Downloaded 09/29/21 05:09 AM UTC

288 JOURNAL OF CLIMATE VOLUME 19

Extension of Drought Records for Central Asia Using Tree Rings: West-Central Mongolia*

N. K. DAVI Tree-Ring Laboratory, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

G. C. JACOBY Tree-Ring Laboratory, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York, and Department of Forestry, National University of Mongolia, Ulaanbaatar, Mongolia

A. E. CURTIS Tree-Ring Laboratory, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

N. BAATARBILEG Tree-Ring Laboratory, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York, and Department of Forestry, National University of Mongolia, Ulaanbaatar, Mongolia

(Manuscript received 22 November 2004, in final form 3 August 2005)

ABSTRACT Central Asian drought has had drastic impacts on vast regions over recent years. Longer records and insight into temporal drought patterns could aid greatly in anticipating extreme events and agrarian plan- ning. Mongolia is representative of the central Asian region, and tree-ring resources are used herein to extend the climate record and test for solar influence and/or Pacific Ocean teleconnections. Absolutely dated tree-ring-width chronologies from five sampling sites in west-central Mongolia were used in precipitation models and an individual model was made using the longest of the five tree-ring records (1340–2002). The tree-ring sites are in or near the Selenge River basin, the largest river in Mongolia and a major input into Lake Baikal in Siberia. Regression models resulted in a reconstruction of streamflow that extends from 1637 to 1997 and explains 49% of the flow variation. Spectral analysis indicated significant variation in the frequencies common to Pacific Ocean variations [Pacific decadal oscillation (PDO) and ENSO] and also some quasi-solar and lunar-nodal periodicities similar to previous Mongolian hydrometeorological reconstructions in eastern Mongolia based on tree rings.

1. Introduction edge and historical records provide partial estimation of drought recurrence (e.g., Mijiddorj and Namhay Drought across central Asia has devastated many re- 1993) but better quantitative information is needed to gions over the last few years and many times in the past. define long-term drought variations in means, ex- Much of Mongolia was especially hard hit with 4 yr of tremes, and trends, and to test for evidence of cyclical extensive drought in 1999–2002. The paucity of long- variations. Human observations leading to hypotheses term records interferes with efforts to anticipate prob- of drought variations need to be supported by quanti- able occurrence and extent of future droughts in central tative reconstructions and analyses of temporal and Asia and in particular Mongolia. Traditional knowl- spatial patterns that may hopefully lead to understand- ing of probable future events. * Lamont-Doherty Earth Observatory Contribution Number Mongolia typifies the steppe terrain and semiarid to 6825. arid continental climate that extends across much of central Asia (Zhang and Lin 1992). Mongolia also typifies large areas of Asia in another way, in that there are Corresponding author address: Nicole Davi, Tree-Ring Labo- ratory, Lamont-Doherty Earth Observatory, Columbia Univer- few long records of hydrometeorological data. Most sity, 61 Route 9W, Palisades, NY 10964. records barely extend beyond 50 yr. Our previous study E-mail: [email protected] demonstrated that tree-ring analysis can be used to re-

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC

JCLI3621 15 JANUARY 2006 DAVI ET AL. 289

FIG. 1. Map of Mongolia showing tree-ring sampling sites (triangles), precipitation stations (numbered), and streamflow gauge station (star). construct precipitation and streamflow in eastern Mon- were tested against individual precipitation stations and golia (Pederson et al. 2001) and that there was evidence regional averages of precipitation for the central region supporting quasi-solar and lunar-nodal variations in of Mongolia (Fig. 1). Significant correlations were drought for that region. The streamflow of the Kherlen found and models developed to estimate seasonal pre- River and precipitation for part of the region were re- cipitation using the tree-ring data. We also developed constructed back to 1651. Spectral analysis showed an independent streamflow reconstruction for the Se- quasi-solar and lunar-nodal variability, and the extreme lenge River. The resulting reconstruction, based on five drought of 1999–2002 fit the long-term pattern caused tree-ring chronologies, forms a 360-yr (1638–1997) rec- by the combination of solar (22 and 11 yr) and lunar- ord for streamflow variations of the Selenge River, a nodal (18.6 yr) variations (Pederson et al. 2001). Test- major source of freshwater for Mongolia, as well as the ing of the spatial consistency of such patterns is needed major input to Lake Baikal (Ma et al. 2003). to evaluate their importance. The precipitation models and streamflow reconstruction presented here are based on annual ring-width 2. Tree-ring data variations of moisture-stressed, old-aged trees. We a. Site information sampled nondestructively at several lower-elevation sites where precipitation appears to be the major factor The Zuun Salaa Mod (ZSM) site (Fig. 1, Table 1) was limiting growth. We also sampled at one slightly higher named for a “hundred-branched tree” of old-aged ap- elevation site where the highly permeable rock sub- pearance just to the east of the site. It is greatly revered strate produces an edaphic desert. The tree-ring data by people and draped with cloth and other items placed

TABLE 1. Tree-ring site information table.

