1 Introduction

Marked increase in global temperature has been observed since the early twentieth century (IPCC 2013; Karl et al. 2015). However, this warming is spatially non-uniform, and a large percentage of rapid warming rates are found over the high-altitude regions in addition to the Arctic (Wang et al. 2014a; Pepin et al. 2015). The Tibetan Plateau (TP), for example, experienced a temperature increase of 0.33 °C/decade during 1961–2012, which is roughly 0.13 °C/decade higher than the global average (Yan and Liu 2014). The TP warming rate did not abate during the recent global warming hiatus since 1998 (Yan and Liu 2014; Pepin et al. 2015), suggesting it being a region of robust response to anthropogenic radiative forcing. State-of-the-art climate models projected that rapid temperature increase on the TP will persist throughout the twenty-first century (Rangwala et al. 2013; Su et al. 2013).

The effects of rapid high-altitude warming are dramatic and widespread. On the TP, extensive glacial shrinkage and permafrost degradation have been observed since the beginning of instrumental measurements in the mid-twentieth century, with accelerating rates over recent decades (Kang et al. 2010; Yao et al. 2012; Wu et al. 2013). Meanwhile, earlier thawing and later freezing of soil have occurred, leading to a substantial reduction in the number of frozen days (Li et al. 2012). The length of the growing season has increased at a rate of roughly 3 days per decade during the past half century, largely owing to an earlier start of the growing season (Dong et al. 2012). Interestingly, some plants on the TP delayed the onset of their growth in spring due to rapid winter temperature increase that triggered a later fulfillment of chilling requirements (Yu et al. 2010), although other factors may complicate such an explanation (Chen et al. 2011).

Moisture-related change accompanying rapid warming on the TP is complicated and exhibits considerable spatiotemporal heterogeneity over the past few decades. Seasonally speaking, precipitation has overall increased in winter and spring but decreased slightly in summer and autumn (Li et al. 2010; Chen et al. 2013). Spatially, the annual precipitation has increased in the northeastern and southeastern regions, but decreased in the northwest and the east edge of the TP (Chen et al. 2013; Yang et al. 2014a). Although spatially coherent patterns are found for an increase in evaporation and snow cover and a decrease in surface wind, other factors such as cloud cover, solar radiation and river runoff exhibit large spatiotemporal heterogeneity that complicates moisture change over the TP (Kang et al. 2010; Yang et al. 2014a; Duan and Xiao 2015). As a result, the implications of rapid high-altitude warming on moisture change over the TP are poorly understood, and one critical reason is the lack of extensive, long-term observations (Qiu 2014). Here we use tree-rings to study common moisture change on the southeastern TP during the past five centuries, and examine whether regional moisture change is related to large-scale TP surface temperature anomaly from a long-term perspective. Tree-rings are employed as a proxy in light of their precise dating, annual resolution, and high sensitivity to climate change in the study area (Fan et al. 2008a; Fang et al. 2010; Liu et al. 2012; Duan and Zhang 2014).

2 Data and methods

2.1 Tree-ring data

We collected tree-ring samples from two sites in the southern Shaluli Mountains, southeastern TP (Fig. 1). The two sites are close to each other, and both are situated on a steep, leeward slope dominated by subalpine old-growth forests of Forest Fir (Abies forrestii). Two increment cores per tree were collected from living trees of A. forrestii at breast height (1.3 m above ground). All sampled trees are healthy and relatively isolated, an optimal condition for maximizing climate signals in tree rings. In total 56 cores from 28 trees and 50 cores from 25 trees were retrieved at the site of MAX and MXG, respectively (Table 1).

Fig. 1
figure 1

Map of the Tibetan Plateau showing the location of the tree-ring sampling sites (triangle), the Daocheng (DC) and Lijiang (LJ) meteorological station (black circle), and the four scPDSI grid points (open circle) used in this study. The tree-ring sites are as follows: Black triangle denotes the two sites of this study (MX). Blue triangles denote the three moisture-sensitive sites [BM (Fan et al. 2008a), LX (Liu et al. 2012), and LZ (He et al. 2012)]. Red triangles denote the three temperature-sensitive sites [ML (Fan et al. 2008b), BD (Duan and Zhang 2014), and QM (Wang et al. 2014b)]

Full size image
Table 1 Statistics of the two tree-ring sampling sites, the nearest meteorological station, and the scPDSI grid points developed by van der Schrier et al. (2013)
Full size table

After being properly mounted and sanded in the laboratory, all samples were measured using a Velmex ring-width measuring system at 0.001 mm precision. Calendar year was assigned to each growth ring by both visual and the COFECHA program assisted statistical cross-dating methodology (Holmes 1983). Eight (three) cores from the MAX (MXG) site were eliminated during this process due to their irregular growth patterns.

