Teberda valley runoff variability (AD 1850–2005) based on tree-ring reconstruction (Northern Caucasus, Russia)

V V Matskovsky1, E A Dolgova1 and O N Solomina1
1 Institute of Geography, Russian Academy of Sciences, 29 Staromonetniy pereulok, Moscow, Russia
E-mail: matskovsky@gmail.com
Abstract. Seven tree-ring chronologies are used to reconstruct Teberda river (Northern Caucasus, Russia) smoothed runoff for May, July and August. Six chronologies were developed from Pinus sylvestris and one from Abies nordmanniana. Tree growth is significantly, but weakly, correlated with maximum temperatures (negatively) and relative humidity (positively) during summer.All possible combinations of seven predictors were tried to get the best result on the cross-validation. Two of three reconstructions showed high wavelet coherence with instrumental data on decadal timescales and were analysed for spectrum stability. Minima of moving trends at the end of the reconstructions along with weakening of decadal cycles may be a marker of significant change of Teberda river hydrological regime during the second half of the 20th century.
1. Introduction
Teberda river is a tributary of Kuban' (Azov Sea basin), the largest river of Krasnodarskiy Kray, an important agricultural region of the Russian South. Teberda is 60 km long with the watershed surface equal to 1080 km2, mean watershed altitude is 2210 m. 55.8% of Teberda river’s runoff (measured at Teberda hydrological station) is provided by snow and ice melt. 60% of runoff occurs in summer, 17% – in the fall, 5% – in winter, 18% – in spring. Mean annual water discharge is 26.8 m3/sec [1]. Teberda monthly runoff correlates with spring temperature (up to 0.62 for April) and with autumn precipitation (up to 0.72 for October). Significant negative trends were identified both in summer and full year runoff for AD 1927-2005. Extending instrumental observations into the past is essential for understanding of the runoff long-term variability and response to changing climate.
2. Materials and methods
2.1. Tree-ring data
The study area is located in the Caucasus Mts. at the territory of Teberda Biosphere reserve and Elbrus National Park. The current timberline of Scots pines (Pinus sylvestris L.) rises up to the elevation of 2500-2700, depending on the slope orientation and local environmental conditions. Tree ring samples (cores) were collected from nine sites in the vicinity of the upper-tree limit (from 2200 to 2500 m a.s.l.) and at glacier’s forefields. Location of these sites are shown on the figure 1.
Living trees of Scots pine (Pinus sylvestris L.) and fir (Abies nordmanniana (Stev.) Spach) were sampled according to standard dendroclimatological principles and conventions [2], [3]. LINTAB equipment was used for measuring of the annual ring-width with precision of 0.01 mm. To cross-date tree-ring series and for graphical comparison we used Rinntech TSAP 0.53 software. To cross-date, check for missing rings or dating errors COFECHA software was used [4]. Samples that showed low correlation (<0.33) with master-chronology were excluded from the analysis. The final dataset comprised 145 tree-ring series.
We used program ARSTAN software [5] for developing the chronologies. Tree-ring series were standardized using a conservative procedure to maintain in the series as much low-frequency variance as possible. Negative exponential or linear curves were used for removing the non-climatic signals.
 
image description
Figure 1. Tree-ring sites (yellow) and Teberda river valley (green). Elbrus mount (white) is clearly seen at the right side of the figure.
 
