Model-based orbital-scale precipitation δ18O variations and distinct mechanisms in Asian monsoon and arid regions

Abstract The past Asian precipitation δ18O (δ18Op) records from stalagmites and other deposits have shown significant orbital-scale variations, but their climatic implications and regional differences are still not fully understood. This study, as the first attempt of a 300-kyr transient stable isotope-enabled simulation, investigated the characteristics and mechanisms of the orbital-scale δ18Op variations in three representative regions of Asia: arid Central Asia (CA), monsoonal South Asia (SA) and monsoonal East Asia (EA). The modelling results showed that the variations in the CA, SA and EA annual δ18Op exhibited significant but asynchronous 23-kyr precession cycles. Further analyses revealed that although the precession-induced insolation variation was the ultimate cause of the δ18Op variation in all three regions, the dominant mechanisms and the involved physical processes were distinct among them. For the CA region, the rainy-season (November–March) temperature effect and water vapour transport by the westerly circulation were identified as the key precession-scale processes linking the October–February boreal mid-latitude insolation to the rainy-season or annual δ18Op. In the SA region, the rainy-season (June–September) precipitation amount effect and upstream depletion of the monsoonal water vapour δ18O served as the main mechanisms linking the rainy-season or annual δ18Op to the April–July insolation variation at the precession scale. For the EA region, however, the precession-scale annual δ18Op was mainly controlled by the late-monsoon (August–September) and pre-monsoon (April–May) water vapour transport patterns, which were driven by the July–August insolation and the global ice volume, respectively. These results suggest that the climatic implications of the orbital-scale Asia δ18Op variations are sensitive to their geographic locations as determined by the combined effects of insolation and regional circulation patterns associated with the respective rainy seasons. This study provides new insights into understanding the regional differences and formation mechanisms of the Asian orbital-scale δ18Op variations.

A series of multivariate statistical methods were used for analysis of the simulation results. The periodicity of climate variables was identified by employing power spectrum analysis [8]. Cross spectrum analysis [9] was used to reveal the phase relationships among the  18 O p of different regions and forcing factors, such as the insolation variation of different months due to the orbital parameters. Principal component analysis (PCA) [10] was used to identify major modes of simultaneous variation patterns among multiple variables that have various degrees of correlations among them. We used the Pearson's correlation analysis to explore the associations and potential causal relationships between different variables [11], while partial correlation was used to evaluate the strength of the relationship between two variables as the effects of other variables (control variables) are eliminated [12]. In order to reveal the spatial response pattern of a given variable to possible influencing factors, we used regression analysis [13] to identify the empirical relationships first and then mapped the regression coefficients representing the changes in the dependent variable corresponding to per-unit change of the independent variable. Note S2. Validations of the annual cycles of the simulated air temperature, precipitation and  18 O p Comparing with the modern observations of temperature (CRU temperature data) [14] and precipitation (CPC merged analysis of precipitation) [15], the annual cycles of the simulated 300kyr-averaged temperature and precipitation in the CA, SA and EA regions are generally consistent with their respective observational characteristics of the 30-year (1981-2010) averages (Fig. S2), although the simulated monthly temperatures are consistently lower than the observations due to higher atmospheric CO 2 content at the present day and the presence of the Quaternary glacial periods. The simulated mean annual cycles of the  18 O p values in the three study regions are also similar to the limited modern observations in Central Asia (e.g., [16]), South Asia (e.g., [17]), and East Asia (e.g., [18]), although the simulated annual variation amplitudes are relatively small. These long-term mean annual cycles of temperature, precipitation and  18 O p in the three regions are relatively stable, even in different glacial and interglacial stages (figures omitted).

Note S3. Comparison of the simulated  18 O p and Chinese stalagmite records
The stalagmite δ 18 O (δ 18 O c ) is jointly controlled by δ 18 O of dripwater and the temperature inside the cave [19]. Since the dripwater δ 18 O mainly depend on the local precipitation of δ 18 O (δ 18 O p ) and the temperature-dependent fractionation is relatively small, δ 18 O c can inherit the rain water δ 18 O p to a large extent [20,21]. Here, we attempted to compare the simulated precipitation  18 O p series with relevant geological records. We only considered those stalagmite  18 O records with long time spans, high temporal resolutions, and reliable dating from China in the EA region [22]. Since the variation amplitudes of the simulated  18 O p series at the orbital scale are relatively small, similar to some of the previous studies [23][24][25][26], the simulated EA annual  18 O p series and the Chinese stalagmite oxygen isotope records are first standardized and then compared (Fig. S3). The result indicates that the variations of the oxygen isotope  [22] over the past 300 kyr; (b) Same as (a) but for the standardized series. composition in the simulated and geological records are highly consistent during the past 300 kyr (r=0.647), and the corresponding cross-spectral analysis also indicates that both of them vary almost in phase at the 23-kyr precession band, with only a 0.6-kyr phase difference. This also partially verifies the reliability of our simulation results, suggesting that the model simulated  18 O p values are highly comparable to the geological records after considering all major atmospheric and hydrological processes that affect the fractionation of oxygen isotope, at least for the EA region.