The raw ring-width measurements contain non-climatic growth trends that need to be removed for dendroclimatic study, a procedure termed as “tree-ring standardization” (Fritts 1976). We applied an initial power transformation to reduce the heteroscedastic behavior commonly found in tree-rings (Cook and Peters 1997), and then detrended all series conservatively by fitting negative exponential curves or linear regression curves of any slope. Tree-ring indices were calculated as the residuals between the raw measurements and the fitted curve values, which can effectively avoid potential index value inflation associated with the ratio method (Cook and Peters 1997). The resulting index series were merged to develop a biweight robust mean chronology, with its variance stabilized using the Rbar weighted method (Osborn et al. 1997; Frank et al. 2007). Finally, we applied the “signal-free” approach to mitigate potential trend distortion problem in traditionally detrended chronology (Melvin and Briffa 2008). The resulting “signal-free” chronology was used for further analysis.

2.2 Climate data

Monthly temperature and precipitation records, spanning 1957–2013, were obtained from Daocheng (DC), the nearest weather station to our sampling sites (Fig. 2). The half-degree gridded Climatic Research Unit (CRU) TS 3.23 temperature and precipitation datasets (Harris et al. 2014) were used to investigate the spatial relationship of our tree-rings with large-scale climate anomalies. We only used the CRU data starting from 1951, as there were few observations on the TP before the 1950s.

Fig. 2
figure 2

a Monthly mean temperature and b monthly total precipitation records at the Daocheng meteorological station during 1957–2013

Full size image

The self-calibrating Palmer Drought Severity Index (scPDSI, van der Schrier et al. 2013) was used as a drought metric. The PDSI is a metric of meteorological drought (Palmer 1965), and has been proven suitable for describing moisture conditions across China (Li et al. 2009a). The scPDSI is a new variant of the PDSI and is more suitable for regions with diverse climatology (van der Schrier et al. 2013). As the nearest Daocheng climate records were not included in the development of the scPDSI dataset, we averaged four half-degree scPDSI grid points relatively close to our sampling sites to represent regional moisture condition (Fig. 1). The four grids were chosen because of their proximity to both our sampling sites and the Lijiang (LJ) weather station, which has the longest observations in the area (i.e., 1944–2012) and was included in the scPDSI calculation.

The European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis dataset (ERA-Interim, Dee et al. 2011) was used for dynamic analysis. ERA-Interim is a global atmospheric reanalysis product covering the data-rich period from 1979 to the present. ERA-Interim was chosen because of its marked improvements on certain key aspects, such as the representation of the hydrological cycle, the quality of the stratospheric circulation, and the handling of biases and changes in the observing system (Dee et al. 2011). As a result, it performs better than other reanalysis products over the TP (Bao and Zhang 2013).

3 Results

A 498- (1509–2006) and 516-year (1498–2013) chronology was developed for the MAX and MXG site, respectively (Table 1). The two chronologies agree very well with each other, with a correlation of 0.67 (p < 0.001) and an explained variance of 83.6 % by the first principal component during the common period 1509–2006. Further considering the close location of the two sites and their high environmental homogeneity, we merged all the ring-width index series to develop one composite chronology (hereafter, MX) to represent a regional-scale climate signal. This composite chronology spans from 1498 to 2013, and is composed of 95 cores from 47 trees, with a mean segment length of 291 years (Fig. 3). According to the generally accepted expressed population signal (EPS, Wigley et al. 1984) cutoff value of 0.85, the chronology is considered most reliable during 1523–2013 when sample size exceeds five cores from four trees. The running Rbar ranges from 0.45 around the 1970s to 0.74 around the 1520s with a mean value of 0.58 (Fig. 3). These statistics indicate that the chronology contains fairly strong and stable common signals, and is valid for dendroclimatic studies described below.