No. of Tree-ring sites Lat (N) Lon (E) Alt (m) trees Years Species EPSa RBARb Undur Ulaan (UU) 48°59Ј 103°14Ј 1400 15 1473–2002 Larix sibirica Ͼ0.96 0.50–0.77 Suulchyin Medee (SM) 49°29Ј 100°50Ј 1800c 15 1573–2002 Larix sibirica Ͼ0.97 0.68–0.84 Khorgo Lava (KL) 48°10Ј 99°52Ј 2060 39 1340–2000 Larix sibirica Ͼ0.92d 0.54–0.78 Telmen Hövöö (TH) 48°46Ј 97°07Ј 1841 16 1638–1998 Larix sibirica Ͼ0.93 0.41–0.69 Zuun Salaa Mod (ZSM) 48°09Ј 100°17Ј 1900 20 1513–2001 Larix sibirica Ͼ0.89 0.45–0.85 a Expressed population signal statistic. b Rbar ϭ the mean correlation coefficient between all tree-ring series used in a chronology. c Estimated from map. d KL Rbar is Ͼ0.92 except for pre-1400s where EPS is 0.81.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 290 JOURNAL OF CLIMATE VOLUME 19 there for religious purposes. We did not sample this TABLE 2. Five-chronology correlation matrix over common tree or any nearby trees. There is a major road about period 1638–1990. 100 m to the east of the site but little evidence of other UU SM KL TH ZSM disturbance. ZSM is gently sloped (ϳ5°) to the east and UU 1 0.56* 0.27* 0.31* 0.37* sparse grass grows between each tree. There was sub- SM 1 0.42* 0.43* 0.50* stantial heartrot in some of these trees, which fre- KL 1 0.45* 0.62* quently occurs in larch trees throughout Asia, even in TH 1 0.46* dry sites. ZSM 1 Telmen Hövöö (TH, or Telmen Beach; Fig. 1, Table * Significance level (0.05). 1) was the only stand of trees for several tens of kilo- meters in this area. This small forest grows on part of a sand dune complex located next to Telmen Lake. Local dardization is to reduce the effects of age and other residents indicated that there had been some logging nonclimatic factors on the resulting series. The stan- (we saw very few stumps), no fires, and no dramatic dard chronologies were used because we wanted to pre- change in lake levels over the past 60 yr. serve low-frequency variations that would be removed The Khorgo Lava (KL) site (Fig. 1, Table 1), located by prewhitening. All the stands were open canopy and in the Khorgo-Terkhiin Tsagaan Nuur National Park, is the standard chronology is more appropriate than the on a geologically young basaltic lava flow with very ARS chronology for such sites (Cook 1985). little soil development. There has been previous success We tested for autoregressive properties of the tree- in reconstructing drought using trees from such sites ring and meteorological data for the common period (Grissino-Meyer et al. 1997). The extreme dryness of (1945–98). The meteorological data are autoregressive the site likely reduces the prevalence of heartrot in order one based on the Aikaike criteria and three of the these trees, resulting in great longevity (1340–2002). Al- five chronologies are order one. The other two are or- though near a small town, the rugged and rocky terrain der zero. Thus, the data are compatible. We did not appeared to reduce the cutting and disturbance of the prewhiten in making the model because we wanted to site. preserve low-frequency variation in the reconstruction. Undur Ulaan (UU, or Red Hill; Fig. 1, Table 1) is on Individual chronology information and statistics are a steep hill, about 200 m north of a main road. Hillside shown in Table 1. The expressed population signal grazing was the only evidence of disturbance. Trees had (EPS) statistic, a measure of chronology reliability substantial heartrot. (Wigley et al. 1984), is greater than 0.89 over the entire Suulchyin Medee (SM, or Last Statement; Fig. 1, length of each chronology, with the exception of the Table 1) is on the north and west facing slopes of a dry early portion of KL (Table 1). A level of 0.85 is con- ridge with very sparse vegetation. A lot of dieback, sidered to be an indication of satisfactory quality for a attributed to insect infestation, was seen in the canopy; chronology. The RBAR statistics, the mean correlation however, there was no evidence of decreased growth coefficient for all tree-ring series in each individual due to infestation in the growth rings. This indicates chronology, are shown in Table 1. Average series in- that the severe infestation is very recent. Presently, in- tercorrelation for the five chronologies is 0.51 over the sect infestations are very common in Mongolian forests 1638–1990 common period. Correlations for each indi- and the stand of trees may not survive many more vidual chronology pair are shown in Table 2. years. We tried to sample healthy trees to exclude ef- The five tree-ring chronologies were loaded similarly fects of insect infestation. in a principal component analysis (PCA) (Table 3). The b. Methods TABLE 3. Ring-width loadings on the first eigenvector from Standard tree-ring methods were used for chronol- PCA for the five chronologies. Current year (T) and current year with prior-year (T Ϫ 1) correlations with recorded Selenge ogy development and climate analysis (Fritts 1976; streamflow data. Cook and Kairiukstis 1990). Ring widths were measured to the nearest 0.001 mm using a Velmex system. Site Eigenloading TTϪ 1 Individual series measurements and dating were UU 0.394 0.435* 0.124 checked using COFECHA software (Holmes 1983) and SM 0.473 0.358* 0.264* standardized using ARSTAN software (Cook 1985). KL 0.452 0.501* Ϫ0.011 Conservative detrending methods (negative exponen- TH 0.425 0.578* 0.359* ZSM 0.485 0.539* 0.300* tial and straight-line curve fits) were used to generate the ring-width index chronologies. The purpose of stan- * Significance level (0.05).

