at 5.41, 4.18, 5.75, 4.38, 4.0, and 2.49 years fall within the typical 2.5–7-year periodicity of ENSO (
Trenberth, 1997;
Rittenour et al., 2000;
Huber and Caballero, 2003), suggesting a possible influence of ENSO on the interannual variability of WMT in the study area. In addition, the longer cycles (~7.67 and 7.08 years) may be associated with NAO (
Rogers, 1984;
Hurrell and Van Loon, 1997).
4 Discussion
4.1 Relationship between tree-ring growth and climate
The study area, located in the south-western Caspian Sea region, shows that tree-ring width indices are significantly and negatively correlated with January–February precipitation, but positively correlated with May–June precipitation. In terms of temperature, tree-ring width is positively correlated with temperatures in the previous December and the current January through March. The negative correlation with winter precipitation is likely due to increased snowfall, which can delay spring soil thaw and shorten the growing season by reducing surface temperatures through increased albedo. This leads to narrower tree rings. In contrast, higher precipitation in late spring and early summer supports active cell division and expansion during the peak growing period, resulting in wider rings. Positive correlations between tree-ring width and WMT indicate that higher WMT sustains metabolic activity and prevents root damage, allowing trees to resume growth more rapidly in spring (
Fritts, 1976). Conversely, extremely low winter temperatures can cause frost or freeze damage, hindering growth and resulting in narrower rings (
Muffler et al., 2024). The strong positive correlation with March minimum temperature suggests that warmer early-spring conditions promote earlier onset of growth, effectively extending the growing season and enhancing radial growth. Overall, tree radial growth in the study area is strongly governed by local hydroclimatic conditions, with winter temperature and late spring to early summer precipitation being the primary limiting factors. Among these, WMT shows the strongest correlation, underscoring its central role in controlling tree radial growth.
4.2 Analysis of the reconstructed WMT
4.2.1 Spatial representation
Spatial correlations between observed WMT and CRU gridded WMTs (1952–2016) reveal strong positive correlations (r > 0.5) across much of West Asia, particularly the Iranian Plateau and Mesopotamian region (Fig. 7(a)). Significant correlations also appear across Central Asia, north-western South Asia, and western China, including the Tianshan Mountains and the Tarim Basin, indicating broad spatial representativeness. Moderate correlations (r = 0.3−0.5) are found in north-eastern North Africa, Eastern Europe, western Turkey, south-western China, and central India, suggesting regionally consistent temperature variability. In contrast, significant negative correlations occur in northern Europe (especially Scandinavia, the Baltic States, north-western Russia), implying inverse temperature relationships. No significant correlations were found for East Africa, central South Asia, or south-eastern China (Fig. 7(a)).
The reconstructed series shows strong positive correlations (r > 0.5) across nearly all of West Asia, especially from the Mesopotamian Basin to the western Iranian Plateau, and extending into south-western Central Asia and central Anatolia (Fig. 7(b)). Moderate correlations (r = 0.3−0.5) are observed in eastern Iran, central Afghanistan, western Xinjiang (China), northern Sudan, eastern Egypt, the Mediterranean coast west of Syria, the Caucasus, and south-western Kazakhstan. Compared to instrumental data, the reconstructed series shows a slightly smaller area of strong correlation and generally lower coefficients. Weak correlations appear in northern India, eastern Xinjiang, the western Qinghai-Xizang Plateau, and along the Saharan margins. Negative correlations with northern Europe persist.
Overall, areas of strong correlation correspond to the Central Asia–West Asia region, encompassing parts of the core area of the westerly-dominated climate model region (
Chen et al., 2019). This region shows the most stable spatial correspondence with WMT signals from the study area. In contrast, monsoon-dominated areas such as northern India, the southern Qinghai-Xizang Plateau, and south-western China show weak or fragmented correlations, suggesting limited climatic connectivity. Therefore, the reconstructed WMT series shows strong spatial representativeness of winter temperature variability within the westerly-dominated region of Central and West Asia.
4.2.2 Comparison of the reconstruction with winter temperature changes in surrounding areas
The reconstructed
WMT series indicates a gradual warming trend beginning in the 1980s, followed by a reversal and cooling in the early 21st century. Despite these fluctuations,
WMT variability has remained relatively stable, likely due to the high-frequency signals preserved in the RES chronology. This pattern contrasts with tree-ring-based reconstructions in regions such as East Asia (
Wang et al., 2023), which show more pronounced warming trends, underscoring the complex and heterogeneous nature of regional climate responses. Reconstructions from nearby areas, including the Alborz Mountains (
Gholami et al., 2017;
Bayramzadeh et al., 2018) and Turkey (
Heinrich et al., 2013), also diverge from the broader global warming trajectory.
