Large-Scale Precipitation Variability Over Northwest China Inferred from Tree Rings

1JULY 2011 F A N G E T A L . 3457

Large-Scale Precipitation Variability over Northwest China Inferred from Tree Rings

KEYAN FANG,XIAOHUA GOU, AND FAHU CHEN Key Laboratory of Western China’s Environmental Systems (Ministry of Education), Lanzhou University, Lanzhou, China

EDWARD COOK,JINBAO LI,BRENDAN BUCKLEY, AND ROSANNE D’ARRIGO Tree-Ring Lab, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

(Manuscript received 23 June 2010, in ﬁnal form 18 January 2011)

ABSTRACT

A preliminary study of a point-by-point spatial precipitation reconstruction for northwestern (NW) China is explored, based on a tree-ring network of 132 chronologies. Precipitation variations during the past ;200– 400 yr (the common reconstruction period is from 1802 to 1990) are reconstructed for 26 stations in NW China from a nationwide 160-station dataset. The authors introduce a ‘‘search spatial correlation contour’’ method to locate candidate tree-ring predictors for the reconstruction data of a given climate station. Calibration and veriﬁcation results indicate that most precipitation reconstruction models are acceptable, except for a few reconstructions (stations Hetian, Hami, Jiuquan, and Wuwei) with degraded quality. Additionally, the authors compare four spatial precipitation factors in the instrumental records and reconstructions derived from a rotated principal component analysis (RPCA). The northern and southern Xinjiang factors from the instrumental and reconstructed data agree well with each other. However, differences in spatial patterns between the instrumentation and reconstruction data are also found for the other two factors, which probably result from the relatively poor quality of a few stations. Major drought events documented in previous studies—for example, from the 1920s through the 1930s for the eastern part of NW China—are reconstructed in this study.

1. Introduction areas (China Daily, 17 February 2009), which regret- tably extends to the 1950s for most stations. The brevity At present, water resources are crucial in dry north- of the instrumental record can limit our ability to eval- western (NW) China for both the civic and communal uate current drought severity relative to long-term vari- water supplies, as well as for agriculture activities. The ability inherent in climate systems. We therefore draw severe drought in more than eight provinces in the upon the paleoclimate records available to us to place semihumid-to-arid region of northern China in early the current drought regimes into a longer context of the 2009 is one such case. During that period, 4.6 million natural variability of climate in our study region in NW people were short of drinking water and nearly 30 mil- China (the area north of 358N and west of 1108E). Several lion people suffered from water shortages for agricul- tree-ring based climate reconstructions have been pub- ture activities, forcing the central government to enact lished for a geographic point or small areas of dozens its top-level emergency drought response for the ﬁrst of square kilometers in this region (e.g., Fang et al. 2009, time. This 2009 drought ranks as the most extreme dur- 2010b; Gou et al. 2008; Li et al. 2006; Liu et al. 2004; Shao ing the period of instrumental monitoring for many et al. 2005; Sheppard et al. 2004; Yuan et al. 2003; Zhang et al. 2003). In this paper, we aim to establish the drought footprint for the entire region and to analyze its broad Corresponding author address: Keyan Fang, Key Laboratory of spatiotemporal features. Western China’s Environmental Systems (Ministry of Education), Lanzhou University, 222 South Tianshui Road, Lanzhou 73000, The wide distribution and annual resolution of tree China. rings make it possible to use quantitative methods to E-mail: [email protected] not only reconstruct local climate variability but also

