Synergy of the westerly winds and monsoons in the lake evolution of global closed basins since the Last Glacial Maximum and implications for hydrological change in central Asia

The monsoon system and westerly circulation, to which climate change responds differently, are two important components of global atmospheric circulation interacting with each other in the middle to low latitudes. Relevant research on global millennial-scale climate change in monsoon and westerly regions is mostly devoted to multiproxy analyses of lakes, stalagmites, ice cores, and marine and eolian sediments. Different responses from these proxies to long-term environmental change make understanding climate change patterns in monsoon and westerly regions difficult. Accordingly, we disaggregated global closed basins into areas governed by monsoon and westerly winds, unified paleoclimate indicators, and added lake models and paleoclimate simulations to emphatically track millennial-scale evolution characteristics and mechanisms of East Asian summer monsoon and westerly winds since the Last Glacial Maximum (LGM). Our results reveal that millennial-scale water balance change exhibits an obvious boundary between global monsoon and westerly regions in closed basins, particularly in the Northern Hemisphere. The effective moisture in most closed basins of the midlatitude Northern Hemisphere mainly exhibits a decreasing trend since the LGM, while that of the low latitudes shows an increasing trend. In the monsoon-dominated closed basins of Asia, a humid climate prevails in the early to mid-Holocene, and a relatively dry climate appears in the LGM and late Holocene. In the westerlywind-dominated closed basins of Asia, the climate is characterized by a humid LGM and mid-Holocene (MH) compared with the dry early and late Holocene, which is likely to be connected to precipitation brought by the westerly circulation. This study provides insight into the long-term evolution and synergy of westerly winds and monsoon systems as well as a basis for the projection of future hydrological balance.


Introduction
As important components of atmospheric circulation systems, the midlatitude westerly winds and low-latitude monsoon systems play key roles in global climate change. Whether on the decadal or the millennial scale, research on this aspect always attracts widespread attention from scientists. Examination of global monsoon precipitation changes on land suggests an overall weakening over the recent halfcentury   (Zhou et al., 2008). Individual monsoon indexes reconstructed by Wang et al. (2017) indicate that moisture in the tropical Australian, East African, and Indian monsoon regions has exhibited a gradual decrease since the early Holocene. It is widely accepted that the East Asian summer monsoon usually follows the variation of lowlatitude summer solar radiation Chen et al., 2006;An et al., 2015). Charney (1969) and Wang (2009) also proposed that the seasonal migration of the intertropical convergence zone (ITCZ) profoundly influences the seasonality of the global monsoons. However, the global westerly winds and their associated storm tracks dominate the midlatitude dynamics of the global atmosphere and affect extratropical large-scale temperature and precipitation patterns (Oster et al., 2015;Voigt et al., 2015). Since the Last Glacial Maximum (LGM), the climate in central and southern regions of the North American continent has gradually dried out as the ice sheet melted and the westerlies moved north (Qin et al., 1997). As mentioned in previous studies, millennial-scale evolution in global monsoons and westerly winds probably shows different patterns as a result of complex driving mechanisms. Arguments about an asynchronous pattern of moisture variations between monsoon and westerly wind evolution underscore the importance of studying their millennial-scale differentiation An and Chen, 2009;Li et al., 2011;An et al., 2012).
A way to examine past climate variability is traditional methods of studying various archives that truly document the evolution of regional climate, including lake sediments , stalagmites (Dykoski et al., 2005;Wang et al., 2008), and tree rings (Linderholm and Braeuning, 2006). However, due to the limited timescale of paleoclimate records, most research on the evolution of monsoons and westerly winds is concentrated in the Holocene and lack an exploration during the LGM. With the development of paleoclimatology in recent decades, numerical simulations of paleoclimate continue to emerge and develop to a relatively mature system, which provides a useful tool for reviewing paleoclimate change over long timescales. On account of the water balance system constantly responding to climatic condition changes, a combination of numerical simulations and lake water balance models can be used to effectively track past climate change and make up the deficiency in qualitative methods of multi-proxy analysis (Qin and Yu, 1998;Xue and Yu, 2000;Morrill et al., 2001Morrill et al., , 2004Li andMorrill, 2010, 2013;Lowry and Morrill, 2019;Li et al., 2020). Covering one-fifth of the terrestrial surface, global closed basins are distributed in both low-latitude monsoon regions and midlatitude westerlies. Furthermore, closed basins with a relatively independent hydrological cycle system have plenty of terminal lake records that provide more evidence for retrospecting climate change (Li et al., 2017), and these can be regarded as ideal regions to study the spatiotemporal difference between monsoons and westerly winds.