Fig. 3
figure 3

a The composite chronology developed from two sites of A. forrestii on the southeastern TP. b The running EPS statistics. Dashed line denotes the 0.85 cutoff value. c The running Rbar statistics. Horizontal line denotes the mean value. d The corresponding sample size

Full size image

As shown in Fig. 4a, statistically significant (p < 0.05) positive correlations between tree-rings and precipitation are found in previous August–September and current May–June. Significant positive correlations with temperature are observed from prior October to current April. Negative but non-significant correlations with temperature are found in current May–June. These results indicate a typical moisture stress on tree growth (Li et al. 2007, 2008; Fan et al. 2009; Fang et al. 2015a). We therefore examined the correlations of tree-rings with the scPDSI during their common period 1944–2012. As shown in Fig. 4b, significant positive correlations with the scPDSI are found in all months investigated, with the highest values in late spring to early summer (April–June). This suggests that the early growing season (EGS) moisture is the most critical factor that limits subalpine tree growth on the southeastern TP.

Fig. 4
figure 4

Correlations of tree-rings with a monthly precipitation (solid bars) and temperature (light bars) records from previous June to current September during 1957–2013, and with b monthly scPDSI data during 1944–2012. The dashed lines indicate the corresponding 95 % confidence level

Full size image

The above climate-tree growth relationship indicates that our chronology is most suitable for the reconstruction of April-June moisture change in the study area. We used a simple linear regression model to build the reconstruction, and assessed its fidelity by split sample calibration and verification tests (Cook and Kairiukstis 1990). As shown in Table 2, the actual and reconstructed scPDSI correlate at 0.715 during 1944–2012 (p < 0.001), which means the reconstruction accounts for 51.2 % (R 2adj  = 50.4 %) of the actual scPDSI variance during this period. The values of two most rigorous tests of model validation, the reduction of error (RE) and the coefficient of efficiency (CE), are both positive, indicating a good model skill (Cook and Kairiukstis 1990). The results of the sign test, which describes how well the tree-ring estimates track the direction of actual data year to year, exceed the 99 % confidence level. These statistical tests sufficiently validate our regression model. A visual comparison also suggests the reconstruction tracks well the actual scPDSI values at both high- and low-frequency scales, despite that it tends to overestimate the persistence but slightly underestimate the severity of the pluvial condition during the 2000s (Fig. 5). Based on this model we reconstructed April-June moisture change on the southeastern TP for the past 491 years (Fig. 6a).

Table 2 Statistics of calibration and verification test results
Full size table
Fig. 5
figure 5

Comparison of the actual (solid line) and estimated (dotted line) April–June scPDSI values during their common period 1944–2012

Full size image
Fig. 6
figure 6

Comparison of the EGS scPDSI reconstruction with three tree-ring records that are most sensitive to the EGS moisture condition on the southeastern TP. a The April–June scPDSI reconstruction from this study, b BM (Fan et al. 2008a), c LX (Liu et al. 2012), d LZ (He et al. 2012). Data in (b)–(d) have been normalized for direct comparison. Bold line in each panel denotes a 21-year low-pass filter. Vertical shading denotes wet periods in our reconstruction

Full size image

4 Discussion

Our results show that subalpine tree growth on the southeastern TP is mainly controlled by the EGS moisture availability (Fig. 4). This type of climate-tree growth relationship is commonly found over the eastern TP (Li et al. 2008; Fan et al. 2009; Wang et al. 2012; Fang et al. 2015a). Physiological studies revealed that the EGS moisture to a great extent controls the onset of xylogenesis and xylem cell production, and thus largely determines ring formation of subalpine conifers on the eastern TP (Wang et al. 2012; Ren et al. 2015). Significant positive correlations with precipitation and non-significant negative correlations with temperature in May and June suggest that xylem growth is primarily controlled by precipitation rather than temperature at our sampling sites (Ren et al. 2015). However, at sites where precipitation is more abundant, temperature could be the most critical limiting factor on subalpine tree growth on the southeastern TP (Liang et al. 2010; Yang et al. 2010; Liu et al. 2016). Under that situation, low air and soil temperature may limit tree growth by causing direct leaf and root damage and/or by reducing photosynthetic rate and cambial activity (DeLucia 1986; Gruber et al. 2009; Liang et al. 2009, 2010). Therefore, we caution that moisture is not necessarily the most critical factor limiting subalpine tree growth across the southeastern TP. Temperature may become most critical when moisture is sufficient for tree growth, and the threshold for such a transition requires future investigation.