DOI: 10.1175/2011JCLI3911.1

Ó 2011 American Meteorological Society Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3458 JOURNAL OF CLIMATE VOLUME 24 reconstruct large-scale spatial climate patterns (Cook difference is that we used a modified PPR method for et al. 1999). Previous dendroclimatic studies have used reconstruction based on a ‘‘search spatial correlation a tree-ring network to reconstruct the two leading drought contours’’ (see section 2d for details). fields of central High Asia (Fang et al. 2010a), including Since regional precipitation variability for much of large parts of NW China. This research first delimited the NW China is directly related to monsoon dynamics study region into the western and eastern portions be- (Fang et al. 2010a; Li et al. 2009; Shi et al. 2007), a re- cause of their distinct drought variability and recon- constructed PPR precipitation dataset can be used to structed the first principle component of drought indices investigate long-term precipitation history and its re- for both portions. The field reconstruction, however, has lations with the Asian monsoon. Spatial precipitation a limited ability to investigate drought history over space reconstruction would be more challenging relative to because it contains a priori spatial patterns of instrumental a PDSI reconstruction, because tree rings in this area are data in its design. For example, temporal variations in the often highly correlated with PDSI (Fang et al. 2009, spatial boundaries of the specified drought modes cannot 2010b; Li et al. 2006). We describe the tree-ring network, be identified with this method. A point-by-point regres- climate data, and regression and other analytical methods sion (PPR) spatial climate reconstruction method can in section 2. Special attention is paid to the regression ameliorate this situation by reconstructing each climate methods because they differ slightly from previous stud- point independently, thereby allowing for ‘‘local control’’ ies. For each station, the regression models are validated over the reconstruction at each point (Cook et al. 1999). by independent data to test for temporal stability. The One of the most notable examples of these methods in the regression models are further examined according to recent literature is the development of the North America their ability to model the temporal variability and recover Drought Atlas (NADA), the methodology of which was the spatial patterns. These results and a discussion are first described by Cook et al. (1999) and later fully pub- presented in section 3, and section 4 summarizes the re- lished in Cook et al. (2004). Spatial reconstruction of all sults of this study. climate points in a region also facilitates an examination of underlying climate forcing via a comparison of the reconstruction with results from various general circulation 2. Data and methods models (GCMs). Several subsequent papers utilized this a. Study region set of gridded reconstructions for detailed analyses of the causes of drought over a large portion of North America Located in east-central Asia (Fig. 1), NW China fea- (e.g., Seager 2007). tures a dry continental climate with large temperature The PPR-based spatial reconstructions have been con- gradients between winter and summer. According to the ducted for many areas, such as continental North America precipitation records used herein, it is relatively humid 21 (Cook et al. 1999, 2004; Meko et al. 1993). Recently, Cook (more than ;200 mm yr ) for the northwestern and et al. (2010) published the first ensemble-based PPR re- southeastern portions of NW China, while it is very dry 21 construction for East Asia by calibrating 327 tree-ring (less than 60 mm yr ) in between, such as in southern chronologies with 534 grid points of the Palmer drought Xinjiang, western Gansu, and western Neimenggu (Fig. 2). severity index (PDSI) for monsoonal Asia. We introduce Peak rainfall occurs in June for western NW China herein a precipitation reconstruction using modified PPR (Xinjiang), where the May–July seasonal precipitation methods. Our study region in the arid northwest has accounts for 36% of the annual precipitation. For the perhaps the densest tree-ring network in China, because eastern region (areas other than Xinjiang), peak rainfall of the availability of old-growth and climate-sensitive occurs in August, and July–September precipitation forests, though with far less spatial coverage than the accounts for 64% of the yearly total precipitation. Pre- previously noted studies of Cook et al. (1999, 2004) and vious studies documented differences in climate vari- Meko et al. (1993) over North America. The target data ability between western and eastern NW China (Chen for reconstruction are derived from a nationwide 160- et al. 2008; Li et al. 2009; Qian and Qin 2008; Shi et al. station precipitation dataset (see climate data section) 2007; Zou et al. 2005). For example, in recent decades an that contains 26 stations within NW China. We also se- increase in precipitation has been observed for western lected this climate dataset as predictands because it is a NW China, while the opposite is true for the eastern basic and frequently used dataset for China (e.g., Gemmer portion (Shi et al. 2007). et al. 2004; Lau and Weng 2001). One difference from b. Tree-ring data the study by Cook et al. (2010) is that we aim to reconstruct a 26-station precipitation dataset for NW China Tree-ring samples were collected from the old-growth from a tree-ring network of 132 chronologies. Another forests in scattered mountainous ranges distributed

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 1JULY 2011 F A N G E T A L . 3459

FIG. 1. Locations of meteorological stations (circles) in NW China and surrounding tree-ring chronologies (triangles). The search spatial correlation coefficients are also illustrated for the Wulumuqi and Xining stations, which were calculated using precipitation of the station tar- geted for reconstruction and nearby precipitation grids. Tree-ring chronologies located within the area of a specified correlation contour are included as candidate chronologies for reconstruction purpose. within the treeless, dry lowlands. The patchiness of high tree-ring indices were averaged to generate chronologies mountains and old-growth forests leads to the uneven based on robust mean methodology (Cook 1985). In this distribution of tree-ring sites, as shown in Fig. 1. As lis- study, we use the standard and autoregressive standard- ted in Table 1, our tree-ring network in NW China and ization (ARSTAN) chronologies but not the residual vicinity are derived from previously published tree-ring chronologies, for which low-frequency variations are chronologies (e.g., Fang et al. 2009, 2010b; Li et al. 2006; largely eliminated (Cook 1985). The following analyses Liu et al. 2004; Shao et al. 2005; Sheppard et al. 2004; are only based on the reliable portion of tree-ring Yuan et al. 2003; Zhang et al. 2003), the International Tree Ring Data Bank (ITRDB; http://www.ngdc.noaa. gov/paleo/ftp-treering.html), our contributed tree-ring chronologies, and the Chinese Tree Ring Data Center (CTRDC; http://ctrdb.ibcas.ac.cn/index.asp). All of the samples were taken from coniferous species (Pinaceae and Cupressaceae,asshowninTable1). Most raw tree-ring measurements were standardized to remove age-related biological trends by fitting straight lines or negative exponential curves. These fitted ‘‘conservative’’ curves preserve low-frequency climate sig- nals that can often be removed by more data-adaptive methods (Briffa and Jones 1990). We used a rigid cubic- spline curve, with a 50% cutoff equal to two-thirds of each series’ length, to treat some tree-ring series with strong disturbances that were not well fitted by more conservative curves. Tree-ring indices were calculated FIG. 2. Mean annual precipitation (mm) map for NW China based as ratios between the actual and fitted values. Detrended on the 26 meteorological stations in this area.

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3460 JOURNAL OF CLIMATE VOLUME 24

TABLE 1. Descriptions of the tree-ring network in NW China and vicinity.