By constructing virtual lake systems, here we applied lake models and a transient climate evolution model to continuously simulate water balance change since the LGM in global closed basins. Meanwhile, precipitation minus evaporation (P − E) simulations and 37 lake status records in the LGM, mid-Holocene (MH), and preindustrial (PI) were supplemented to validate results of the continuous simulations. The prominent spatial differentiation of monsoons and westerly winds revealed by simulations leads us to focus on the Northern Hemisphere midlatitude closed basins, which are simultaneously influenced by midlatitude westerly winds and low-latitude monsoons. In the midlatitude closed basins of the Northern Hemisphere, the good match between a water balance simulation and a reconstructed moisture index from 27 paleoclimate records verifies the reliability of the simulation results. Further, we disaggregated the Northern Hemisphere midlatitude closed basins into areas dominated by monsoons and westerly winds and emphatically explored the temporal evolution of the East Asian summer monsoon and westerly winds since the LGM. According to the climate records, we comprehensively considered the determinants that have controlled the trend of climate change in the Northern Hemisphere westerlies and East Asian summer monsoon regions since the LGM. This study not only reveals millennial-scale climate change from the perspective of water balance, but also provides a new method for studying the synergy of the westerly winds and monsoons.  (He, 2011). We applied this model to continuously simulate effective moisture change represented by virtual water balance variation since the LGM. Likewise, the CCSM4, CNRM-CM5, FGOALS-g2, GISS-E2-R, MIROC-ESM, MPI-ESM-P, and MRI-CGCM3 models participating in CMIP5/PMIP3 were also used to simulate the relative change in P − E during three particular periods (LGM, MH, PI). Here the PI period, which is considered a typical period of the late Holocene, is mainly used to measure the changes in hydroclimate conditions during the LGM and MH periods relative to the late Holocene, as well as verify the feasibility of the lake models by comparing the lake-level simulations with the lake status records among the three periods. PMIP3 protocols define the boundary conditions of these models, with a few exceptions (Table 1). Precession, obliquity, and eccentricity values are specified according to Berger (1978). CO 2 , CH 4 , and N 2 O values are set on the basis of reconstructions from ice cores (Monnin et al., 2004;Flückiger et al., 1999Flückiger et al., , 2002. A remnant Laurentide ice sheet in the LGM and a modernday ice sheet configuration in the MH and PI simulations are specified by the ICE-5G reconstruction (Peltier, 2004), while the vegetation is prescribed to modern values. Ice sheet configuration and vegetation distribution are used by the GISS model.
LGM radiative forcing changes in the MIROC model and MRI model are the exceptions of the PMIP3 boundary conditions; details are shown in Licciardi et al. (1998) and Lowry and Morrill (2019). Before calculating, we linearly interpolated differentresolution grid cells of the TraCE model and multi-model ensemble into a uniform resolution of 0.5 • × 0.5 • . For all grid cells in closed basins, we assumed that the virtual lake in each grid cell is a 1 m deep lake with fresh water, and then the virtual lake evaporation is calculated by a lake energy balance model that is modified according to the Hostetler and Bartlein (1990) model. The evaporation of lake surface depends on the heat capacity of water, water density, lake depth, lake surface temperature, shortwave radiation, longwave radiation absorbed by the water surface, longwave radiation emitted by the water surface, latent heat flux, and sensible heat flux. If the surface energy balance is negative (positive), ice forms (melts). Lake depth and lake salinity are important input parameters influencing lake surface evaporation (Dickinson et al., 1965); however, only small changes appear in lake evaporation when adding lake depth to 5 and 10 m and increasing lake salinity to 10 ppt. More details on the lake energy balance model are described in Morrill (2004) and Li and Morrill (2010).