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 15 JANUARY 2006 DAVI ET AL. 291

TABLE 4. Hydrometeorological data. sum of annual precipitation averaged for the three stations is 252.3 mm. The growing season in Mongolia, at Precipitation station Lat (N) Lon (E) Alt (m) Years our lower-elevation sites, is roughly June to late Au- Uliastai 47°45Ј 96°51Ј 1751 1937–2003 gust. Average monthly temperatures from Mörön, Uli- Khujirt 46°54Ј 102°46Ј 1650 1943–96 astai, and Khujirt are above freezing starting in May Mörön49°38Ј 100°10Ј 1283 1941–2003 Mongolian station 48°12Ј 99°54Ј 2010* 1971–83 and drop below freezing starting in October. Average N. Han annual temperature for west-central Mongolia based on Ϫ Hydrological station Lat (N) Lon (E) Alt (m) Years the three stations is 1.9°C the 1943–96 common pe- Selenge 49°14Ј 101°21Ј 1945–2002 riod.

* Estimated. b. Hydrological data Streamflow data for the Selenge River at the Hutag chronologies were combined using PCA to reduce the streamflow measuring station (1945–2002) were ob- number of predictors and create one series representing tained from the Hydrometeorological Institute of Mon- common regional variations. PCA was calculated using golia, and were updated by N. Baatarbileg (2004, un- a correlation matrix because the data are standardized. published manuscript). Most of the runoff comes from The first eigenvector (EV1) contains 55% of the total precipitation in the mountains surrounding Lake Hövs- variance. göl and the mountains to the southwest. Most streamflow discharge occurs over a longer season (April– 3. Hydrometeorological data October) than precipitation, with the highest discharge a. Precipitation in July and August (Fig. 2). Spring streamflow comes from melting winter snowpack and much of the late Monthly instrumental precipitation and temperature season streamflow comes from infiltration and bank records from Mörön (1941–2003), Uliastai (1937–2003), storage after the direct precipitation runoff decreases. and Khujirt (1943–96) (MUK) were obtained from the The June–August precipitation averaged over the three Global Historic Climate Network (information avail- stations used for modeling (Mörön, Uliastai, and able online at www.ncdc.noaa.gov/oa/climate/research/ Khujirt) correlates with April–October streamflow (r ϭ ghcn/ghcngrid.html) (Table 4). These stations are the 0.68, p ϭ 0.01) data over the common period 1945–96. closest to the tree-ring sites with continuous records. There are other stations in the region but they either have much shorter records or have many missing val- 4. Analysis and results ues. There is also a precipitation station near the a. Precipitation Khorgo Lava site with a short (1971–83) record called Mongolian station N. Han (station ID 2184423701). The five individual tree-ring chronologies were Although there are only 13 yr of record, it is uncommon tested with three local meteorological stations: Mörön, to find a meteorological station adjacent to a sampling Uliastai, and Khujirt (MUK). We tried to include more site (see Tables 1 and 5). This record is too short to stations but using records with many missing values es- include in the regional precipitation series but we use it timated or shorter records detracted from the results. to further test the precipitation signal in the long Correlations with monthly precipitation for individual Khorgo Lava chronology. Average mean monthly pre- MUK stations varied from site to site. The most con- cipitations for the three stations (Mörön, Uliastai, and sistent positive correlations were with the current sea- Khujirt) shows that most (72%) of the annual precipi- son June. Correlations improved when using the EV1 tation falls during June, July, and August (Fig. 2). The from the five chronologies with average monthly pre-