No. of Full Mean chronology Reliable mean Region chronologies coverage length (yr) length* (yr) Species** Source Altai (Russia) 48 1581–1994 350 327 LASI ITRDB NW China 51 0–2004 594 486 PISC, LASI, PIAR, Contributed JUPR, etc. ITRDB CTRDC Mongolia 13 832–2002 563 477 LASI, PISI, PISY ITRDB Kirgizstan 20 1604–1995 308 272 PISY ITRDB

* The reliable length of individual chronologies was determined when more than six tree-ring series were found. ** LASI 5 Larix sibirica; PISC 5 Picea schrenkiana; LASI 5 Larix sibirica; PIAR 5 Pinus armandii; JUPR 5 Juniperus przewalski; PISI 5 Pinus sibirica; PISY 5 Pinus sylvestris; and PISC 5 Picea schrenkiana. chronologies with more than six individual series (Meko d. Regression methods et al. 1993). A tree-ring network with 132 chronologies was developed, and the mean length of the reliable por- We use the PPR method (Cook et al. 1999) to re- tions of all chronologies is 391 yr (Table 1). Most tree- construct the precipitation data of each station from the ring chronologies were developed from total ring-width surrounding tree-ring chronologies. PPR is essentially data, and a few of them were established from early-wood a variant of the principal component regression (PCR), width, late-wood width, early-wood density, minimum which is used to reconstruct multiple predictands from density, late-wood density, or maximum density. multiple predictors (Briffa et al. 1986), restricted to the one-predictand case (Cook et al. 1999). Previous PPR- c. Climate data based reconstructions chose a ‘‘search radius’’ method The target climate records for reconstruction are the to locate candidate tree-ring chronologies (e.g., Cook monthly precipitation data from 26 stations (Table 2), et al. 1999, 2004), whereby candidate tree-ring predictors derived from a nationwide 160-station dataset beginning are those that fall within the ‘‘circle’’ of a given search in the 1950s [courtesy of the National Climate Center radius centered at the tree-ring site. If insufficient tree- (NCC) in Beijing; http://ncc.cma.gov.cn/cn/]. Since the ring chronologies are found within the spatial area of a data for the period 1991–2005 are missing for the Shanba search radius, then the search radius is extended to a specific station, we only include data from 1954 to 1990 for distance to gather more tree-ring data for regression. that station (Table 2). The regression models based on One innovation of this study is the use of search spa- the data since the 1950s are verified with independent tial correlation contours to locate candidate tree-ring data of monthly precipitation prior to the 1950s from chronologies for the calibration of a given climate sta- the Global Historical Climatology Network (GHCN) tion, instead of the search radius. Spatial correlation (www1.ncdc.noaa.gov/pub/data/ghcn/v2). Our selection contours are calculated between precipitation data of of these degraded pre-1950s data (Li et al. 2009) for the target climate station for reconstruction and the verification was made to take advantage of the minimal surrounding gridded precipitation records from the post-1950s data for calibration. Missing values were re- CRU TS3 dataset. This spatial correlation map indicates placed by mean values of neighboring years. For those the similarity of precipitation variability between the stations without historical data prior to the 1950s, we use location of the target climate station and surrounding the interpolated data of the nearest precipitation grid areas. Tree rings located in areas with comparable pre- from the Climatic Research Unit (CRU) TS3 dataset at cipitation to the target climate station tend to show 0.5830.58 resolution (New et al. 2000) for verification higher correlations with the precipitation records of that purposes. Because the pre-1950s data could be in- station. Tree-ring chronologies falling within the spatial terpolated from distant stations if there were no con- ranges of the initial spatial correlation coefficient con- current data nearby, we only selected the period when tour (0.4) are selected as candidate predictors. Similar to station historical records could be found within a rea- the search radius method (Cook et al. 1999), we expand sonable distance (;500 km) or used an arbitrary 15-yr the areas surrounding the climate station if insufficient (the average length of historical data before 1951 is candidate tree-ring chronologies are found. In this study, 14.3 yr) verification period (Table 2). For example, be- the search spatial correlation contours expand in 0.05 cause data collection at the Lanzhou station began in increments until at least 20 candidate tree-ring chronol- 1933, the gridded data for verification used near the ogies are available (Fig. 1). The search radius is based on Lanzhou station were truncated at 1933. the hypothesis that the climate is evenly distributed in

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 1JULY 2011 F A N G E T A L . 3461

TABLE 2. Descriptions of the locations and statistics of reconstruction models of the 26 stations from the (top) western to the (bottom) eastern part of NW China. Note that the instrumental station data prior to the 1950s are extracted from the GHCN dataset. Interpolated data from the nearest grid of the CRU dataset prior to the 1950s were used where the GHCN data are not available. Calibration periods are the common period between tree-ring chronologies and the most recent instrumental data after the 1950s, and the veriﬁcation periods are based on the instrumental data from the GHCN or CRU datasets before the 1950s.

