Climate Dynamics

, Volume 26, Issue 2, pp 295–303

Tree-ring-based hydrological records for western Himalaya, India, since a.d. 1560

Authors

    • Birbal Sahni Institute of Palaeobotany
    • School of Forest Resources, College of AgricultureChungbuk National University
  • Won-Kyu Park
    • School of Forest Resources, College of AgricultureChungbuk National University
  • Ram R Yadav
    • Birbal Sahni Institute of Palaeobotany
Article

DOI: 10.1007/s00382-005-0089-1

Cite this article as:
Singh, J., Park, W. & Yadav, R.R. Clim Dyn (2006) 26: 295. doi:10.1007/s00382-005-0089-1

Abstract

We analysed 565 increment cores from 325 Himalayan cedar [Cedrus deodara (Roxb.) G. Don] trees growing at 13 moisture-stressed, widely distributed sites in the western Himalayan region. We found a strong positive relationship between our tree-ring width chronologies and spring precipitation which enabled us to reconstruct precipitation back to a.d. 1560. This reconstruction is so far the longest in this region. The calibration model explains 40% variance in the instrumental data (1953–1997). The most striking feature of the reconstruction is the unprecedented increase in precipitation during the late twentieth century relative to the past 438 years. Both wet and dry springs occurred during the Little Ice Age. A 10-year running mean showed that the driest period occurred in the seventeenth century while the wettest period occurred in the twentieth century. Spectral analysis of the reconstructed series indicated a dominant 2-year periodicity.

1 Introduction

India is the seventh largest and second most populous country in the world, where most of the population lives in rural areas (Office of the Registrar General 2001) with their livelihood largely dependent on rain-fed agriculture. Excess rainfall or failure of summer monsoon rainfall can affect millions of people in India every year (Gadgil 1996). The intensity of summer monsoon rainfall over India is controlled by the thermal contrast between land and sea (Webster et al. 1998). The Himalaya, the highest mountain system on the earth, plays a significant role in building this land–ocean thermal contrast that helps drive south Asian summer monsoon rainfall (Prell and Kutzbach 1992; Douville and Royer 1996). Long-term high-resolution proxy climate records from the region would provide valuable information on the variability of past monsoon activity and its relevance to human society.

Tree-ring samples of Himalayan cedar were collected from a wide geographic area in the western Himalaya. Regular augmentation of our tree-ring data with successive years’ collections helped to develop a well-replicated chronology network. These chronologies prepared from moisture-stressed sites have been used to develop several century long temperature reconstructions (Yadav and Singh 2002; Yadav et al. 2004). However, the precipitation reconstructions for the region are few and either based on single predictor chronology (Yadav and Park 2000) or too short in length (Hughes 1992; Singh and Yadav 2005) to put variability in precipitation in longer perspective. With our recent extensive collections several chronologies have been extended back by several centuries with sufficient sample replication. A common chronology period with sufficient sample replication covering the major part of the Little Ice Age period [~a.d. 1450–1850 (Grove 1988)] has been used here to develop spring precipitation reconstruction for the first time for western Himalaya. It is hoped that such climatic information for the Himalayan region will greatly help in understanding the precipitation behaviour over the region and its relationship with climate phenomena operating over other distant regions.

2 Data and methods

The network of tree-ring data used here was developed from Himalayan cedar trees growing in 13 distantly located, moisture-stressed natural forest stands (Fig. 1). Rocky hill slopes with thin soil cover characterize these forest sites. The sites located in the interior Himalayan area distant from human habitations are least disturbed. Two increment cores from healthy trees with no visible disturbance were taken at breast height (1.4 m) from the stem with the aim to extract all the growth rings present in trees, but in several cases this was not possible due to limitations with the corer length or heart rot.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig1_HTML.gif
Fig. 1

Location map of the tree-ring sites (numbered dark triangles) and meteorological stations (circles with dot in centre)

Increment cores were processed in the laboratory using standard dendrochronological techniques (Fritts 1976). Skeleton plotting was used to assign the exact calendar year to each growth ring (Stokes and Smiley 1968). Ring widths were measured with 0.01 mm accuracy using a linear encoder attached to a personal computer. Dating accuracy was cross-checked and possible measurement errors were identified using the quality control programme COFECHA (Holmes 1983), which identifies tree-ring segments that may have dating or measurement errors. Samples having dating problems were rechecked and errors corrected. A total of 565 cores from 325 trees derived from 13 sites were selected for analysis (Table 1).
Table 1

General details of Cedrus deodara chronologies used in reconstruction

1

2

3

4

5

6

7

1

Harshil

2,730

709 (1294–2002)