Our EGS moisture reconstruction covers the period of 1523–2013 (Fig. 6a). Due to the “segment length curse” (Cook et al. 1995), our reconstruction is capable of resolving interannual to interdecadal moisture variations, but may not be able to represent the centennial-scale variability very well. We therefore focus our discussion on sub-centennial scale moisture variability. As shown in Fig. 6a, our reconstruction reveals marked interdecadal variations in regional EGS moisture over the past five centuries. Severe dry conditions occurred during the 1630s–1640s, 1670s–1690s, 1730s–1770s, 1790s–1820s, 1860s–1880s, 1910s–1930s, and 1950s–1980s, and pronounced wet conditions were observed during the 1520s–1590s, 1610s–1620s, 1700s–1720s, 1830s–1850s, 1890s–1900s, and 1990s–2000s. The most severe and prolonged drought occurred in the 1730s–1770s. The most recent pluvial during the 1990s–2000s was likely the wettest for the past five centuries, although its duration was exceeded by the generally wet conditions during the sixteenth century. It is worth noting that tree-rings overestimated the persistence but slightly underestimated the severity of this pluvial (Fig. 5). Nonetheless, the 1990s–2000s pluvial is probably unprecedented at least for the past five centuries, as revealed by this and other moisture sensitive tree-rings on the southeastern TP (Fig. 6).

Spatial correlation analysis with instrumental scPDSI during 1951–2012 indicates that our reconstruction is representative of large-scale EGS moisture change on the southeastern TP (Fig. 7a, b). To examine whether it represents large-scale moisture change back in time, we compared our reconstruction with three tree-ring records (BM, LX, and LZ, Fig. 1) that are also most sensitive to the EGS moisture condition on the southeastern TP (Fan et al. 2008a; He et al. 2012; Liu et al. 2012). As shown in Fig. 6, our record agrees well with the other three over most of the past five centuries, with a significant correlation value (p < 0.001) of 0.30 with the BM for 1655–2005 (351 years), 0.25 with the LX for 1523–2010 (488 years), and 0.27 with the LZ record for 1523–2009 (487 years). However, one discrepancy is observed during the sixteenth century when our record indicates a generally wet while the LX record shows a dry condition. We found that our record also shows generally low values if without the “signal-free” adjustment, suggesting that the generally dry condition with the LX record is likely due to the trend distortion introduced by the traditional detrending method (Melvin and Briffa 2008). At any rate, these records exhibit a high degree of coherency with regard to interdecadal variations, indicating common EGS moisture change on the southeastern TP over the past five centuries.

Fig. 7
figure 7

Spatial correlation patterns for the period of 1951–2012. a Actual and b reconstructed April–June scPDSI correlated with regional gridded scPDSI. Reconstructed April–June scPDSI correlated with the CRU minimum temperature in c prior winter (October–February) and d current EGS (April–June). The correlation coefficient at the 0.05 significance level is about 0.25, based on a two-tailed Student’s t test. The box in (d) denotes the region over which the temperature is averaged

Full size image

An ensuing question is what caused the coherent EGS moisture change on the southeastern TP. One possibility is the Asian monsoon. However, the EGS is largely ahead of monsoon season (Fig. 2), thus the Asian summer monsoon is unlikely to play a key role. This is corroborated by the non-significant correlations of the actual April-June scPDSI with the East Asian (Li and Zeng 2002) and South Asian (Wang et al. 2001) summer monsoon indices (Fig. S1). Moreover, both monsoon systems have weakened during recent decades (Yu et al. 2004; Li et al. 2009b; Turner and Annamalai 2012), which is in contrast to the EGS moisture increase on the southeastern TP. Another possibility is the large-scale ocean-atmospheric circulations. However, as shown in Fig. S2, the EGS moisture change on the southeastern TP shows no significant correlation pattern with the precedent or concurrent tropical sea surface temperatures (Rayner et al. 2003), supporting the notion that large-scale ocean-atmospheric circulations do not play a key role on the wetting trend on the TP (Fang et al. 2015b). The third possibility is snow cover on the TP (Estilow et al. 2015). However, the actual April–June scPDSI shows no significant correlation pattern with the precedent winter snow cover on the TP (Fig. S3a). Although it shows significant positive correlations with concurrent snow cover in the study area (Fig. S3b), the covariability more likely suggests a response of snow cover to the EGS moisture availability. The fourth possibility is precipitation on the TP. Similar to snow cover, the actual April–June scPDSI shows no significant correlation pattern with the precedent winter precipitation on the TP (Fig. S3c), suggesting that the latter is not a critical factor that affects the EGS moisture. Instead, it shows significant positive correlations with concurrent precipitation in the study area (Fig. S3d), indicating the EGS moisture is largely determined by precipitation in the same season.