Station Lat Lon Instrumental Screening No. of retained Calibration Explained name (8N) (8E) time span probability chronologies Prewhitening? period variance R2 Aletai 47.73 88.08 1939–2008 0.20 24 Yes 1954–92 0.344 Hami 42.82 93.52 1941–2008 0.20 11 Yes 1951–94 0.305 Hetian 37.13 79.93 1941–2008 0.05 3 No 1953–94 0.309 Kashi 39.47 75.98 1936–2008 0.25 13 Yes 1951–94 0.297 Kuche 41.72 82.95 1941–2008 0.30 13 Yes 1951–92 0.479 Qiemo 38.15 85.55 1941–2008 0.10 7 No 1952–94 0.400 Ruoqiang 39.03 88.17 1941–2008 0.30 13 No 1953–92 0.470 Tacheng 46.73 83.00 1941–2008 0.15 23 No 1954–93 0.430 Tulufan 42.93 89.20 1941–2008 0.30 14 Yes 1952–92 0.313 Wulumuqi 43.78 87.62 1941–2008 0.20 15 No 1951–92 0.462 Wusu 44.43 84.66 1939–2008 0.20 13 Yes 1954–92 0.295 Yining 43.95 81.33 1941–2008 0.30 12 Yes 1951–92 0.374 Dunhuang 40.15 94.68 1938–2008 0.20 17 Yes 1951–92 0.311 Jiuquan 39.77 98.48 1935–2008 0.20 10 Yes 1951–93 0.293 Lanzhou 36.05 103.88 1933–2003 0.15 11 Yes 1951–87 0.557 Linxia 35.58 103.18 1939–2008 0.20 9 Yes 1951–87 0.323 Wuwei 38.05 102.55 1933–2008 0.30 15 Yes 1951–92 0.283 Xifengzhen 35.73 107.63 1937–2008 0.20 6 No 1951–87 0.334 Zhangye 38.93 100.43 1936–2008 0.30 11 Yes 1951–87 0.497 Baotou 40.66 109.85 1940–2008 0.10 12 Yes 1951–87 0.470 Shanba 40.58 107.10 1933–1990 0.15 7 No 1954–87 0.359 Yinchuan 38.48 106.22 1933–2008 0.10 5 No 1951–95 0.339 Zhongning 37.48 105.67 1940–2008 0.10 6 No 1953–87 0.392 Xining 36.62 101.77 1933–2008 0.30 14 Yes 1954–87 0.384 Yan’an 36.60 109.50 1933–2008 0.10 6 Yes 1951–87 0.309 Yulin 38.23 109.70 1933–2008 0.30 7 No 1951–94 0.302 space (Cook et al. 1999, 2004). However, the climate of freezing, and tree rings are already formed before the a given region could be highly variable because of factors cold seasons. Second, we tested the reconstructions with such as complex topography. We consider this method and without the prewhitening of the autocorrelations of more appropriate in NW China, where variable climate both predictors and predictands. The pooled autocor- conditions and complex topographic features are com- relations of various chronologies modeled through this mon. In this study, only tree-ring sites proximal to a given prewhitening process need to be added back into the station were involved in regression (Fig. 1) because the reconstruction, to preserve the low-frequency vari- spatial correlations with precipitation of distant areas are ability of climate change. Third, the retained tree-ring less meaningful (Cook et al. 1999). predictors were processed by PCA to generate or- Four steps were used in regression after the candidate thogonal eigenvectors. To reduce the dimension of the chronologies were located (Cook et al. 1999). First, the predictors and to highlight the common climate signal, candidate tree-ring chronologies were screened using we excluded the higher-order tree-ring eigenvalues correlations between climate and tree rings with a cutoff (less than 1) that accounted for little variance. Last, the two-tailed probability, to exclude chronologies in- retained orthogonal tree-ring eigenvectors were en- sensitive to the precipitation at a given station. Because tered into the regression model based on a minimum tree rings in our study region may show a 1-yr lag climate Akaike information criterion (AIC; Akaike 1974). response, tree rings of both the current year (t) and the Statistics employed to evaluate the model calibration year before (t 2 1) were taken into account to generate and verification include correlation coefficients of the doubled sets of predictors (Cook et al. 1999, 2010). The calibration period (calibration r) and verification period target precipitation data for reconstruction are the av- (verification r), as well as the reduction of error (RE). eraged monthly data from January to October. We Different from the statistics of explained variance, the chose not to use the data in November and December calculation of RE is mainly based on the actual and because temperatures in these months may fall below estimated data in the verification period. This statistic is

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3462 JOURNAL OF CLIMATE VOLUME 24

FIG. 3. Reconstructions of monthly precipitation (mm) averaged from January to October for 26 stations in NW China. The stations are listed from the (top) western to the (bottom) eastern part of NW China. thus sensitive to the quality of the climate records made between the averaged actual and reconstructed during the verification period. The theoretical range data during the warm season (from January to October) of RE is from 2‘ to 1, and a RE statistic with a value from all stations. Following the methodology of Cook et al. greater than zero is an indicator of acceptable model (1999), we additionally calculate the Pearson correlations accuracy (Cook et al. 1999). To test the homogeneity between the instrumental warm-season precipitations at and quality of the verification data, we simply examine all stations in a given year and the reconstructed data at the abrupt shifts by calculating the mean differences various stations in the same year. That is, the observa- (differences divided by the mean) between the calibra- tions for this correlation calculation are the precipitation tion and verification datasets. Visual comparisons are values of various stations in a given year, instead of a

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 1JULY 2011 F A N G E T A L . 3463

FIG. 4. Contour maps of the correlation coefficients of the calibration r and verification r, RE of the verification period, as well as mean differences between the calibration and verification periods. RE with values greater than zero indicates acceptable model ability (Cook et al. 1999). High value of ‘‘mean difference’’ indicates a large difference between mean values before (mostly from interpolated gridded data) and after the 1950s, which may be an artifact of the interpolation process. time series of precipitation records between stations. chronologies, except for the stations of Dunhuang, Qiemo, The above-mentioned procedures check for model fi- Ruoqiang, and Yulin (with the lowest search correlation delity at each station over time. However, there is no coefficients ranging from approximately 0.2 to 0.3). The guarantee that the reconstruction recovers large-scale screening probability varies from 0.05 to 0.3 to include spatial features because the only spatial component in- a sufficient number of candidate moisture sensitive chro- corporated into the PPR method is the search radius nologies to produce a reliable reconstruction. The number (Cook et al. 1999) or, at the local scale, the searching spa- of chronologies retained for the reconstruction of precip- tial coefficients. Verifications of the reconstruction in a itation at a given station ranges from 3 to 23 (Table 2), spatial sense are made using the varimax- (Kaiser 1960) which is related to the screening probability and the dis- and promax-based (Hendrickson and White 1964) rotated tribution of chronologies. For example, more chronolo- principal component analysis (RPCA; Richman 1986) to gies are involved in the regression models for Aletai and consider the variable (meteorological station) and the Tacheng stations, where a denser tree-ring network is found observations (precipitation records of each station). nearby (Table 2). Autocorrelations of the retained tree-ring chro- 3. Results and discussion nologies for regression vary from site to site. This prewhitening procedure is designed to correct the dif- a. Precipitation reconstruction ference in autocorrelations of the candidate tree-ring A number of experiments were conducted for each chronologies within a region probably cause by an issue target station to generate the best possible regression such as growth disturbances, which vary between sites model with different settings (Table 2). The search spa- (Cook et al. 1999). However, there is no guarantee that tial correlation coefficients normally did not drop be- the reconstruction with prewhitening overwhelms the low 0.4 to locate sufficient numbers of candidate tree-ring model without prewhitening (Cook et al. 1999). Most

