Precipitation reconstruction based on tree-ring width over the 1 past 270 years in the central Lesser Khingan Mountains , 2 Northeast China 3

Abstract. Inter-annual variations in precipitation play important roles in management of forest ecosystems and agricultural production in Northeast China. This study presents a 270-year precipitation reconstruction of winter to early growing season for the central Lesser Khingan Mountains, Northeast China based on tree-ring width data from 99 tree-ring cores of Pinus koraiensis Sieb. et Zucc. from two sampling sites near Yichun. The reconstruction explained 43.9 % of the variance in precipitation from the previous October to current June during the calibration period 1956–2017. At the decadal scale, we identified four dry periods that occurred during AD 1748–1759, 1774–1786, 1881–1886 and 1918–1924, and four wet periods occurring during AD 1790–1795, 1818–1824, 1852–1859 and 2008–2017, and the period AD 2008–2017 was the wettest in the past 270 years. Power spectral analysis and wavelet analysis revealed cyclic patterns on the inter-annual (2–3 years) and inter-decadal (~11 and ~32–60 years) timescales in the reconstructed series, which may be associated with the large-scale circulation patterns such as the Arctic Oscillation and North Atlantic Oscillation through their impacts on the Asian polar vortex intensity, as well as the solar activity.



31
Precipitation is one of the most important climate variables in the global climate system and affects 32 human society via its impacts on water resources, agricultural production, and ecosystems. In recent 33 years, extreme droughts and flooding events repeatedly occurred in many regions of the world, 34 which have brought heavy losses in economy and human life. However, the scarcity of long-term 35 instrumental climatic data and historic records in many regions impedes our understanding to the 36 spatiotemporal precipitation variability and hampers our ability to plan for future. Additionally, 37 unlike temperature variation displaying relatively persistent patterns over large regions, 38 precipitation tends to have strong spatial variability. Therefore, spatially explicit and long-term data 39 are essential for understanding the current variation patterns and trends in the historical and spatial 40 context, which is also important for both validation of climate models and integration and 41 comparison with other historical, archaeological, and proxy data (Cook et al., 2010). 42 Tree-ring based reconstructions play an important role in paleoclimate studies due to their 43 accurate dating, annual resolution, wide distribution and good replication (Briffa et al., 1990;Cook 44 et al., 2000;Scuderi, 1993;Lamarche, 1974;Jacoby et al., 1996;Hughes et al., 1984;Shao et al., 45 2005). In China, many tree-ring-based paleoclimate reconstructions are available in different 46 regions, such as the Tibetan Plateau (Zhang et al., 2003;Liang et al., 2009), Xinjiang Province 47 (Chen et al., 2014, Helanshan Mountain (Liu, 2004), and Hengduan Mountain (Fan et al., 2008;Li 48 et al., 2017). In comparison with these regions, long-term tree-ring based paleoclimatic records are 49 still relatively sparse for eastern China overall, including Northeast China, mostly due to long 50 history of human activities that have removed most old-growth forests. 51 The Lesser Khingan Mountains in Northeast China (Fig. 1) extends over 450 km from south 52 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License.

Meteorological and circulation data 126
Climatic data records at the Yichun meteorological station (128.92˚E, 47.73˚N; 240.9 m a.s.l., 127 (http://climatedataguide.ucar.edu/cliamte-data/standardized-precipitation-evapotranspiration-133 index-spei) to calculate spatial correlations with the TRW chronology, and the gridded CRU TS 4.02 134 precipitation data for the period 1956-2017 (www.cru.uea.ac.uk) to further explore the spatial 135 representativeness of the reconstructed precipitation. 136 In addition, in order to discuss the possible driving factors that affected the precipitation regime, 137 we collected the Asian polar vortex intensity (APVI) data, a measure determined by the total air 138 mass quantity or density between 500 hPa geopotential height field and the isohypsic surface located 139 that the polar vortex southern boundary characteristic contour covering 60-150°E in Northern 140 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License.