The ability of each chronology to represent the ideal population signal was assessed using the mean between-tree correlation and the expressed population signal (EPS) statistics [6].
2.2. Runoff data
Monthly values of Teberda river runoff at Teberda hydrological station for AD 1927-2005, corrected for nonstationarity were provided by Teberda Biosphere reserve. They contain missing values from August of 1941 to December of 1947 and from January of 1971 to January of 1972.
2.3. Methods
We built principal component chronology to extract a common climatic signal from our local chronologies. It showed the strongest response to climatic variability (calculations in Dendroclim2002 [7]), correlating negatively with May-July maximum temperature and positively with May-July relative humidity. A significant correlation (at 99.9% confidence level) is identified between PC chronology indices and July Teberda river runoff (r=0.41). But none of the chronologies showed really strong response to any individual climatic variable. However, the absence of a strong linear correlation with a single climate variable does not mean that tree-growth in this region is not sensitive to climate [8].This suggests more complex tree-climate relationships, but this complexity shouldn’t stop our efforts to get as much as possible from determined connections.
We used principle component regression [9] to generate the runoff reconstruction models, separately for each month and for the full year. Seven standard chronologies that are listed below were used as predictors. To determine the best set of predictors, all possible combinations were tried and the set, that showed maximum R2 statistics on leave-one-out cross-validation, was chosen.
Wavelet coherence analysis [10] was applied to determine common properties of instrumental and reconstructed data. Continuous wavelet transform with the Morlet base wavelet [11] was used for analysis of spectral characteristics of reconstructions through time.
Significance of all trends was tested under the assumption of signal as trend plus AR(1) noise.
Missing values of instrumental runoff data were ignored in all the analyses.
3. Results
3.1. Chronologies
We developed 6 pine (KHTP, KV, KYZ, CHS, GAZ and BAZ) and 1 fir (ALI) ring-width chronologies. KHTP chronology consists of samples from KHTP and KHAT sites, CHS chronology consists of samples from CHS and CHE sites (figure 1). Commonly used statistics such as standard deviation, first order autocorrelation, mean sensitivity that is a measure of the relative difference in width between consecutive rings [2] are listed in table 1. All the chronologies are characterized by relatively low mean sensitivity and high autocorrelation.
3.2. Reconstructions
Annual reconstructions showed unreliable results on cross-validation that’s why we smoothed reconstructed series with 10-years moving average filter. Three of thirteen smoothed reconstructions (for each month and for the full year) showed R2 higher then 0.6 on cross-validation. They are: May runoff (the best set of predictors is ALI, BAZ, CHS and KHAT, R2=0.6487), July runoff (ALI, BAZ and KHAT, R2=0.6567) and August runoff (ALI, CHS and KHAT, R2=0.6615) (figure 2a). During these months occurs 12%, 22% and 17% of annual runoff correspondingly (according to instrumental records), all together more than a half. We considered only those parts of the reconstructions, for which EPS of all chronologies-predictors is higher than 0.8. So, in all the cases, ALI chronology limited them to AD 1850.
Wavelet coherence analysis was applied to instrumental and reconstructed (not smoothed) data to estimate the ability of reconstruction to reproduce low-frequency variations. May and July reconstructions showed strong and stable coherence with instrumental records on periods more than 25 years, August reconstruction didn’t (figure 2b).
Table 1. Tree-ring chronologies statistics.
Site name
KYZ
KV
KHTP
GAR
CHS
BAZ
ALI
First year
1550
1640
1678
1693
1738
1660
1800
Last year
2006
2002
2005
2002
2002
2004
2005
Cores (trees)
20 (10)
8 (4)
57 (29)
13 (7)
15 (8)
18 (11)
24 (13)
Inter-series correlation
0.649
0.551
0.507
0.614
0.526
0.436
0.282
Standard deviation
0.221
0.176
0.221
0.221
0148
0.140
0.125
Mean sensitivity
0.157
0.125
0.155
0.128
0.110
0.119
0.118
1st order autocorrelation
0.648
0.550
0.505
0.612
0.524
0.435
0.282
Mean segment length
227
283
175
160
160
233
114
EPS > 0.8
1797
1863
1752
1939
1833
1768
1850
 
 
Two of three reconstructions and sum of three reconstructed months rich their minimum values in 1997 (ignoring last 5 years because of 10-years smoothing). May’s reconstruction has the second lowest value in 1990.
Continuous wavelet transform for May and July reconstructions (not smoothed) (figure 2d) both show weakening of low frequency cycles (16 to 32 years periods) for the years after AD 1927 comparing to the previous years. We didn’t perform wavelet analysis for August reconstruction because of weak wavelet coherence of reconstructed and instrumental data. For wavelet spectra diagrams watery colors show the cone of influence, black border shows regions of greater than 95% confidence for red-noise process.
 