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3464 JOURNAL OF CLIMATE VOLUME 24

tests (Gerstengarbe and Werner 1999), suggesting a degraded quality of the instrumental records before 1950 for most stations. The big difference in mean values between the estimated and actual data leads to abnormally low RE values according to its definition (e.g., Cook et al. 1999). Although RE is negative for the Qiemo and Ruoqiang stations, the two stations are well calibrated with reliable precipitation data since the 1950s. RE statistics for the Kuche and Yining stations in central Xinjiang are negative, but the values of the verification r for these two stations are relatively high (0.44 and 0.52, respectively). We consider these four reconstructions reasonably accurate in modeling past precipitation, and the poor verification results may be an artifact because of the degraded quality of interpolated FIG. 5. (top) Comparison of instrumental and reconstructed precipitation before the 1950s. However, there is limited precipitation, as well as the associated number of stations. (bottom) success in either calibration or verification for the Hetian Correlations are calculated for precipitation at various stations in reconstruction, suggesting special caution should be taken a given year between the instrumental and reconstructed data during the common period 1941–87. All the correlations are sig- in the case of this reconstruction (Fig. 4). nificant at the 0.000 01 level. There are reconstructions (stations Hami, Jiuquan, Lanzhou, Wuwei, and Xining) in the central and eastern NW areas with relatively poor verification results. The of the reconstructions (16 stations) are better cali- reconstructions at the Lanzhou and Xining stations still brated after prewhitening each chronology before the have some potential for recovering past climate because modeled pooled ‘‘red noise’’ signal was added back in the calibration r for Lanzhou is extremely high and the (Table 2), suggesting that corrected autocorrelation verification r for Xining is not very low. However, the modeling is more suitable for reconstruction in this reconstruction models for stations Hami, Jiuquan, and area with sparsely distributed chronologies. The re- Wuwei show rather poor model fitness for both cali- constructeddatawerethenscaledtotheinstrumental bration and verification, suggesting that caution should data to equalize the mean and variance during the pe- be used when considering these three stations. Poor riod of overlap (Esper et al. 2005). The most recent calibration and verification results for the stations over instrumental data, that were not included in the re- central NW China may be related to the sparseness of gression process, were added to the scaled reconstructed tree-ring chronologies over that region. Future spatial data to generate the longest possible precipitation time reconstruction models could be improved when more series. In so doing, the common period for all the re- tree-ring chronologies are available, particularly for areas constructions is from 1802 to 1990, and the longest with sparse coverage, such as central NW China. In sum- single reconstruction is from 1711 to 2008 at the Hetian mary, we advise caution when using the reconstructions at station (Fig. 3). four stations (Hetian, Hami, Jiuquan, and Wuwei). For a given year, correlations between the instrumental b. Calibration and verification and reconstructed data of all stations are highly signifi- The explained variance varies from 28.3% at Wuwei cant (0.000 01 level), with some relatively low values in station to 55.7% at Lanzhou (Table 1). The relatively 1941, 1942, 1947, 1955, and 1972 (Fig. 5). The map cor- poor verification results found in southern Xinjiang, as relations indicate that the actual and estimated data indicated by low verification r and RE at Hetian, Qiemo, agree with each other better after the 1950s, as indicated and Ruoqiang, may be related to the poor quality of by a pattern of generally increasing correlation values verification precipitation data because the differences (Fig. 5). This is probably an indicator that the instru- between the verification and calibration periods are mental data after the 1950s are more reliable. Rela- particularly high for southern Xinjiang (Fig. 4). For ex- tively low correlations in 1941, 1942, 1947, and 1955 are ample, precipitation for Ruoqiang station in the veri- indicated by clear mismatches of the averaged actual fication period (1941–52) is 10 times higher than during and estimated values in these years (Fig. 5). However, the calibration period (1953–92). Actually, there is a the relatively low correlation in 1972 does not corre- significant (95%) abrupt shift around 1950 for most spond to an obvious mismatch of the same year (Fig. 5). stations (21 stations). as indicated by Mann–Kendall This is probably because the minor disagreements (e.g.,

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 1JULY 2011 F A N G E T A L . 3465