Radial growth -climate relationships, reconstruction calibration and verification 149
To investigate the tree growth-climate relationships, we calculated the Pearson's correlation 150 coefficients between the TRW chronology and TMEAN, TMAX, TMIN and PPT during the 151 instrumental period of 1956-2017. Since the climate of a given year could have a lagged effect on 152 the growth in the following year (Fritts, 1976), climate data from the previous October to the current 153 September were used in the correlation analysis. To test whether the correlation coefficients were 154 affected by variations in the low-frequency domain, we also calculated the correlation coefficients 155 using the first-differences of the chronology and the climatic data. The results can give us hints on 156 which climate variable served as the major limiting factor of radial growth of trees, the potential 157 target for reconstruction. 158 In reconstruction, we first established a transfer function using linear regression in which the 159 TRW chronology was used as the independent variables and the selected climatic factor as the 160 dependent variable for the full calibration period. To validate the transfer function, the cross-161 validation procedure (Michaelsen, 1987) and independent split-period validation procedure (Fritts,162 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. 1976) were used in this study. The validation statistics include the sign tests on both the original and 163 first-difference data and t test of product means to show how well the model-predicted values 164 following the directions of variation in the observed values (Fritts, 1976). Also included are 165 reduction of error (RE), coefficient of efficiency (CE) and correlation coefficient. RE is a measure 166 of comparison between the predicted and observed values (Fritts, 1976), and CE is a relative 167 measure of the analysis error variance to the variance in the true state (Nash and Sutcliffe, 1970;168 Tardif et al., 2014). Positive RE and CE values are evidence for a valid transfer function (Fritts, 169 1976;Nash and Sutcliffe, 1970). 170

Power spectral analysis and wavelet analysis 171
Spectral analysis is the process of estimating the power spectrum of a signal from its time-172 domain representation. To examine the temporal variation pattern of precipitation in our study area 173 in different frequency domain, we performed power spectral analysis (Fowler, 2010) and wavelet 174 analysis (Torrence and Compo, 1998). 175

177
The two sites are very close, and the correlation coefficients between each series and master 178 dating series of flagged 50-year segments (lagged 25-year) filtered with 32-year spline were 0.61-179 0.80 calculated using the COFECHA software. Therefore, we combined the tree-ring width data 180 when developing the TRW chronology. The TRW chronology covered the periods AD 1685-2017 181 (Fig. 3). The statistical characteristics of the chronology are given in Table 2. The mean sensitivity 182 (a measure of the inter-annual variability in tree-ring series) was 0.223, indicating that the TRW 183 chronology showed relatively low inter-annual variability compared to those chronologies from 184 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. semi-arid area (Shao et al., 2010). The first-order autocorrelation of the TRW series was 0.31, 185 suggesting that the radial growth was probably influenced by conditions of previous years. The Rbar 186 (overall mean correlations between the sample series), Rbt (mean between-tree correlations), and 187 Rwt (mean within-tree correlations) were 0.258, 0.251 and 0.801, respectively. They were 188 comparable to other tree ring studies in the region (e.g., Yin et al., 2009  and TMAX in June for the first-difference data (Fig. 4B). We also calculated the correlations 214 between the TRW chronology and climatic variables for different combinations of months/seasons. 215 The strongest correlation was produced using a combined variable of previous October-current June 216 total precipitation for the origin data (r=0.663, p<0.01), which was also statistically significant for 217 the first-difference data (r=0.438, p<0.01). These results suggest that the cold-season and early 218 growing-season precipitation is a major factor of radial growth of trees at our sampling site, with its 219 effects detectable in the TRW series variations in both low-and high-frequency domains.  Table 3). The sign test is statistically significant at the 0.01 level for the original data, but it 230 was not significant for the first-difference data. The result indicated that the match between the 231 reconstructed and observed rainfall data was better in the low-frequency domain than that in the 232 high-frequency domain. The relatively high values of RE and product mean t indicated reasonable 233 skill in the reconstruction with a leave-one-out correlation coefficient of 0.63. The results of split-234 period validation are also presented in Table 3. In the first split-period validation, the calibration 235 period was set to be 1956-1986, and validation period as 1987-2017. The calibration model 236 explained 21.4% of the variance in PPTp10-c6. Results of the signs tests for the original data (ST) and 237 first-difference data (ST1) were not significant at the 95% confidence level, but the RE and CE 238 values are above zero and the t value of product mean is high, again suggesting reasonable skills for 239 reconstruction with a correlation coefficient of 0.729 for the original-reconstructed climate in the 240 verification period. For the second split-period validation, the period 1987-2017 was used for 241 calibration and 1956-1986 for validation. The model explained 53.2% of the variance in PPTp10-c6. 242 The sign test of the original data reached the 95% confidence level, but the result of the first-243 difference data was not statistically significant. The correlation coefficient, RE, and CE were lower 244 than those of the first split-period validation, but remained positive, and the product mean t value 245 remained high. The validation results suggested that the model was relatively robust with sufficient 246 skills of estimation. The reconstructed precipitation series derived from the model showed a good 247 agreement with the observed precipitation values during the calibration period (Fig. 5B). 248 <Insert Table 3 here> 249 Temporal variation of the reconstructed precipitation 250 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License.
The reconstruction period began in AD 1748 when the TRW series' EPS exceeded 0.85 (Table  251 2 and Fig. 3). but weaker than those between the TRW chronology and the precipitation from previous October to 282 current June. These results also supported the conclusion that moisture is the major factor affecting 283 the growth of Korean pine at our study sites. 284 <Insert Fig. 6 here> 285 In this region, there have been more reconstructions of the past temperature than precipitation. 286 For example, even at a site very close to ours (also in the Wuying District), Yin et al. (2009)  287 reconstructed temperature variations of the previous October using the same tree species. A further 288 comparison revealed that they used the residual chronology rather than the standard chronology, and 289 also used climatic data from a different meteorological station (Wuying rather than Yichun). In their 290 study, the only month of statistically significant correlations was October of the previous year and 291 they did not conduct correlation analysis for the first-difference data. After first differencing in our 292 analysis, all positive correlations with temperature variables became statistically insignificant at the 293 0.95 confidence level. Therefore, we are confident that the growth-precipitation relationship as 294 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. displayed in Fig. 4 is more robust than the relationship between tree growth and temperature. 295 However, we also speculate that this relationship may have been strengthened due to the recent 296 and PDSI series (Fig. 7), suggesting persistent large-scale weather conditions affecting the entire 319 Northeast China. 320 <Insert Fig. 7 here> 321 The 1920s drought was one of the most severe and well-documented natural hazards in the last 322 200 years in the semi-arid and arid areas of northern China (Liang et al., 2006). In the Wuying area, 323 the 1920s was a dry period with the driest year in 1920 (Fig. 7). For the entire 1920s, however, the 324 moisture conditions gradually recovered from the low. Based on gridded temperature and 325 precipitation data, Ma et al. (2005)