Our moisture reconstruction shows strong and positive correlations with large-scale TP surface temperature anomaly in prior winter (October–February) and current EGS (April–June). The strong and positive correlations with prior winter minimum temperature (Tmin) are concentrated on the southeastern TP (Fig. 7c), while the correlations with the EGS Tmin are centered over the interior of the TP (Fig. 7d). Similar but weaker correlation patterns are found for the maximum temperature (Tmax) in both seasons (Fig. S4). The seasonal shift in spatial correlation pattern suggests that temperature of different seasons affects the EGS moisture through different processes. The strong and positive correlations of the EGS moisture with prior winter temperature are found within the study area (Fig. 7c). The atmosphere has a relatively short memory where the climate signals in winter may not be able to exert a time-lagged effect on the warm season moisture, and instead soil moisture is more likely the medium for such a long climate memory (Barnett et al. 1989; Hsu and Liu 2003; Chow et al. 2008). Indeed, we found that the EGS moisture shows persistently high correlations with prior winter scPDSI at our sampling sites (Fig. 4b), consistent with previous studies at other moisture-stressed sites on the southeastern TP (Fan et al. 2008a; Fang et al. 2010; He et al. 2012). These results suggest that prior winter temperature affects the EGS moisture availability by modulating water storage in the soil. In winter, frozen ground prevents infiltration of snowmelt or rainfall into the soil, leading to higher-than-normal springtime runoff (Niu and Yang 2006). High winter temperature causes thawing of ground and slow melting of snowpack, which result in more infiltration of water into deep soil. Meanwhile, high temperature means more winter precipitation falls as rain instead of snow (Barnett et al. 2005), a change that facilitates winter soil water infiltration. These processes under high winter temperature help retain more water in the local system, which will otherwise be likely lost as surface runoff and river flow during the rapid snow melting in late spring to early summer. The above notion is supported by the observed increase in wintertime low-level clouds at both daytime and nighttime on the TP during recent decades (Duan and Xiao 2015), which is a result of increased surface warming, snowpack melting and evaporation. Overall the increase in wintertime low-level clouds is more pronounced at nighttime than at daytime (Duan and Xiao 2015), supporting our finding that the Tmin is more strongly correlated to the EGS moisture change on the southeastern TP.

The EGS moisture is not strongly related to concurrent Tmin in the study area (Fig. 7d), and its correlation with concurrent Tmax is even negative (Fig. S4b). These results suggest that high EGS temperature leads to regional moisture loss by enhancing evapotranspiration (Fang et al. 2015a; Ren et al. 2015). In contrast, our record shows strong and positive correlations with concurrent surface temperature anomaly over the interior of the TP (Fig. 7d), suggesting that our study area gains moisture when anomalous warming occurs over the interior of the TP. Regression analysis using the ERA-Interim reanalysis data was performed in order to understand the dynamic process. As shown in Fig. 8a, corresponding to positive TP surface temperature anomalies in April–June, positive 200 hPa geopotential anomalies are found over the TP and surroundings, with the center above the interior of the TP with an extension to northwest China. The appearance of strong upper-level anti-cyclone indicates a large-scale upward convection in the region as a response to anomalous surface warming on the TP. The convection leads to an increase in lower tropospheric humidity over the southeastern TP, as represented by positive anomalies of 700 hPa specific humidity (Fig. 8b). In contrast, the convection does not induce more atmospheric humidity over the interior of the TP, largely because the underlying surface is characterized by gobi deserts with limited moisture supply. As a result, strong convection plus increased lower tropospheric humidity lead to an increase in precipitation and effective precipitation [precipitation–evaporation (P–E)] in the southeastern TP, whereas a strong convection plus less lower tropospheric humidity result in a decrease in precipitation and effective precipitation in the interior and western TP (Fig. 8c, d).