TABLE 5. Calibration and verification statistics for Selenge River streamflow reconstruction. Adjusted r2 is the variance accounted for by the calibration model, adjusted for degrees of freedom; RE is the reduction of the error statistic; and P is the Pearson correlation coefficient. Levels of significance are indicated in parentheses.

1945–70 calibration 1971–97 verification 1971–97 calibration 1945–70 verification Adjusted r2 0.356 — 0.543 —– RE 0.382 0.411 0.561 0.282 Sign test 20ϩ/6Ϫ (0.005) 22ϩ/5Ϫ (0.001) 22ϩ/5Ϫ (0.001) 19ϩ/7Ϫ (0.01) P 0.618 (0.001) 0.685 (0.001) 0.749 (0.001) 0.536 (0.002)

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 292 JOURNAL OF CLIMATE VOLUME 19

FIG. 2. (left) Average monthly precipitation (mm) for Mörön, Uliastai, and Khujirt EV1 over the 1943–96 time interval and (right) average monthly streamflow (m3 sϪ1) for Selenge-Hutag over the 1945–2002 time interval. cipitation from MUK. The EV1 shows positive signifi- local precipitation record (1971–83) at Mongolian sta- cant correlation with prior September, and current sea- tion N. Han showed strong significant correlations be- son February–March and June–July (Fig. 3). February– tween the tree-ring indices and April and July precipi- March correlations do not have any ready explanations. tation: 0.57 (p ϭ 0.05) and 0.80 (p ϭ 0.01), respectively. Liang et al. (2001) also found significant correlation The ring-width indices from the standard chronology between Meyer spruce growth and February precipita- and the April–July total precipitation have a simple r of tion in Inner Mongolia. We found that there are no 0.69 (p ϭ 0.01) and the r-squared adjusted for degrees significant intercorrelations of precipitation with tem- of freedom lost due to the regression is 0.43. Although perature for these months. The ring-width EV1 corre- the N (13) of the recorded precipitation data is low, the lates highest with an average of June, July, and August correlations are significant and confirm the precipita- precipitation from MUK (r ϭ 0.52) (p ϭ 0.01) over the tion signal in the Khorgo Lava chronology shown in 1943–95 common period (Fig. 4). Fig. 5. Regression between the Khorgo Lava trees and the Temperature station data were also tested using the same three local meteorological stations (Muren, Uli- astai, and Khujirt). These temperature data show a consistent negative relationship with the five chronologies for previous July and current season April and June. This relationship strengthened when data from the three temperature stations were averaged and then correlated with ring-width EV1. The consistent negative correlation for the current warm season months ranged from April (r ϭϪ0.471) to July (r ϭϪ0.239). April is a month of low precipitation, and warm springs may increase an already high rate of evapotranspiration (Ma et al. 2003). This may further lessen water availability for the growing season and explain the high negative correlation for April.

b. Streamflow FIG. 3. Correlation plot comparing average monthly precipita- Each individual chronology was tested with Selenge tion from Mörön, Uliastai, and Khujirt with the first principal component of the five ring-width chronologies over the 1943–95 streamflow data for the current year (T) and prior year common period. Horizontal lines indicate the 95% significance (T Ϫ 1) (Table 3). Three of the five chronologies show level. Stars represent significant correlations. significant correlations with prior-year streamflow. Site

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 15 JANUARY 2006 DAVI ET AL. 293

FIG. 4. Average June–August precipitation from Mörön, Uliastai, and Khujirt EV1 from PCA based on the five ring-width chronologies over the 1943–95 common period.