Note S4. Determination of the CA, SA and EA rainy seasons
Considering that the annual  18 O p is obtained as the average of individual month's  18 O p in the whole year weighted by the precipitation amount of each month, the annual  18 O p usually depends on the rainy-season  18 O p . Therefore, it is necessary to specify the rainy season for each study region for analyzing the changes in the annual  18 O p . By calculating the percentages of the monthly precipitation and the product (RD) of mean monthly precipitation and monthly  18 O p from January to December in the past 300 kyr (Table S1), it is found that high RD values occur during November-March for the CA region, June-September for the SA region, and May-September for the EA region, accounting for 81.7%, 73.6%, and 72.8% of their annual totals, respectively, and hence the respective rainy seasons are defined.

Note S5. Comparisons of periodicities of the GIV, GHG, and annual and rainy-season insolation series
Power spectral analysis results of the climate forcing factors in the late Quaternary ( Fig. S4) indicate that the variations of GIV [5] and GHGs (as the CO 2 equivalent concentration; [4]) are dominated by the quasi-100-kyr cycles (Fig. S4a,b), which also include the signals of the 41-kyr and 23-kyr cycles related respectively to the Earth's orbital obliquity and precession [27]. In the meantime, while the mid-to low-latitude rainy-season insolation has significant 23-kyr cycles ( Fig. S4d-f), the annual average insolation has the dominant 41-kyr cycle rather than the 23-kyr cycle ( Fig. S4c).  (Table S2). From the surface to 300 hPa, PC1 of each level mainly reflects a consistent variation pattern of the temperature from October to February, closely matching the November-March rainy season, and a reversed variation pattern of the temperature in the warm-season months (Table S2). In other words, the most dominant temperature variation pattern shows that when the PC1 scores are positive, there is the tendency for the surface to 300-hPa temperatures to be above normal during the cold-season months and below normal during the warm-season months.
When the PC1 scores are negative, the opposite temperature variation pattern occurs.     Table S4 presents the correlations between the January-December monthly temperatures at 700 hPa in the CA region and the insolation at approximately the same latitudes (45ºN ). It can be seen that the correlation coefficients between the monthly average temperature and the leading 1-2 month insolation are usually the highest. For example, the correlation coefficient between the 700-hPa December temperature and the leading November insolation is 0.569. Therefore, we examined the correlation between the CA rainy-season (November-March) average temperature and the mean insolation leading by 1 month (October-February). The result shows that the correlation coefficient between the CA average rainy-season 700-hPa temperature and the average October-February insolation is 0.619 (Table S5), which is significantly higher than those in the lower troposphere below 700 hPa, while the strengths of the correlations maintain for the levels up to 300 hPa (Table S5). Since the focus of this analysis is the potential linkage between the insolation and temperature for the months and isobaric levels relevant to the CA rainy season, those stronger correlations in the summer months (Table S4) or those in the upper troposphere isobaric levels with less moisture content (Table S5) are excluded.
Similarly, the correlation coefficients (Table S5)  For example, the correlation between the whole-troposphere  18 O v and insolation is as high as 0.877 (Table S5). It is noted that the correlation of S 10_2 with the rainy-season surface  18 O v11_3 is strong with r=0.883. In contrast, there is no correlation between the rainy-season precipitation and insolation (r=-0.048), which once again shows that the CA precipitation isotopic composition is related to the water vapor transport rather than the local precipitation variation.

Note S11. Correlations of the SA precipitation,  18 O p and the source region  18 O v with the 30º N insolation
To confirm the linkage between the variation of the SA rainy-season precipitation and the precession cycle, we correlated the January-December monthly precipitation series with the monthly insolation at 30º N. Table S7 shows that the precipitation of each month in the rainy season is usually most closely related to the insolation leading by two months. For example, the highest correlation coefficient of June (September) precipitation is with April (July) 30º N insolation as 0.827 (0.736).
Thus, the positive correlations between SA June-September precipitation and April-July insolation tend to be highest, while the negative correlations between the  18 O p or source region  18 O vs during the SA rainy season and April-July insolation are also the strongest (Table S7).  To explain the insolation-induced atmospheric physical processes affecting the SA precipitation and the source-region  18 O v , Fig. S7 shows the regression coefficient fields of the June-September whole-troposphere water vapor transport flux (Fig. S7a) and atmospheric water vapor content (Fig. S7b) in the Asia-Indian Ocean region with the April-July average 30º N insolation as the independent variable. When the insolation is strengthened in the NH spring and summer, the water vapor transport into the SA from the southwest is enhanced (Fig. S7a), resulting in increases in water vapor content (Fig. S7b) and precipitation over the SA region.    (Table S10).  weaker relations with these forcing factors (Table S12). Meanwhile, the August-September whole-troposphere water vapor content averaged for the EA shows strong positive correlations with the insolation during July and August, while the April-May water vapor content has strong positive correlations with March and April insolation (Table S11). These results suggest that the EA August-September and April-May  18 O p are likely controlled by the insolation and GIV (possibly including the GHG concentrations as well), respectively, due their modulating effects through regulating the large-scale atmospheric circulation patterns.