analyzed the shift of dry/wet boundaries for different regions in 326
China during 1900-2000. They discovered that for Northeast China, there was a wetting trend during 327 the 1920s, with the boundary of the semi-arid and sub-humid regions shifting westward from 128°E 328 to 124°E, which was then reversed in the early 1930s (Ma and Fu, 2005). In the meantime, most 329 other regions in China experienced the peak drought conditions during the late 1920s and early 330 1930s (Liang et al. 2006). Therefore, most likely this severe drought did not reach our study region 331 where the 1920 drought was a separate event impacting various regions in Northeast China (Fig. 7). 332 To further explore the spatial representativeness of the reconstructed precipitation series, we 333 calculated correlation coefficients between the observed (Fig. 6B) and reconstructed (Fig. 6C) 334 PPTp10-c6 data for the period 1956-2017 using the gridded CRU TS 4.02 dataset (www.cru.uea.ac.uk) 335 and plotted the results using KNMI Climate Explorer (https://climexp.knmi.nl/). The reconstructed 336 PPTp10-c6 correlated significantly with the gridded precipitation over a region covering 337 approximately 42-52˚N and 124-132˚E (r>0.5, p<10%) (Fig. 6C), displaying a similar spatial 338 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. structure of the correlations (although weaker) between the observed PPTp10-c6 and the gridded 339 precipitation data (Fig. 6B). These results indicated that our precipitation reconstruction can capture 340 the occurrences of drought events in a large area in the northern part of Northeast China. 341