FIG. 5. Khorgo Lava (1340–2000) standard chronology of ring-width indices and sample size.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 294 JOURNAL OF CLIMATE VOLUME 19

season April–October and prior-year September and October (Fig. 6). Based on these correlations we use principal component regression (which recreates EV1) to develop an April–October tree-ring-based model (using EV1 for year T and year T ϩ 1 to predict streamflow for year T) for Selenge streamflow over the 1638– 1997 time period (Fig. 7). The climate–tree-growth reconstruction (Figs. 8a,b) accounts for 49.3% of the streamflow variance and demonstrates valid calibration and verification statistics (Table 5). Additional eigen- vectors were tested, but they did not improve the model. Most streamflow models using tree-ring data are lin- ear. However, in some circumstances the relationships can be nonlinear (e.g., Fritts 1976). With increasing amounts of precipitation (and resulting streamflow), FIG. 6. Monthly correlation coefficients for Selenge at the levels are reached that do not cause ring widths to in- Hutag streamflow station with five-chronology EV1 over the 1945–97 common interval. Horizontal bars indicate the 95% sig- crease at the same rate that they do for the increases at nificance level. lower levels because the trees experience other growth restraints such as nutrient supply or internal biochemi- cal limitations. Plotting the estimated and recorded val- UU (Table 1) shows a positive, but not significant, cor- ues can reveal whether “a curvilinear model is more relation with prior-year streamflow (Table 3). This may appropriate” (Fritts 1976). Previously a log model im- reflect the downstream distance from the Selenge proved results in a reconstruction of the Virgin River in streamflow gauge to the UU site (Fig. 1). Khorgo Lava Utah (Hereford et al. 1995). Nonlinear models were is a precipitation sensitive site that has a high degree of tested in Pederson et al. (2001) but showed no improve- agreement between trees, but does not have a prior- ment. (This negative result was not included in that year signal. The main differences between the Khorgo publication.) In this Selenge model, the plots suggested Lava site and the others are the low moisture retention a nonlinear relationship between the tree-ring data and of the lava substrate and the site elevation. It is the streamflow (Fig. 8a). The nonlinear model improved highest elevation of the five sampling sites (Table 1). the explained variance from 42.1% to 49.3%; therefore, The regionalized EV1 based on five chronologies the nonlinear model is used (Figs. 8a,b). shows positive significant correlations with current- Selenge streamflow variability was evaluated using

FIG. 7. Tree-ring-based reconstruction of Selenge River streamflow. The fine line is the average streamflow rate for the months April–October 1637–1997. The thicker line is a 25-yr moving average.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 15 JANUARY 2006 DAVI ET AL. 295

FIG. 8. (a) Scatterplot of reconstruction model results using the log of streamflow. (b) Estimated and recorded Selenge River streamflow rate converted back to m3 sϪ1. Note that even with a log model the tree-ring data still underestimate the extreme flow of 1993.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 296 JOURNAL OF CLIMATE VOLUME 19