To better assess the relative change in water balance since the LGM, the virtual lakes are assumed to be in hydrological equilibrium at steady state. The lake water balance equation is shown as follows: where D is discharge from the lake (m 3 yr −1 ), A B is the area of the drainage basin (m 2 ), R is runoff from the drainage basin (m yr −1 ), A L is the area of the lake (m 2 ), P L is precipitation over the lake (m yr −1 ), and E L is lake evaporation (m yr −1 ). Given the application of Eq. (1), requiring specific values of A B and A L , this equation is simplified for grid cells wherein P L − E L ≥ 0 and grid cells wherein P L − E L < 0. Grid cells wherein P L − E L ≥ 0 represent open lakes and maintain water balance by discharging more or less water. The runoff into the lake compensates for the net water loss in grid cells wherein P L − E L < 0, and these regions maintain water balance by changes in the ratio of A L to A B , as described by setting D = 0 in Eq. (2): where A L /A B represents the virtual lake level. Accordingly, for grid cells with P L − E L < 0, the A L /A B values are calculated and compared to represent relative water balance change, and more details about the lake water balance model are described in Li and Morrill (2010). We combined the values of P L , E L , and R with Eqs. (1) and (2) and simulated the continuous water balance change since the LGM using the TraCE 21 kyr model.

Record selection and moisture index inference
Status information for 37 lakes in or near global closed basins was collected to compare relative changes among the three characteristic periods. Lake status information sorted by latitude is shown in Table 2. Then, 27 climate records were compiled in or near the midlatitude closed basins of the Northern Hemisphere with reliable chronologies and successive sedimentary sequences from published literature, which can reflect the continuous dry and wet change (Table 3). We interpolated climate data at intervals of 10 years and unified the timescale according to the chronological accuracy of the extracted data. Finally, the data were standardized to indicate a humid climate with a relatively high value and a dry climate with a relatively low value, and the signals of moisture change were transformed into an index range of 0 to 1. Due to the different timescales of the collected continuous paleoclimate records, we can only reconstruct the regional moisture change from the early to late Holocene after unifying the timescales, but the purpose of this part is only to check the simulation results.

Mathematical methods
Linear tendency estimation is a common trend analysis method, which was chosen to measure the variation degree of simulated water balance in this paper. We also used the empirical orthogonal function (EOF), a method of analyzing the structural features in matrix data and extracting the feature  Lake elevation Goldsmith et al. (2016) vector of main data to examine the spatial and temporal variability of simulated water balance. The spatial distribution of the EOF first (second) mode is denoted by EOF1 (EOF2), and the time series of the first (second) mode is denoted by PCA1 (PCA2).

Results and discussion
3.1 Observed and simulated water balance change in global closed basins As Fig. 1 shows, we intercepted LGM (18 000-22 000 yr), MH (5000-7000 yr), and PI (1800-1900 AD) periods from the TraCE 21 kyr dataset to better match the multi-model ensemble. Because runoff anomalies are highly correlated with precipitation anomalies, it is therefore feasible to consider the contribution of runoff to the water balance to be the contribution of precipitation to the water balance. Differences between the time period we chose subjectively and the time periods defined by the multi-model ensemble may affect the comparison results. However, the precipitation and evaporation difference of TraCE 21 kyr among the three periods exhibits a similar spatial pattern as the P − E difference of the multi-model ensemble. The simulations and lake status records of the midlatitude westerlies (low-latitude monsoon regions) show that the LGM is humid (dry) relative to MH and PI, which generally corresponds to the hydroclimate patterns of previous research (Street and Grove, 1979;Qin et al., 1997;Quade and Broecker, 2009;Lowry and Morrill, 2019). It is not our intent to simulate relative lake status change among the three periods, but to validate continuous water balance simulations and to track continuous water balance fluctuations on the millennial scale using TraCE 21 kyr simulations.
In continuous simulations, we partitioned the trend map of water balance into positive and negative components to highlight the spatial patterns of water balance change (Fig. 2). In the global midlatitude westerlies, simulations indicate that widespread effective moisture has declined since the LGM except in the northern Caspian Sea, whereas effective moisture has increased since the LGM over the global tropics. Meanwhile, the trend map exactly exhibits the spatial differentiation of the millennial-scale water balance change between the global low-latitude monsoon-dominated regions and the midlatitude westerly-wind-dominated regions. This differentiation provides the basis to explore the continuous evolution of monsoons and westerly winds in closed basins since the LGM.