42/28

0.34

0.16

2

Dharali III

2,720

478 (1522–1999)

28/15

0.32

0.09

3

Dharali IV

3,060

478 (1522–1999)

32/16

0.37

0.12

4

Dharali V

2,960

586 (1414–1999)

19/11

0.38

0.18

5

Mukhaba

2,820

442 (1558–1999)

73/41

0.27

0.08

6

Jangla

2,900

1,051 (952–2002)

53/31

0.36

0.11

7

Kopang

2,875

797 (1206–2002)

57/34

0.37

0.17

8

Bhaironghati

2,880

716 (1287–2002)

24/15

0.29

0.13

9

Karchha

3,020

1,232 (771–2002)

50/28

0.53

0.21

10

Gangotri III

3,250

444 (1556–1999)

41/23

0.42

0.14

11

Gangotri IV

3,200

626 (1374–1999)

25/14

0.48

0.15

12

Juma

2,780

664 (1340–2003)

65/39

0.43

0.07

13

Kosa

2,800

810 (1194–2003)

56/30

0.47

0.06

1 Serial no., 2 site, 3 altitude (m asl.), 4 chronology length (year) and period (a.d.), 5 sample numbers (cores/trees), 6 correlation lag 1 year, 7 correlation lag 2 years

Negative exponential curves or straight lines were used to remove the biological age trend and retain the majority of the climate-related variance in the ring-width series (Cook and Kairiukstis 1990). The computer programme ARSTAN (Cook 1985) was used for standardization of tree-ring series. Indices for each series were derived by dividing the measured value by the predicted curve value for each year. The next step involved the removal of autocorrelation using an autoregressive (AR) model selected on the basis of the minimum Akaike criterion to produce series of tree-ring indices with low-frequency trends now removed. These indices were combined across all series in each year using a bi-weight robust estimation of the mean to discount the influence of outliers. Three chronology types were produced: (1) Standard, based solely on the negative exponential and straight line detrending, (2) Residual, which has only high-frequency variations preserved due to the AR modelling, and (3) Arstan, developed by reincorporating the pooled autoregressive properties back into the Residual chronology. The common period 1560–1999 represented in all 13 site chronologies (Table 1) was selected for further study. The expressed population signal (EPS), which helps determine the minimum number of series that can provide a reliable estimate of the mean chronology (Wigley et al. 1984), was 0.85 or above in all 13 chronologies from a.d. 1600 onward. Cross-correlation analysis, which measures the common variability in chronologies (1560–1999) revealed that all chronologies showed highly significant cross-correlations (Table 2), indicating a common forcing factor (i.e. climate) influencing tree growth.
Table 2

Cross-correlation among standard chronologies (a.d. 1560–1999, p<0.0001)

 

Dharali III

Dharali IV

Dharali V

Mukhaba

Jangla

Kopang

Bhaironghati

Karchha

Gangotri III

Gangotri IV

Juma

Kosa

Harshil

0.86

0.79

0.82

0.86

0.87

0.84

0.85

0.81

0.79

0.80

0.75

0.69

Dharali III

 

0.84

0.78

0.93

0.91

0.92

0.87

0.84

0.82

0.81

0.77

0.75

Dharali IV

  

0.76

0.80

0.81

0.85

0.81

0.81

0.76

0.73

0.72

0.69

Dharali V

   

0.79

0.83

0.80

0.84

0.75

0.66

0.74

0.61

0.58

Mukhaba

    

0.96

0.93

0.90

0.87

0.83

0.83

0.81

0.76

Jangla

     

0.92

0.91

0.88

0.83

0.83

0.76

0.73

Kopang

      

0.93

0.89

0.82

0.78

0.77

0.74

Bhaironghati

       

0.87

0.79

0.80

0.74

0.67

Karchha

        

0.86

0.88

0.77

0.73

Gangotri III

         

0.87

0.75

0.72

Gangotri IV

          

0.74

0.69

Juma

           

0.92

For dendroclimatic studies, climate data from stations near the tree-ring sites are needed. However, in the Himalayan region, meteorological stations are situated relatively at lower elevations and distant from the high-elevation tree-ring sites. Furthermore, climate in Himalayan region varies within short distances because of the complex topography. By using the mean based on various stations, microclimatic variations among stations can be minimized (Blasing et al. 1981; Jacoby et al. 2000). We prepared mean series by averaging two homogeneous data sets from Mukteswar (29°28′N, 79°39′E, 2311 m asl.) and Shimla (31°10′N, 77°17′E, 2205 m asl.) in the western Himalaya over a common period from 1898 to 1998 for dendroclimatic calibrations and reconstruction (Fig. 1).