Fig. 8
figure 8

Spatial regression patterns for the period of 1979–2014. Regression patterns of a 200 hPa geopotential height (m2/s2), b 700 hPa specific humidity (g/kg), c precipitation (mm/day), and d effective precipitation (P–E, mm/day) with the interior TP surface temperature in April–June. The interior TP surface temperature was averaged over a region as denoted in Fig. 7d, using the gridded CRU dataset

Full size image

Previous studies found that high temperature leads to strong surface and soil water evaporation that favors the formation of convective precipitation, a crucial process for water supply on the TP before the arrival of monsoon rainfall (Yanai and Li 1994; Lau et al. 2010). This warm-wet relationship has been found in many regions of the TP (Li et al. 2010, 2014; Yang et al. 2014b). Therefore, high EGS surface temperature on the TP benefits moisture supply in its southeastern region through enhancing large-scale evaporation and convective precipitation.

To validate whether the warm-wet relationship has persisted at a long-term scale, we compared our EGS moisture reconstruction with three tree-ring records (ML, BD, and QM, Fig. 1) that represent large-scale temperature change on the TP (Fan et al. 2008b; Duan and Zhang 2014; Wang et al. 2014b). Admittedly only the ML record is from the core area of high correlations shown in Fig. 7c, d. However, it shows very coherent relationship with the other two temperature records (Fig. 9), indicating temperature change is highly uniform on the TP. As shown in Fig. 9, the moisture and temperature reconstructions exhibit a high degree of coherency with regard to their interdecadal variations, with increased moisture coincident with periods of high temperature, and vice versa for the dry and cool periods. In particular, the wettest pluvial of the past five centuries occurred during the 1990s–2000s, which is also the warmest period on the TP during the past millennium (Wang et al. 2014b). The coincidence of the 1990s–2000s warm and pluvial conditions on the southeastern TP may not be enhanced by any persistent trend, as both temperature and moisture records exhibit strong interdecadal variations during the twentieth century (Fig. 9), which is in contrast to the persistent warming and wetting trend on the northeastern TP (Yang et al. 2014b). Therefore, the warm-wet association on the southeastern TP has persisted at least for the past five centuries.

Fig. 9
figure 9

Comparison of the EGS scPDSI reconstruction with three temperature-sensitive tree-ring records on the TP. a The April–June scPDSI reconstruction from this study, b ML (Fan et al. 2008b), c BD (Duan and Zhang, 2014), d QM (Wang et al. 2014b). Data in (b)–(d) have been normalized for direct comparison. Bold line in each panel denotes a 21-year low-pass filter. Vertical shading denotes wet periods in our reconstruction

Full size image

Our chronology contains prior winter temperature signal (Fig. 4a), which to some extent complicates the interpretation of a warm-wet relationship between the TP surface temperature and the EGS moisture change in its southeastern region. However, two moisture-sensitive chronologies used in the study contain very weak or no prior winter temperature signal (Fan et al. 2008a; Liu et al. 2012) and that exhibit coherent variations with our chronology (Fig. 6), proving that the warm-wet relationship is not due to the inclusion of winter temperature signal in our chronology. At any rate, future sampling of pure moisture-sensitive chronologies on the southeastern TP is needed in order to validate our conclusion. Moreover, the warm-wet relationship breaks down in a few short periods such as the late 1950s to the early 1960s (Fig. 9). Other factors that may affect the EGS moisture change on the southeastern TP await future investigation.

5 Conclusions

We developed a 491-year EGS moisture reconstruction with tree-rings, by far the longest for the southeastern TP. Our and other reconstructions together reveal common EGS moisture change on the southeastern TP, and provide a long-term context for evaluating their relationship with large-scale climate anomaly. Our study indicates a coherent relationship between large-scale TP surface temperature and the EGS moisture change in its southeastern region. High TP surface temperature may affect the EGS moisture supply through the modulation of winter soil water storage and the enhancement of regional EGS evaporation and convective precipitation. State-of-the-art climate models projected that rapid temperature increase on the TP will persist throughout the twenty-first century as a result of continuing anthropogenic greenhouse forcing. The coherent warm-wet association identified in the study implies a generally wetter condition on the southeastern TP under future warming.