Possible driving mechanisms of millennial-scale water balance change
In this section, the possible driving mechanism that affects the millennial-scale water balance change in global closed basins is explored. Positive signs of EOF1 represent most monsoon regions in the midlatitudes and low latitudes, while negative signs are mainly located in the Northern and Southern Hemisphere westerlies. Spatial characteristics of EOF2 have an opposite trend to EOF1, except for the Caspian Sea. The contribution rate of PCA1 and PCA2 is 51 % and 14 %, respectively; therefore, the following discussion mainly focuses on PCA1 with a high contribution rate (Fig. 3). PCA1 extracted from the water balance simulation tends to represent the effective moisture fluctuation of closed basins in low-latitude monsoon regions, indicating a relatively humid climate during the early to mid-Holocene. By comprehensively analyzing a variety of paleoclimate proxies, Wang et al. (2017) suggested that moisture change revealed by the Australian monsoon, the East African monsoon, and the Indian monsoon regions reaches the wettest status in the early Holocene, while the wettest condition in the East Asian summer monsoon regions occurs between 8 and 6 ka. Likewise, Qin et al. (1997) presented the wettest period in the African and South Asian monsoon regions as the early to mid-Holocene, coinciding well with our results. The climatic significance of δ 18 O in the Asian speleothem records is always a long-standing debate, and some influential hypotheses regard δ 18 O of the monsoon regions as a proxy for "Asian monsoon intensity", "Indian monsoon intensity", "summer monsoon rainfall amount", and "circulation conditions" Chen et al., 2016). Although the climatic significance is controversial, it is well accepted that δ 18 O changes should bear the imprint of variations in the oxygen isotopic composition of precipitation Chen et al., 2016). According to the close similarity of PCA1 to the speleothem records from Dongge and Hulu caves, our simulations are more inclined to suggest that the δ 18 O stalagmite records indicate a change in water vapor brought by the monsoons. In addition, we not only compared PCA1 to the stalagmite records of Dongge Cave, with controversial climatic significance, but also with the summer solar radiation at low latitudes in the Northern Hemisphere. This comparison provides evidence for the view that the evolution of low-latitude monsoons is generally controlled by summer insolation in the Northern Hemisphere Chen et al., 2006;An et al., 2015). Thus, we further speculated that the water balance change in monsoon regions of global closed basins is mainly driven by midlatitude and low-latitude summer solar radiation.

Evolutionary characteristics and causal factors of millennial-scale hydroclimate change in Northern Hemisphere midlatitude closed basins
On the basis of the spatial characteristics of the EOF analysis, closed basins in the Northern Hemisphere, affected both by low-latitude monsoons and midlatitude westerly winds, are ideal regions for revealing the synergy of the westerly winds and monsoons. Between 30 • and 60 • N, 27 paleoclimate records indicating dry or wet climate were collected from the Northern Hemisphere midlatitude closed basins. As described in Sect. 2.2, we reconstructed a moisture index from the early to late Holocene around that region (Fig. 4). The simulated mean water balance curve corresponds well to the mean moisture index in the Northern Hemisphere midlatitude closed basins, indicating a transition from a humid climate in the early to mid-Holocene to an arid climate in the late Holocene. Therefore, continuous simulations, well validated by paleoclimate indicators, could be better used to track climate change during the LGM. Water balance simulations since the LGM show that a humid climate not only appears in the early to mid-Holocene but also occurs during the LGM, while the climate is relatively dry in the late Holocene. The maintained high moisture in the LGM is possibly influenced by low evaporation and high precipitation (Fig. 5). Using paleoclimate modeling, Yu et al. (2000) mentioned that the low temperature during the glacial period causes a decrease in evaporation and a reduction of lake water loss, resulting in the appearance of a high lake level. Afterward, solar radiation, atmosphere radiation, temperature, evaporation, and precipitation simulations gradually increase (Fig. 5). When entering the warm Holocene, precipitation continues to increase and reaches a maximum in the MH, while solar radiation, atmospheric radiation, and evaporation decrease during the early to mid-Holocene and then increase around the late Holocene. Low (high) evaporation and high (low) precipitation are responsible for the MH (late Holocene) relatively humid (dry) climate (Fig. 5).