To assess regional coherence and divergence, we compared our
WMT reconstruction with January–March mean maximum temperatures from south-western Iran (
Arsalani et al., 2022) and October–March mean temperatures from Georgia (Caucasus region) (
Kvaratskhelia and Gavashelishvili, 2025) (Fig. 8). Most warm periods in our reconstruction correspond with peaks in these regional data sets, though some discrepancies exist. For instance, the 1880s and late 1970s warm periods are less distinct in south-western Iran, which instead experienced a sustained warming from the 1930s to the mid-1940s. In the Caucasus, warm phases during the 1960s and late 1970s are muted, and a prolonged cooling trend extends from the late 1940s to early 1990s, with only a slight rebound thereafter. Notably, the timing of peak warmth differs: the late 1970s in our records, the 1960s in south-western Iran, and the 1940s in the Caucasus. These differences likely reflect variations in reconstruction periods, geographic setting, topographic complexity, and elevation. Compared with the other two series, our series also displays comparatively subdued fluctuations, not due to detrending methods but rather to the inherent growth characteristics of the sampled trees and the region’s relatively stable winter climate. Consequently, the tree-ring-derived WMT variations exhibit a relatively muted pattern.
Despite the limited length of our 137-year reconstruction, the transfer function used explains a high proportion of variance, and the reconstructed series demonstrates strong spatial representativeness. These results contribute valuable regional insights into the spatial heterogeneity of climate change and help fill geographical gaps in the understanding of WMT variability.
4.2.3 Large-scale circulation impacts on WMT
The reconstructed WMT is significantly negatively correlated with the NAO index in January (r = − 0.195, p < 0.05) and February (r = − 0.196, p < 0.05), with an even stronger negative correlation for the January–February average (r = − 0.241, p < 0.05). Positive correlations are observed with the Niño 3.4 SST index in the previous November (r = 0.186, p < 0.05), and the current January (r = 0.189, p < 0.05) and March (r = 0.205, p < 0.05)
The XWT and WTC results between WMT and the NAO (Figs. 9(a) and 9(b)) show significant coupling at multiple time scales. Between 1930 and 1960, strong coherence appears in the 6–12-year band, and phase arrows mostly point left or upper left, indicating a negative correlation, with NAO slightly leading temperature. Additional high-coherence regions appear in the early 1900s and post-1970s at both shorter (2–4 years) and longer (~16 years) periods. These findings suggest a multi-scale influence of the NAO, particularly on decadal variability. Higher NAO values correspond to colder winters, consistent with previous studies linking the NAO’s positive phase to stronger westerlies and increased cold air intrusions across the region, including the Zagros Mountains (
Arsalani et al., 2022) and the broader Mediterranean (
Scaife et al., 2008).
XWT and WTC analyses between WMT and Niño 3.4 SST (Figs. 9(c) and 9(d)) reveal similarly strong coupling, especially from 1920 to 1960 in the 6–12-year band, and phase arrows mostly point to the lower right, indicating a positive correlation, with SST changes slightly leading temperature. Intermittent coherence in the 2–4-year band is also observed during the late 1940s and 1990s, indicating episodic ENSO influence at interannual timescales. Positive correlations between WMT and Niño 3.4 SST suggest that El Niño events – associated with higher SSTs – correspond to warmer winters in the region. This may reflect weakened Siberian high pressure systems and a diminished East Asian winter monsoon during El Niño phases (
He and Wang, 2013), which reduce the southward flow of cold air and lead to milder winter temperatures. Similar findings have been reported across East Asia over the past six centuries (
Wang et al., 2025a), where El Niño years generally bring warmer winters, and La Niña years bring colder ones.
In sum, these analyses further confirm that both NAO and ENSO significantly modulate WMT variability in the south-western Caspian Seas region. The interplay between Arctic, mid-latitude, and tropical systems through large-scale teleconnections plays a critical role in shaping the region’s winter climate.
5 Conclusions
This study found that tree-ring width indices in the Lerik region, located in the south-western Caspian Sea area, are significantly influenced by climatic factors. Specifically, tree radial growth is positively correlated with late-spring to early-summer precipitation and WMT, and negatively correlated with winter precipitation. Among these, the strongest relationship is with WMT, indicating that WMT is the primary climatic driver of tree radial growth in this region.
Leveraging this strong correlation, we reconstructed WMT variability from 1879 onward. The reconstruction identifies several multi-year warm periods (lasting more than five years), including 1886–1890, 1914–1918, 1921–1925, 1933–1939, 1961–1965, 1975–1980, and 1997–2005. Of these, the 1997–2005 period was the longest and 1975–1980 was the warmest. Despite these episodes, the reconstruction remains within a relatively stable range, aligning with similar tree-ring-based reconstructions from the nearby Alborz Mountains and Turkey with no warming trend. This consistency underscores both the reliability of our results and the spatial heterogeneity in regional responses to global climate change.
Future efforts should focus on identifying longer and well-preserved chronologies in this region. Extending the reconstruction further back in time will enhance our understanding of historical climate variability and improve insight into the complex and localized nature of warming patterns in response to global climate change.
PDF
(6232KB)