TABLE 6. Wettest and driest 5-yr periods for Selenge streamflow The amount of variance explained by the quasi-solar reconstruction. and lunar-nodal frequencies and the lunar-nodal fre- Rank Dry Wet quency is estimated by using single-spectrum analysis (SSA) (Vautard and Ghil 1989). The frequencies we 1 1854–58 1764–68 include are 22 and 11 yr (solar) and 18.6 (lunar nodal). 2 1779–83 1990–94 3 1736–40 1797–1801 Summing the variance explained by these frequencies 4 1737–41 1763–67 gives an estimate of about 27%. One must regard this 5 1863–67 1796–1800 estimate with caution as the SSA is very sensitive to 6 1901–05 1798–1802 small changes in input parameters. 7 1641–45 1765–69 We also checked spectral coherency between the 8 1696–1700 1917–21 9 1855–59 1794–98 tree-ring series and SOI and PDO. The SOI results 10 1778–82 1795–99 show coherency in the 3–4- and 6–8-yr ranges between SOI and both EV1 and Khorgo Lava. There is spectral coherency between PDO and both series in the 32- the top-10 wet and dry 5-yr intervals (Table 6). The and 3.8–3.9-yr ranges (Figs. 10a,b). Between 4 and 21 wettest 5-yr period was 1764–68 and the driest period yr, there are differences between the coherencies of the was 1854–58. The most extended wet period is 1794– PDO with the two tree-ring-based series. 1802 and an extended dry period is 1778–83. From the beginning of the reconstruction through the mid-1800s, 5. Discussion the reconstruction shows a high degree of variability in Streamflow can be a better indicator of regional wa- streamflow. This trend is also seen in the Kherlen River ter resources than individual station precipitation reconstruction but ends around 1850 (Pederson et al. records. Streamflow integrates the areal effects of pre- 2001). The severe extended droughts during the 1800s cipitation input and evapotranspiration losses. This sys- (Fig. 7) are more extreme than those found in Pederson tem of moisture balance is more similar to the hydro- et al. (2001). The twentieth century looks wetter than logic environment of trees than precipitation alone. prior centuries but is within the range of long-term vari- Hence, a closer relationship between ring-width data ability (also seen in Pederson et al. 2001). The strongest and streamflow is not surprising. We did not make a regional difference is in the very late 1700s. The eastern precipitation reconstruction with the EV1 because the region was dry while this central region was very wet. variance explained was only about 25% whereas with the streamflow it was almost 50%. The reconstruction c. Spectral analysis extends beyond the season of cambial (ring) growth to We processed the streamflow reconstruction, the better represent the longer season of streamflow, and five-chronology EV1, and the chronology from the the longer season of biological activity, April–October. Khorgo Lava site through a multitaper method (MTM) The correlations between EV1 and September and Oc- spectral analysis (Mann and Lees, 1996) and a cross- tober streamflow, after the end of the season for cam- spectral analysis with climatic series, specifically the bial growth, are caused by streamflow for these months Southern Oscillation index (SOI) and Pacific decadal reflecting the precipitation input of the prior months oscillation (PDO), for the season having the strongest and river basin storage components. The earlier pre- climatic signal. We describe the EV1 and Khorgo Lava cipitation directly benefits the tree growth. The precipi- results (EV1 and the streamflow reconstruction have tation results confirm the primary influence on the the same spectral characteristics). The MTM plots tree’s growth. In addition the results for the Khorgo (Figs. 9a,b) show that both of the series have spectral Lava site alone support the value of this series for power in an ENSO range of about 4 yr (99% signifi- analyses of longer-term spectral properties. The large cance) and 5–6 yr (90% significance). There is also difference between the eastern (Pederson et al. 2001) spectral power in a quasi-solar range of about 22–24 yr and the west-central (herein) reconstructions in the late (95%) and for Khorgo Lava there is some power in the 1700s is an example of the high spatial variability that lunar-nodal range of about 16–19 yr (90%–95%). Evi- can occur in Mongolia. This type of regional difference dence of power in the range of PDO variations of 35–50 was also found by Yatagai and Yasunari (1995). yr is also present (99%). Differences between the EV1 Spectral periodicities found in Pederson et al. (2001) and Khorgo Lava results are not unexpected because are similar to some of the properties found in the west- the latter had no lag effect of the prior year’s precipi- central Selenge River reconstruction. Differences may tation and it is a substantially longer series by 238 yr. be due to the more mountainous topography to the

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 15 JANUARY 2006 DAVI ET AL. 297

FIG. 9. MTM spectral frequencies of (a) the five-chronology EV1 (1638–1997) and (b) the Khorgo Lava site (1400–2000). west as well as regional differences due to circulation Kulkarni (1999) and Morinaga et al. (2003), although shifts. The spectral properties similar to Pacific Ocean they focused on winter conditions. Possible teleconnec- spectra suggest teleconnections in agreement with pre- tions in summer precipitation variations are described vious findings. Evidence for these types of oceanic tele- in Yatagai and Yasunari (1995) based on analysis of connections have also been presented by Kripalani and recorded precipitation from 1951 to 1990.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 298 JOURNAL OF CLIMATE VOLUME 19

FIG. 10. Cross-spectral analysis of (a) the five-chronology EV1 with PDO and (b) the Khorgo Lava site and PDO, over the 1900–97 common period.