Cross-correlation analysis was performed to understand the tree growth and climate relationship. Mean monthly temperature and monthly precipitation from the previous year’s October to the current year’s October were correlated with the residual chronology for the common period 1898–1998. The previous year’s climate was considered in the analysis because tree growth during the current year can integrate climate factors from the previous growing season (Fritts 1976). This analysis showed that the precipitation from the previous year’s October, November, December and current year’s March, April, May has a direct positive influence on tree growth at all sites (Fig. 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig2_HTML.gif
Fig. 2

Cross-correlation coefficients between residual ring-width chronology and monthly climate variables (blank vertical bars denote temperature and dark vertical bars denote precipitation). Horizontal dotted lines are 95% confidence levels. The months range from October of the previous year to October of the current year

The existence of strong and significant relationship between precipitation during the spring months and tree growth enabled us to develop a reconstruction of spring precipitation. To obtain maximum possible low-frequency variations in the reconstruction, standard chronologies were used. Because the chronologies showed strong associations during the previous 2 years, we used standard chronologies for years t (13 chronologies), t–1 (12 chronologies) and t–2 (2 chronologies) in a principal component regression analysis. To avoid problems with multicollinearity among the predictor chronologies, the chronologies were transformed into uncorrelated principal components. Principal components with eigenvalues greater than 1.0 were used in the regression analyses. To check the fidelity in tree-ring and spring precipitation relationship, calibrations were performed in different periods (Table 3). The calibration model for the period 1953–1997 captured the highest variance (Radj2=40%); however, it captured only 22% variance in the instrumental data in the 1898–1952 subperiod. Although the calibration for the subperiod 1898–1952 is weaker than 1953–1997, both the models produced significant verification statistics as measured by various tests of similarity (except sign test for the calibration model 1898–1952; Table 3) (Fritts 1976). After this confirmation of the statistical veracity of the models in prediction, we used the 1953–1997 calibration model on the basis of maximum R2 criterion to develop the spring precipitation reconstruction extending back to a.d. 1560. The spring precipitation estimates using this model show close resemblance with the yearly values for the instrumental period (Fig. 3).
Table 3

Calibration–verification statistics obtained in principal component regression analysis

Calibration

Verification

Period

Radj2 (%)

F value

Period

r

T value

Sign test

RE

1898–1947

24

8.08***

1948–1997

0.60***

2.77*

50/27

0.33

1948–1997

36

13.87***

1898–1947

0.50***

4.27*

50/35*

0.29

1898–1952

22

7.80***

1953–1997

0.62***

2.62*

45/27

0.36

1953–1997

40

14.73***

1898–1952

0.47**

3.33*

55/41*

0.19

1898–1997

32

23.24***

     

Radj2 is captured variance in the instrumental data adjusted for degrees of freedom; F is Fisher value; r is Spearman correlation coefficient; T value is used to test the difference between the mean of the positive products and the mean of the negative products (disregarding the sign); Sign test is a measure of the association of the signs of first differences of actual and estimated data. This gives indication of the extent to which the reconstruction reproduces high-frequency components of the instrumental data. The sign test of the first difference series matches the direction of change (positive or negative) in the two series from one year to the next. The magnitude of the change is not considered. RE is the reduction of error, measure of the residual error variance relative to that which would be obtained if the reconstructed values were compared with the mean for the calibration period. Any positive value indicates that there is some sense in the reconstruction (Fritts 1976)

*p<0.05, **p<0.01, ***p<0.001

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig3_HTML.gif
Fig. 3

Comparison of actual (solid line) and reconstructed (dotted line) spring precipitation (mm). The a.d. 1953–1997 calibration model was used for reconstruction

3 Results and discussion

The reconstruction of precipitation shows annual to decadal fluctuations (Fig. 4). Over the entire period (1560–1997), the late twentieth century experienced an unprecedented precipitation increase. This twentieth century increase is also seen in Tibet (Bräuning and Mantwill 2004), although no such anomalous increase was observed in Mongolia (Pederson et al. 2001).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig4_HTML.gif
Fig. 4

Spring precipitation reconstruction (a.d. 1560–1997). Bold line denotes smoothed version obtained after fitting a cubic spline designed to remove 50% of the variance in a sine function with the wavelength of 20 years. The middle horizontal line represents mean spring precipitation (a.d. 1560–1997)