FIG. 6. Instrumental varimax precipitation factors (monthly mean from January to October) for NW China based on RPCA over the common period 1941–90. abnormally high and low values) between the instru- NW factor’’ (Fig. 6). Similar spatial patterns occurring in mental and reconstructed data during that year were the instrumental and reconstructed data suggest that the averaged out, leading to close mean values. It should be reconstructed model does well at recovering these spa- noted that the reconstructed averaged monthly pre- tial features (Figs. 6 and 7). Varimax factor (VF) 2 for cipitation at some very dry stations in Xinjiang fall below the reconstructions corresponds to VF1 of the instru- zero for a few years (Fig. 3). This largely resulted from the mental data (eastern NW factor), though the reconstruc- scaling process, a linear transformation that is designed to tion VF1 is more concentrated on the southeastern equalize the mean and variance of reconstruction with region. VF2 of the reconstructed data (northern Xinjiang) instrumental data (e.g., Esper et al. 2005). and reconstruction VF4 (southern Xinjiang) correspond to instrumentation VF2 and instrumentation VF3 (Fig. 6), c. Spatial patterns of the actual and estimated data respectively. However, instrumentation VF4 (Fig. 6), the The first four principal components account for 61.5% central NW factor centering at Jiuquan, has moved east- and 49.5% of the total variance of the instrumental and ward for the reconstruction VF4 with the highest loadings estimated data, respectively, which were retained for over Wuwei (Fig. 7). This might be another indicator rotation using varimax and promax methods. Since the of the relatively poor reconstructions associated with varimax and promax rotations showed similar results, some central NW China stations (Jiuquan and Wuwei), we only show results for the varimax rotations for in- which modify the actual spatial precipitation patterns strumental and reconstructed data (Figs. 6 and 7). For over these areas. Meanwhile, we need to consider the the varimax rotation of the instrumental data over the possibility that the spatial ranges of different climate common period 1941–90, factor 1 yields the highest divisions might shift over time. loading on eastern NW China, which is thus referred to Both the instrumentation VF1 and reconstruction as the ‘‘eastern NW factor’’ (Fig. 6). Similarly, factor 2 is VF1 (Figs. 6 and 7) are similar to the previous drought named the ‘‘southern Xinjiang factor,’’ factor 3 is the classifications in eastern NW China (Li et al. 2009; Shi ‘‘northern Xinjiang factor,’’ and factor 4 is the ‘‘central et al. 2007; Zou et al. 2005). The previously classified

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3466 JOURNAL OF CLIMATE VOLUME 24

FIG. 7. Varimax precipitation factors (monthly mean from January to October) from RPCA for NW China over the common reconstruction period of 1802–1990. western ‘‘Xinjiang’’ factor in NW China is separated scores of the reconstructed data and the corresponding into southern (instrumentation VF2 and reconstruction varimax factors of the instrumental data are shown in VF4) and northern (instrumentation VF3 and recon- Fig. 8. The scores of the first three varimax factors of the struction VF2) Xinjiang factors. This is confirmed by reconstructed data match the corresponding varimax previous climate classifications at a more local scale factor scores of the instrumental data well. However, (Qian and Qin 2008). The transitional area between there are discrepancies between the scores of the western and eastern NW China, as identified in previous southern Xinjiang factor, that is, reconstruction VF4 and studies (Qian and Qin 2008; Shi et al. 2007), is compa- instrumentation VF2. This may be related to the de- rable to the central NW China factor in this study. It graded quality of instrumental data before the 1950s in should be noted that the southern Xinjiang factor shows the southern Xinjiang area, as indicated by the large a lower explained variance (8.9%) for the reconstruc- differences in mean values before and after 1950, which tion data (reconstruction VF4) than the explained vari- results in high normalized varimax scores before ap- ance (16.8%) for the instrumental data (instrumentation proximately 1950 and relatively suppressed low nor- VF2), which may be due to the limited number of tree- malized varimax scores after 1950 (Fig. 8). Meanwhile, ring chronologies in southern Xinjiang. This indicates the instrumentation VF2 with higher explained variance that future improvements are needed to sample more shows high loadings over the larger area, whereas the evenly distributed chronologies. reconstruction VF4 has smaller explained variance, leading to changed factor scores. d. RPCA factor scores The most severe dry years for VF1 are in 1862 and Moisture variables in NW China are highly diverse— 1932. The dry conditions in 1932 fall within a well-known for example, between its western and eastern parts drought event from the 1920s through the 1930s (Fang (Chen et al. 2008; Shi et al. 2007); therefore, we did not et al. 2009; Liang et al. 2006). The widespread drought extract the first principal component to represent the from the 1920s through the 1930s was most severe for leading mode of the entire NW China, instead we discuss the east-central factor (factor 3 in Fig. 8), an event that the varimax rotated factors. The four varimax factor is particularly significant for stations Baotou, Shanba,

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 1JULY 2011 F A N G E T A L . 3467

FIG. 8. Normalized scores of the four VFs over the common reconstruction period of 1802– 1990 as shown in Fig. 7, as well as their extreme dry (,2 std dev) and wet (.2 std dev) years. The corresponding VF scores of the instrumental data (Fig. 6) are also shown as comparisons.