Evolutionary differentiation of millennial-scale monsoons and westerly winds in Asian closed basins
Spatial distributions of EOF1 and EOF2 clearly exhibit a prominent boundary in the interactional zones between the East Asian summer monsoon and westerly winds in Asia.
Since the boundary of the monsoon will be adjusted accordingly with the change in the East Asian summer monsoon strength, the evolution of Asian lakes on the millennial scale probably does not follow a single climate changing pattern    (Table 3).  (Wu and Guo, 2000;Editorial Committee of China's Physical Geography, 1984;An et al., 2012). The regions dominated by the East Asian summer monsoon and westerly winds were then selected based on the spatial characteristics of the two modes extracted from the EOF to explore millennialscale evolution features of the two climate systems (Fig. 6).
In westerly-wind-dominated regions, the LGM and MH are characterized by a humid climate, and a relatively dry climate prevails in the early and late Holocene, whereas the water balance in monsoon-dominated regions is generally affected by the East Asian summer monsoon, which brought a lot of water vapor over the early to mid-Holocene and led to a relatively dry climate in the LGM and late Holocene. Li (1990) first proposed the "monsoon" and "westerly" modes on the millennial scale since the late Pleistocene in northwest China; since then different climate changing patterns between arid central Asia and monsoonal Asia have been demonstrated by numerous paleoclimate records An and Chen, 2009;Li et al., 2011;Chen et al., 2019). The viewpoint that millennial-scale East Asian summer monsoon change is possibly driven by summer insolation change in low latitudes is the most widely accepted Dykoski et al., 2005;Hu et al., 2008;Fleitmann et al., 2003); the sea surface temperatures (SSTs) of the North Atlantic and air temperatures at high latitudes are responsible for the Holocene effective moisture evolution in arid central Asia, which is dominated by the westerly winds . Moisture transport in arid central Asia mainly comes from the Northern Hemisphere westerlies, the moisture source of which derives from the Black Sea, the Mediterranean Sea, the Arctic Ocean, and the Atlantic Ocean. Winter precipitation accounts for a large proportion of annual precipitation in these regions . The water balance change in the Asian monsoon regions we extracted largely represents the hydroclimate variation in regions dominated by the East Asian summer monsoon since the LGM, while the water balance change in the westerly regions in central Asia can represent the hydroclimate variation in the entire Northern Hemisphere westerlies. Qin et al. (1997) made a large-scale spatial analysis and presented lake levels in south-central North America changing from high to low since the LGM, reaching the lowest levels in the early to mid-Holocene. The LGM proxies indicate that southwestern America experienced a climate that was wetter than present, and the Pacific Northwest through the Rockies experienced a climate that was drier than present, as well as a transition from wetter to drier conditions along a northwest-southeast-trending band across the northern Great Basin (Oster et al., 2015). Our results generally reflect the climate of the westerlies being relatively wet at the LGM and relatively dry at the MH. For the Asian tropics in the Northern Hemisphere, the increased summer solar radiation from the past 12 000 to the past 6000 years ago induced the enhancement of a thermal contrast between the land and sea and further caused the strengthening of summer monsoons so that more water vapor was brought (COHMAP Members, 1988). Collected records in the Northern Hemisphere indicate the evolution of westerly winds and monsoon systems (Fig. 7). Speleothem records from central and southern China confirm that the periods of weak East Asian summer monsoons coincide with cold periods in the North Atlantic , Dykoski et al., 2005Wang et al., 2008). The longest and highest-resolution drill core from Lake Qinghai  indicates that the summer monsoon record generally resembles the changing trends of Asian summer monsoon records derived from Dongge and Hulu speleothems over the last 20 kyr, and the midlatitude westerly climate dominates the Lake Qinghai area in glacial times. Low-latitude summer insolation is broadly recognized as a major control on low-latitude monsoon systems; as a result, the tropical monsoons are weak during the LGM and late Holocene, and strong monsoons prevail in the early to mid-Holocene (Fig. 7). Accordingly, the intensity of monsoon systems and westerly winds varies in different periods so that the main control system in the interactional regions  mussen et al., 2006). (b) Lake Qinghai Asian summer monsoon index . (c) Dongge and Hulu Cave speleothem δ 18 O records (Dykoski et al., 2005;Wang et al., 2008). (d) Moisture indexes in the East Asian monsoon (red line), East African monsoon (green line), Indian monsoon (blue line), and Australian monsoon (orange line) regions . (e) The average moisture index for the arid central Asian region as a whole during the Holocene ). (f) Lake Qinghai westerly climate index ; panels (g, h) show winter 50 • N insolation and summer 50 • N insolation, respectively (Berger, 1978). depends largely on which system is much stronger during that period.