6. Conclusions variation in the long-term tree-ring record than in the limited record of measured precipitation. This new rec- This reconstruction is a reliable record of the Selenge ord provides an important extension of the input flow River streamflow variability in west-central Mongolia to Lake Baikal. In addition, it provides information that increases the knowledge of drought and wet peri- about past and likely future variations in water avail- ods. The spectral properties [in agreement with Peder- ability in central Mongolia. son et al. (2001)] are strong enough to infer some significant predictability and also suggest teleconnections Acknowledgments. We thank R. D’Arrigo, N. Peder- with Pacific Ocean variations. son, K. Peters, E. Cook, and three anonymous review- In Fig. 7 it can be seen that there is much wider ers for their comments. We also thank Oyunsanaa

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC 15 JANUARY 2006 DAVI ET AL. 299

Byambasuren for his great help in the field. This re- on a sandy substrate in semi-arid grassland, north China. search was aided by the Ministry for Nature and Envi- Trees, 15, 230–235. ronment in Mongolia and the National University of Ma, X., T. Yasunari, T. Ohata, L. Natsagdorj, G. Davaa, and D. Oyunbaatar, 2003: Hydrological regime analysis of the Se- Mongolia. This material is based on work supported by lenge River basin, Mongolia. Hydrol. Processes, 17, 2929– the National Science Foundation under Grants ATM 2945. 0117442 and OCE 0402474 Mann, M., and J. Lees, 1996: Robust estimation of background noise and signal detection in climatic time series. Climatic REFERENCES Change, 33, 409–445. Cook, E. R., 1985: A time series analysis approach to tree-ring Mijiddorj, R., and A. Namhay, 1993. Reconstruction of 2-century- standardization. Ph.D. thesis, The University of Arizona, year climate phenomena oscillations existing in Mongolia by Tucson, AZ, 171 pp. historical sources. Proc. First PRC–Mongolia Workshop on ——, and L. Kairiukstis, 1990: Methods of Dendrochronology: Climatic Change in Arid and Semi-arid Region over Central Applications in the Environmental Sciences. Kluwer, 394 pp. Asia, Beijing, China, Mongolian Hydrometeorological Ser- Fritts, H. C., 1976: Tree Rings and Climate. Academic Press, 567 vice and China Meteorological Administration, 11–16. pp. Morinaga, Y., S.-F. Tian, and M. Shinoda, 2003: Winter snow Grissino-Meyer, H., T. W. Swetman, and R. K. Adams, 1997: The anomaly and atmospheric circulation in Mongolia. Int. J. Cli- rare, old-aged conifers of El Malpais—Their role in under- matol., 23, 1627–1636. standing climatic change in the American southwest. Natural Pederson, N., G. C. Jacoby, R. D. D’Arrigo, E. R. Cook, and History of El Malpais National Monument, K. Mabery, Ed., B. M. Buckley, 2001: Hydrometeorological reconstructions New Mexico Bureau of Mines and Mineral Resources, Bul- for northeastern Mongolia derived from tree rings: 1651– letin 156, 155–161. 1995. J. Climate, 14, 873–881. Hereford, R., G. C. Jacoby, and V. A. S. McCord, 1995: Geomor- Vautard, R., and M. Ghil, 1989: Singular spectrum analysis in phic history of the Virgin River in the Zion National Park nonlinear dynamics, with applications to paleoclimatic time area, southwest Utah. U.S. Geological Survey Open File series. Physica D, 35, 395–424. Rep. 95-515, 75 pp. Wigley, T. M., K. R. Briffa, and P. D. Jones, 1984: On the average Holmes, R. L., 1983: Computer-assisted quality control in tree- value of correlated time series, with applications in dendro- ring dating and measurement. Tree-Ring Bull., 44, 69–75. climatology and hydrometeorology. J. Climate Appl. Meteor., Kripalani, R. H., and A. Kulkarni, 1999: Climatology and variabil- 23, 201–213. ity of historical Soviet depth data: Some new perspectives in Yatagai, A., and T. Yasunari, 1995: Interannual variations of sum- snow–Indian monsoon teleconnections. Climate Dyn., 12, mer precipitation in the arid/semi-arid regions in China and 475–489. Mongolia: Their regionality and relation to the Asian sum- Liang, E., X. Shao, and J. Lin, 2001: Dendroclimatic evaluation of mer monsoon. J. Meteor. Soc. Japan, 73, 909–923. climate–growth relationships of Meyer spruce (Picea meyeri) Zhang, J., and Z. Lin, 1992: Climate of China. Wiley, 376 pp.

Unauthenticated | Downloaded 09/29/21 05:09 AM UTC