Yinchuan, Yulin, Zhangye, and Zhongning (Fig. 3). The to locate the candidate tree-ring chronologies, instead of extreme drought in 1829 appears to cover the entire the previously used ‘‘search radius’’ method. The cali- Xinjiang area, which is significant for both the northern bration and verification procedures generally follow the (instrumentation VF4) and southern Xinjiang (recon- standard methodologies used in previous studies. In- struction VF4) factors. A remarkable drought in 1944– dependent data for verification were extracted from his- 45, evident in previous studies (e.g., Li et al. 2006), is torical or gridded datasets. Based on the calibration and most extreme in northern Xinjiang (VF2 in Fig. 8) but verification results, we advise caution when using the not significant in southern Xinjiang. The extreme reconstructions at Hami, Hetian, Jiuquan, and Wuwei. drought in 1945 is particularly apparent in Aletai, Hami, Additional comparisons of spatial patterns of the actual Wulumuqi, Wusu, and Yining (Fig. 3). The extremely and estimated data suggested that the reconstructions wet conditions in 1958, seen in reconstruction VF2 and generally match the spatial patterns of the instrumental reconstruction VF3, correspond to the wet period in the data, although the spatial boundaries of some factors vary 1950s, as revealed in previous studies for this area (e.g., between the instrumental and reconstructed datasets. For Fang et al. 2010a). For reconstruction VF4 in southern example, the central NW factor is shifted a long way Xinjiang, the driest year is 1956, followed by 1960 and eastward in the reconstructed data. This may be related to 1917. According to the instrumental precipitation re- the poor reconstructions at a few stations (Wuwei and cords, it was particularly dry in 1956 in Hetian, Kashi, Jiuquan). Some severe drought events are significant in and Ruoqiang. the factor scores of the reconstructed data. A spatial reconstruction for a climate dataset with much denser spatial coverage could be beneficial to the study of cli- 4. Concluding remarks mate variations over space and time, but it will require We have developed a tree-ring network of 132 tree- the generation of additional long tree-ring chronologies. ring chronologies from ring-width data (limited density data) in NW China and the vicinity. This tree-ring net- Acknowledgments. The authors acknowledge Jianfeng work was utilized to reconstruct a 26-station precipitation Peng, Yongxiang Zhang, Yong Zhang, Qinghua Tian, Tao dataset in NW China, derived from a nationwide 160- Yang, and other persons for their kind help in the field and station dataset, using a PPR spatial reconstruction method. in the lab. We are grateful to Fritz Schweingruber, Gordon We introduced a search spatial correlation contour method Jacoby, Paul Sheppard, Qibing Zhang, Xuemei Shao,

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 3468 JOURNAL OF CLIMATE VOLUME 24