The Northern Hemisphere westerlies shifting northward or southward has a significant impact on global atmosphere circulation and inevitably affects the monsoon systems. Quaternary ice sheets of the Northern Hemisphere in the LGM developed to their maximum extension, and the consequent existence of a persistent strong glacial anticyclone led to the southward displacement of the westerly winds . Most research suggests that the Northern Hemisphere westerlies in the LGM moved to the southwest of the United States and the eastern Mediterranean region (Lachniet et al., 2014;Rambeau, 2010). Therefore, the narrowed temperature difference between the sea and land caused the East Asian summer monsoon to weaken and may have further induced strong westerly winds throughout the year and then precipitation increases . Furthermore, a growing body of evidence shows that the position and orientation of the westerly jet (WJ) probably controlled the Holocene East Asian summer rainfall patterns. A link between the northward seasonal progression of the WJ and the spatial pattern of East Asian summer monsoon precipitation shows that earlier northward progression of the WJ caused abundant precipitation at high latitudes and less precipitation at low latitudes (Nagashima et al., 2013). The northward evolution of the WJ from south of the Tibetan Plateau and seasonal transition exerted especially strong influences on East Asian paleoclimate change (Chiang et al., 2015). Herzschuh et al. (2019) proposed that the position of the summer monsoon rain band changed as the WJ axis shifted gradually southward, leading to the occurrence of a spatiotemporal difference in Holocene China's maximum precipitation. In summary, the above views emphasize that the complex interaction between the monsoon and the westerly systems on the millennial scale should receive more attention.

Conclusions
On the basis of 37 lake status records near global closed basins and 27 paleoclimatic records near midlatitude closed basins of the Northern Hemisphere, we applied a lake energy balance model, a lake water balance model, and paleoclimate simulations to explore the millennial-scale differentiation between global monsoons and westerly winds. A water balance simulation shows that the effective moisture in most closed basins of the Northern Hemisphere midlatitudes has gradually decreased since the LGM, which matches the reconstructed moisture index well. Effective moisture change in most closed basins of the low latitudes (monsoon regions) presents an opposite changing trend to that in the midlatitudes. In Asian midlatitude closed basins, climate change in regions dominated by westerly winds exhibits a relatively humid climate in the LGM and MH and a relatively dry climate in the early and late Holocene, whereas the East Asian summer monsoon generally influences climate change in closed basins dominated by monsoons, which brought more water vapor over the early to mid-Holocene but less water vapor in the LGM and late Holocene. Data availability. The TraCE 21 kyr dataset comes from the Climate Data Gateway National Center for Atmospheric Research (NCAR) website at https://www.earthsystemgrid.org/project/trace. html (last access: 6 November 2020, NCAR, 2020). PMIP3/CMIP5 simulations are available from the Earth System Grid Federation (ESGF) Peer-to-Peer (P2P) enterprise system website at https: //esgf-node.llnl.gov/projects/esgf-llnl/ (last access: 6 November 2020, Lawrence Livermore National Laboratory, 2020). Global closed basin boundaries are derived from hydrological data and maps based on the SHuttle Elevation Derivatives at multiple Scales (HydroSHEDS) website: https://www.hydrosheds.org/page/ hydrobasins (last access: 6 November 2020; WWF, 2020).