Possible driving mechanisms 342
To examine the temporal variation pattern of precipitation in the Wuying area in different 343 frequency domains, which may allow us to explore possible driving factors that affected the 344 precipitation regime, we performed power spectral analysis of the reconstruction series and 345 discovered semi-cyclic variations with periods of 2.2-3.2 years, 11 years, and 30 years (Fig. 8A). 346 Wavelet analysis also confirmed these results, showing cyclic periodicities of 2-3 years, ~11 years, 347 and ~30-64 years (Fig. 8B). 348 <Insert Fig. 8 here> 349 Since early growing season moisture condition is the limiting factor of radial growth of trees 350 and more than 60% of the observed PPTp10-c6 occurs in May and June, explaining more than 71% 351 of the total variance in PPTp10-c6, we will focus on the atmospheric processes that influence May-352 June precipitation in the following. At this time of the year, previous studies indicated that 353 precipitation in this region is mostly caused by extratropical cyclonic activities that are impacted by 354 the Asian Polar Vortex Intensity (APVI) (Zhang and Li, 2013), The correlation between the APVIp10-355 6 and the observed PPTp10-c6 at Yichun was -0.275 (p = 0.033), while its correlation with the 356 reconstructed series was -0.243 (p = 0.051). We argue that the APVI in May and June (APVIc56) 357 would have a significant impact on PPTp10-c6. This was validated by the correlations of the APVIc56 358 with the observed (r = -0.375, p = 0.002) and reconstructed (r = -0.269, p = 0.029) PPTp10-c6 series. 359 Therefore, in the following, we will focus on the relationships between APVIc56 and various large-  , 1997;Hurrell, 1995;Yao et al., 2017).. 366 Both the ENSO and PDO did not show any significant correlation with the APVIc56 (Table 4). 367 However, AO and NAO showed significant positive correlations with APVIc56 (Table 4). Since the 368 AO and NAO time series are highly correlated to each other (Ambaum et al., 2001), we further 369 analyzed the temporal variation patterns of a reconstructed monthly NAO series since 1659 370 (Luterbacher et al., 2002). The correlation coefficient between the reconstructed May-June NAO 371 and reconstructed PPTp10-c6 was -0.118 (p = 0.061) for the common period 1748-2001, while the 372 correlation between the two series after 5-year smoothing was -0.229 (p=0.2) after adjusting degree 373 of freedom according to the formula calculated by Bretherton et al. (1999). On the decadal scale, 374 the inverse correlation between the reconstructed NAOc56 and reconstructed PPTp10-c6 exists (Fig. 375 9). Power spectral analysis of this NAO series showed statistically significant cyclic patterns of 2.7-376 3.2 years and 50-60 years, which matched the periodicities in the reconstructed PPTp10-c6 series (Fig. 377 8a). This specific reconstructed May-June NAO series did not show a 30-year cyclic pattern. 378 However, it existed in a multi-proxy NAO reconstruction by Trouet et al. (2009). Finally, the 11-379 year cycle in the reconstructed series matched the 11-year sunspot cycle, probably due to its impact 380 on the Asian Polar vortex at the 300 hPa geopotential height (Angell, 1992). Overall, we identified 381 the Asian Polar Vortex as the possible regional control factor of winter-early summer precipitation 382 in our study region, while AO and NAO are the most likely large-scale circulation patterns that 383 influence the inter-annual variation of precipitation in the Lesser Khingan Mountains. Contrary to 384 some previous studies (Zhang et al., 2018c), ENSO and PDO were not found to be related to winter-385 early growing season precipitation in our study area. 386 <Insert Table 4 here> 387 <Insert Fig. 9 here> 388

389
In this study we reconstructed winter to early growing-season precipitation based on the ring- 270-year precipitation reconstruction showed good spatial representation and revealed four dry 398 periods that occurred during AD 1748-1759, 1774-1786, 1881-1886and 1918-1924, with AD 1774 1786 as the driest. It also revealed four wet periods occurring during AD 1790-1795, 1818-1824, 400 1852-1859and 2008-2017, and the period AD 2008-2017 was the wettest in the past 270 years on 401 the decadal scale. In addition, although 1920 was a dry year in our study area, the severe drought 402 that hit many regions in North China during the late 1920s most likely spared this region. The results 403 of power spectral analysis and wavelet analysis revealed cyclic patterns of 2.3-3.2 years, 11 years, 404 https://doi.org/10.5194/cp-2020-56 Preprint. Discussion started: 18 June 2020 c Author(s) 2020. CC BY 4.0 License. and 30-64 years in the reconstructed precipitation series, which matched those of a reconstructed 405 NAO series and the 11-year sunspot cycle. Our results suggest that the Asian Polar Vortex is 406 probably the regional control factor of the inter-annual variation of winter-early growing season 407 precipitation, while NAO and AO are the associated large-scale circulation patterns. Results from 408 our study indicated that even in a cold and relatively humid climate, moisture condition can still 409 serve as a control factor for radial growth of trees, which provides more opportunities for climatic 410 reconstructions of precipitation to enhance spatial coverage of sampling sites as precipitation tends 411 to have strong spatial variability. This may also have significant implications in forest and 412 ecosystems management and agricultural production.