Figure 2. (a)
Smoothed instrumental runoff (green) and its reconstruction (black). (b) Wavelet coherence of instrumental and reconstructed runoff, AD 1927-2002. (c) Moving trends of the reconstructed runoff (76 years window), AD 1850-2002. Year corresponds to the first year of the window. For (a), (b) and (c) the upper picture refers to July, middle to August and the lower to May.
(d) Continuous wavelet transform of the reconstructed runoff.
 
3.3. Trends
Runoff of all three months (May, July and August) as well as full year’s runoff have significant negative trends for the period AD 1927-2005, with values -0.063, -0.140, -0.165 and -0.429 x106 m3/year correspondingly.
We calculated moving trends for our three reconstructions (without smoothing) and for their sum for AD 1850-2002 in 76 years moving window. The years on the plots correspond to the left side of the window, namely 1850 means years from 1850 to 1925 (figure 2c). All trends showed their minimum values for the last window (AD 1927-2002). Trends for other moving windows (60, 50 and 40 years, not shown) sometimes have minimum values in another (not last) window, but not more than for one reconstruction for each window. Sum of the reconstructions showed its minimum in the last window in all the cases.
4. Discussion and summary
Although we didn’t create annually resolved quantitative reconstructions of Teberda river runoff because of insufficient correlation between tree growth and hydrological parameters, our smoothed reconstruction of May, July and August runoff can tell much about low-frequency variations of these parameters.
Almost all explored characteristics, such as trends, spectra and smoothed values of the series have minima near the end of the reconstructions. These extremes along with weakening of decadal cycles may be a marker of significant change of Teberda river hydrological regime.
We made our reconstructions only back to AD 1850 yet some of our chronologies begin in 16th century. Adding new samples to these chronologies will increase common signal and let us extend our reconstructions into the past.
References
[1]          Lurye P M, Panov V D and Tkachenko Yu Yu 2005 Kuban’ river: hydrography and runoff regime (in russian) (Sankt-Petersburg: Hydrometeoizdat)
[2]          Fritts H C 1976 Tree Rings and Climate (London: Academic Press)
[3]          Cook E R and Kairiukstis L A 1990 Methods of Dendrochronology: applications in the environmental sciences (Dordrecht: Kluwer)
[4]          Holmes R L 1983 Computer-assisted quality control in tree-ring dating and measurement Tree-Ring Bulletin 43 69-78
[5]          Cook E R and Holmes RL 1984 Program ARSTAN and Users Manual (NY: Lamont Doherty Geological Observatory)
[6]          Wigley T M L, Briffa K R and Jones P D 1984 On the average value of correlated time series, with applications in dendroclimatology and hydrometeorology. J. Climate Appl. Meteor. 23 201–13
[7]          Biondi F and Waikul K 2004 DENDROCLIM2002: A C++ program for statistical calibration of climate signals in tree-ring chronologies Computers & Geosciences 30 303–11
[8]          George S S, Meko D M and Evans M N 2008 Regional tree growth and inferred summer climate in the Winnipeg River basin, Canada, since AD 1783 Quat. Res. 70 158–72
[9]          Aivazyan S A, Yenyukov I S and Meshalkin L D 1985 Applied statistics. Study of relationships (in russian) (Moscow: Finansy i Statistika)
[10]      Grinsted A, Moore J and Jevrejeva S 2004 Application of the cross wavelet transform and wavelet coherence to geophysical time series Nonlinear Processes in Geophysics 11 5/6
561-6
[11]      Torrence C and Compo G P 1998 A practical guide to wavelet analysis Bull. Amer. Met. Soc. 79 1 61–78

← back