Neil Pederson, and other scientists who have contrib- minimum temperatures in northeastern Tibetan Plateau: Evi- uted their tree-ring data in the study region to the dence from tree rings. Sci. China, 51D, 41–55. publicly available database. We appreciate the helpful Hendrickson, A. E., and P. O. White, 1964: Promax: A quick method for rotation to oblique simple structure. Br. J. Stat. comments from Peregrine Gerard-Little and two anon- Psychol., 17, 65–70. ymous reviewers. This research was supported by the Kaiser, H. F., 1960: The application of electronic computers to National Basic Research Program of China (973 Pro- factor analysis. Educ. Psychol. Meas., 20, 141–151. gram) (2009CB421306), the National Science Foun- Lau, K. M., and H. Weng, 2001: Coherent modes of global SST and dation of China (40971119 and 41001115), the NSFC summer rainfall over China: An assessment of the regional impacts of the 1997–98 El Nin˜ o. J. Climate, 14, 1294–1308. Innovation Team Project (41021091), and the Chinese Li, J., X. Gou, E. R. Cook, and F. Chen, 2006: Tree-ring based 111 Project (B06026). drought reconstruction for the central Tien Shan area in northwest China. Geophys. Res. Lett., 33, L07715, doi:10.1029/ REFERENCES 2006GL025803. ——, E. R. Cook, R. D’Arrigo, F. Chen, and X. Gou, 2009: Akaike, H., 1974: A new look at the statistical model identiﬁcation. Moisture variability across China and Mongolia: 1951–2005. IEEE Trans. Autom. Control, 19, 716–723. Climate Dyn., 32, 1173–1186, doi:10.1007/s00382-008-0436-0. Briffa, K. R., and P. D. Jones, 1990: Basic chronology statistics and Liang, E. Y., and Coauthors, 2006: The 1920s drought recorded by assessment. Methods of Dendrochronology: Applications in tree rings and historical documents in the semi-arid and arid the Environmental Sciences, E. R. Cook and L. A. Kairiukstis, areas of northern China. Climatic Change, 79, 403–432. Eds., Kluwer Academic, 137–152. Liu, Y., L. Ma, S. W. Leavitt, Q. Cai, and W. Liu, 2004: A pre- ——, ——, T. M. L. Wigley, J. R. Pilcher, and M. G. L. Baillie, 1986: liminary seasonal precipitation reconstruction from tree-ring Climate reconstruction from tree rings: Part 2, Spatial re- stable carbon isotopes at Mt. Helan, China, since AD 1804. construction of summer mean sea-level pressure patterns over Global Planet. Change, 41, 229–239. Great Britain. J. Climatol., 6, 1–15. Meko, D. M., E. R. Cook, D. W. Stahle, C. W. Stockton, and Chen, F., and Coauthors, 2008: Holocene moisture evolution in M. K. Hughes, 1993: Spatial patterns of tree-growth arid central Asia and its out-of-phase relationship with Asian anomalies in the United States and southeastern Canada. monsoon history. Quat. Sci. Rev., 27, 351–364. J. Climate, 6, 1773–1786. Cook, E. R., 1985: A time-series analysis approach to tree-ring New, M. G., M. Hulme, and P. D. Jones, 2000: Representing standardization. Ph.D. dissertation, The University of Ari- twentieth-century space–time climate variability. Part II: De- zona, 171 pp. velopment of 1901–96 monthly grids of terrestrial surface cli- ——, D. M. Meko, D. W. Stahle, and M. K. Cleaveland, 1999: mate. J. Climate, 13, 2217–2238. Drought reconstructions for the continental United States. Qian, W. H., and A. Qin, 2008: Precipitation division and cli- J. Climate, 12, 1145–1162. mate shift in China from 1960 to 2000. Theor. Appl. Cli- ——, C. A. Woodhouse, C. M. Eakin, D. M. Meko, and D. W. matol., 93, 1–17. Stahle, 2004: Long-term aridity changes in the western United Richman, M. B., 1986: Rotation of principal components. Int. States. Science, 306, 1015–1018. J. Climatol., 6, 293–335. ——, K. J. Anchukaitis, B. M. Buckley, R. D. D’Arrigo, G. C. Seager, R., 2007: The turn of the century North American drought: Jacoby, and W. E. Wright, 2010: Asian monsoon failure and Global context, dynamics, and past analogs. J. Climate, 20, megadrought during the last millennium. Science, 328, 486–489. 5527–5552. Esper, J., D. C. Frank, R. J. S. Wilson, and K. R. Briffa, 2005: Effect Shao, X. M., L. Huang, H. B. Liu, E. L. Liang, X. Q. Fang, and L. L. of scaling and regression on reconstructed temperature Wang, 2005: Reconstruction of precipitation variation from amplitude for the past millennium. Geophys. Res. Lett., 32, tree rings in recent 1000 years in Delingha, Qinghai. Sci. L07711, doi:10.1029/2004GL021236. China, 48D, 939–949. Fang, K., and Coauthors, 2009: Drought variations in the eastern Sheppard, P. R., P. E. Tarasov, L. J. Graumlich, K.-U. Keussner, part of northwest China over the past two centuries: Evidence M. Wagner, H. O¨ sterle, and L. G. Thompson, 2004: Annual from tree rings. Climate Res., 38, 129–135. precipitation since 515 BC reconstructed from living and fossil ——, N. Davi, X. Gou, F. Chen, E. Cook, J. Li, and R. D’Arrigo, juniper growth of northeast Qinghai Province, China. Climate 2010a: Spatial drought reconstructions for central High Asia Dyn., 23, 869–881. based on tree rings. Climate Dyn., 35, 941–951, doi:10.1007/ Shi, Y., Y. Shen, E. Kang, D. Li, Y. Ding, G. Zhang, and R. Hu, s00382-009-0739-9. 2007: Recent and future climate change in northwest China. ——, X. Gou, F. Chen, R. D’Arrigo, and J. Li, 2010b: Tree-ring Climatic Change, 80, 379–393. based drought reconstruction for Guiqing Mountain (China): Yuan, Y., L. Jin, X. Shao, Q. He, Z. Li, and J. Li, 2003: Varia- Linkages to the Indian and Paciﬁc Oceans. Int. J. Climatol., 30, tions of the spring precipitation day numbers reconstructed 1137–1145, doi:10.1002/joc.1974. from tree rings in the Urumqi River drainage, Tianshan Gemmer, M., S. Becker, and T. Jiang, 2004: Observed monthly Mts. over the last 370 years. Chin. Sci. Bull., 48, 1507–1510. precipitation trends in China 1951-2002. Theor. Appl. Clima- Zhang, Q.-B., G. Cheng, T. Yao, X. Kang, and J. Huang, 2003: A tol., 77, 39–45. 2,326-year tree-ring record of climate variability of the north- Gerstengarbe, F. W., and P. C. Werner, 1999: Estimation of the eastern Qinghai-Tibetan Plateau. Geophys. Res. Lett., 30, 1739, beginning and end of recurrent events within a climate regime. doi:10.1029/2003GL017425. Climate Res., 11, 97–107. Zou, X., P. Zhai, and Q. Zhang, 2005: Variations in droughts over Gou,X.,F.Chen,M.Yang,G.Jacoby,K.Fang,Q.Tian,and China: 1951–2003. Geophys. Res. Lett., 32, L04707, doi:10.1029/ Y. Zhang, 2008: Asymmetric variability between maximum and 2004GL021853.

Unauthenticated | Downloaded 10/01/21 02:00 AM UTC 15 FEBRUARY 2012 C O R R I G E N D U M 1357

CORRIGENDUM

KEYAN FANG,XIAOHUA GOU, AND FAHU CHEN Key Laboratory of Western China’s Environmental Systems, Ministry of Education, Lanzhou University, Lanzhou, China

EDWARD COOK,JINBAO LI,BRENDAN BUCKLEY, AND ROSANNE D’ARRIGO Tree-Ring Lab, Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York

In Fang et al. (2011), information about one of the corresponding authors was left out of the contact information at the bottom of the title page of the ﬁnal published paper. The correct contact information for the corresponding authors is the following:

Corresponding authors’ address: Keyan Fang and Xiaohua Gou, Key Laboratory of Western China’s Environmental Systems (Ministry of Education), Lanzhou University, 222 South Tianshui Road, Lanzhou 73000, China. E-mail: [email protected] and [email protected]

The staff of the Journal of Climate regrets any inconvenience this error may have caused.

REFERENCE

Fang K., X. Gao, F. Chen, E. Cook, J. Li, B. Buckley, and R. D’Arrigo, 2011: Large-scale precipitation variability over northwest China inferred from tree rings. J. Climate, 24, 3457–3468.

DOI: 10.1175/JCLI-D-11-00584.1

Ó 2012 American Meteorological Society Unauthenticated | Downloaded 10/01/21 02:00 AM UTC