In the western Himalayan region, the minimum temperature has decreased during the latter part of the twentieth century contrary to the increase in maximum temperature (Yadav et al. 2004). We correlated instrumental spring precipitation with mean minimum spring temperature from Mukteswar and Shimla. This analysis showed inverse and highly significant relationships between spring precipitation and minimum temperature during different periods [correlation coefficient r =–0.67 (1901–1998), r=–0.72 (1901–1959), r=–0.58 (1960–1998); p (probability) in all cases is <0.0001], which indicates that night cooling (warming) is associated with increase (decrease) in spring precipitation in the western Himalayan region. Though the mean maximum temperatures increased during the twentieth century, the mean minimum temperatures decreased drastically since the 1960s. Due to diverging trends in mean maximum and minimum temperatures since 1960s, the diurnal temperature range increased. We compared instrumental spring precipitation and mean minimum temperature from Mukteswar and Shimla during 1901–1959 and 1960–1998 and found that precipitation decreased in the first subperiod and increased in the second subperiod (Fig. 5). This suggests that the precipitation increase in the latter period could be related with increasing daytime temperature, resulting in heating and evaporation, and cooling and condensation in night. These results suggest that a decrease in minimum temperatures might be a major cause for an increase in precipitation in the late twentieth century.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig5_HTML.gif
Fig. 5

a Spring minimum mean temperature anomaly relative to the 1960–1990 mean. b Spring precipitation. The dashed lines denote linear trend for 1901–1959 and 1960–1998, respectively

From 1560s to 1850s dry and wet periods shared almost equal durations, showing that the Little Ice Age (Lamb 1977; Grove 1988) experienced fluctuating spring precipitation in the western Himalayan region. This also suggests that glaciers advanced in unstable climatic conditions during this period. A 10-year running mean through the reconstructed series shows that the driest period occurred in the seventeenth century while the twentieth century experienced the wettest and also the second driest period. Running means calculated over 10, 20 and 30-year intervals clearly showed the occurrence of the wettest periods during the twentieth century (Table 4).
Table 4

The 10, 20 and 30-year running mean of reconstructed spring precipitation (mm)

Highest

Lowest

Period

mm

Period

mm

10-year mean

1982–1991

236.9

1622–1631

90.1

1646–1655

183.6

1940–1949

93.9

1731–1740

181.7

1781–1790

101.9

1573–1582

181.0

1705–1714

109.9

1956–1965

175.0

1677–1686

109.9

1818–1827

174.9

1793–1802

111.6

1666–1675

169.2

1658–1667

114.1

1602–1611

168.7

1806–1815

115.3

1751–1760

167.7

1903–1912

116.7

1727–1736

166.8

1856–1865

117.6

20-year mean

1975–1994

211.6

1934–1953

105.6

1822–1841

163.8

1621–1640

107.8

1592–1611

162.9

1781–1800

108.9

1911–1930

161.7

1856–1875

119.0

1563–1582

161.1

1705–1724

119.4

30-year mean

1968–1997

185.6

1781–1810

112.6

1816–1845

161.3

1616–1645

117.0

1646–1675

158.5

1926–1955

121.9

1732–1761

157.9

1846–1875

124.0

1569–1598

155.7

1685–1724

125.0

Dry and wet periods captured in the reconstruction closely match those found in Tibet (Bräuning and Mantwill 2004). During the second half of the eighteenth century and during the nineteenth century, wet and dry periods are also similar with those found in Mongolia (Pederson et al. 2001). The early and late eighteenth century and early nineteenth century were dry. The period from 1840s to 1910s experienced major and prolonged drought, while 1820s to 1840s experienced wet springs. During the twentieth century springs were wet except during 1930s to 1950s when dry springs occurred (Fig. 4). Other tree-ring-based precipitation reconstructions (Yadav and Park 2000; Singh and Yadav 2005) available for the western Himalayan region show similar trends to those found in this study. However, precipitation reconstructions available for Kashmir region (Hughes 1992; Borgaonkar et al. 1994) are not similar to our reconstructed series, which could be due to differences in the seasonal precipitation that was reconstructed as well as to the different climatic regimes that prevail in the two regions.

We performed spectral analysis to evaluate periodicities in the behaviour of the reconstructed precipitation time series. Coherency between the actual and reconstructed data was tested for the common period 1898–1997, which showed significant high-frequency peaks at around 2.0, 2.04, 2.09 years in both data sets (Fig. 6). Spectral analysis conducted over the entire reconstructed period (1560–1997) indicated low-frequency peaks at 33.5–39.6, 62.3, 109.0, 145.3 and 218.0 years, whereas high-frequency peaks occurred around 2.0–2.3 years (p<0.05) (Fig. 7). Such high-frequency variations as reported here were also found earlier (Diaz and Pulwarty 1994; Briffa et al. 2001; Diaz et al. 2001; Pederson et al. 2001). However, the variability mode around 36 years has been related to large-scale synoptic-scale pressure anomalies over eastern Asia and adjacent equatorial regions (Bradley 1999). It has been suggested that the biennial oscillations constituting an integral part of the Asia-Pacific climate system are caused by the interaction of atmosphere ocean-monsoon systems and extratropics (Meehl 1997). Existence of a strong correlation between reconstructed precipitation and August–December Southern Oscillation Index (SOI) (r=–0.23; p<0.008, 1866–1997) supports a linkage between the two variables. To analyse the association between spring precipitation and Southern Oscillation cross-correlation analysis was performed over the sliding window of 31 years. This showed negative correlations during 1866–1910 and 1940–1997 and positive during the intervening period. Similar shifts in relationship between all India summer monsoon rainfall and different circulation features like the sea surface temperature over the Pacific Ocean (Angell 1981; Mooley and Parthasarathy 1984), snow cover over the Himalayan region (Thapliyal 1987), Southern Oscillation (Elliott and Angell 1988) and pressure over Bombay (Parthasarathy et al. 1991) have also been noted. Such shifts in relationship have been related to large-scale changes in the tropical circulation features from meridional around 1870–1900 to zonal around 1900–1940, and again to a meridional monsoon system from 1940 onwards.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig6_HTML.gif
Fig. 6

Comparison of actual (solid line) and reconstructed (dotted line) power spectrum of spring precipitation for the common period (a.d. 1898–1997). The horizontal lines are 95% confidence levels

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-005-0089-1/MediaObjects/382_2005_89_Fig7_HTML.gif
Fig. 7

Power spectrum of the spring precipitation reconstruction (a.d. 1560–1997). The horizontal line is the 95% confidence level

Winter and spring weather conditions over the Himalayan region have significant influence on south Asian summer monsoon rainfall (Blanford 1884; Fasullo 2004; Dash et al. 2005) which in turn is also associated with El Niño-Southern Oscillation (Dey and Bhanukumar 1983; Khandekar 1991; Vernekar et al. 1995; Li and Yanai 1996). The existence of relationships between spring precipitation over the Himalayan region with tropical circulation features indicates that long-term reconstructions are very useful in understanding the climate system over the region in a long-term perspective. However, the tree-ring network needs to be strengthened to obtain more robust reconstructions.

4 Conclusions

A network of 13 tree-ring chronologies from Himalayan cedar collected from the western Himalayan region showed a direct relationship with spring precipitation and was used to reconstruct spring precipitation back to a.d. 1560. The reconstruction captured 40% of the variance in the calibration model over the period 1953–1997. This is currently the longest precipitation reconstruction covering the major part of the Little Ice Age from the western Himalayan region. In terms of the variance explained in the calibration model, the present reconstruction is also the strongest except for the April–September precipitation reconstruction developed using the ring-width and density data network from the Kashmir valley (Hughes 1992). The most interesting feature of this reconstruction is an unprecedented precipitation increase during the late twentieth century. The western Himalayan region experienced unstable springs during Little Ice Age. Running means for 10, 20 and 30-year intervals indicated the wettest period during the twentieth century, while the 10-year mean showed that the driest period occurred in the seventeenth century and the second driest period in the twentieth century.

Spectral analysis of the reconstructed series indicated high-frequency cycles around 2.0–2.3 years, an important periodicity of the Asia-Pacific climate system. The occurrence of secular shifts in the relationship between spring precipitation over the Himalayan region and SOI shows that long-term reconstructions from the Himalayan region would be useful to understand tropical circulation features in a longer-term perspective. A dense network of tree-ring data similar to those from Karakoram (Esper 2000) and Nepal (Cook et al. 2003) would also be very useful

Acknowledgements

We are thankful to forest officials of the forest department of Uttaranchal for their help during the various field trips. Critical suggestions of Dr. Henri D. Grissino-Mayer, Department of Geography, The University of Tennessee, Knoxville, TN, USA and two anonymous reviewers greatly improved the manuscript, for which we are grateful. India Meteorological Department, Pune, provided meteorological data. Present work was carried out at Chungbuk National University, Republic of Korea, under ISEF 2004–2005 sponsored by The Korea Foundation for Advanced Studies, Republic of Korea, granted to senior author.

Copyright information

© Springer-Verlag 2005