The Global Monsoon across Time Scales

Introduction Conclusions References

westward propagating Rossby waves and the mean westerly flow on their poleward side (Hoskins, 1996;Rodwell and Hoskins, 1996). The resultant coupled monsoondesert interactions govern a large portion of the tropics and subtropics (Fig. 2).
The monsoon domains defined by precipitation are different from, but dynamically consistent with, those defined by low-level winds. The tropical monsoon domains as 15 depicted by the annual reversal of zonal wind can be define by the following criterion: The local summer westerly minus winter easterly at 850 hPa exceeds 50 % of the annual mean zonal wind speed (Wang and Ding, 2008). Such defined monsoon wind domains are shown in Fig. 3 (bold black contours) and they agree closely with previous definitions made by Ramage (1971) using three wind criteria. Unlike Ramage

Evolution of the global monsoon concept
Regional monsoons are bonded by the global divergent circulation. Trenberth et al. (2000) depicted the global monsoon as the global-scale seasonally varying overturning circulation throughout the tropics. Considering that the physical principle of conservation of mass, moisture, and energy applies to the global atmosphere and its exchange 5 of energy with the underlying surfaces, the analysis of broader monsoon variability from a global perspective is imperative and advantageous for understanding fundamental monsoon controls and dynamics.
The GM can be quantitatively defined as the dominant mode of the annual variation of precipitation and circulation in the global tropics and subtropics (Wang and 10 Ding, 2008). The first empirical orthogonal function mode of the annual variation of global precipitation and low-level (850 hPa) winds (Fig. 4) accounts for 71 % of the total annual variance, has a prominent annual peak, and features an inter-hemispheric contrast in precipitation (Fig. 4a). This mode can be simply characterized by the difference in mean precipitation and circulation between the two extended solstitial sea- 15 sons, June-July-August-September (JJAS) and December-January-February-March (DJFM) (Fig. 4b). This solstitial mode depicts seasonal changes in the GM precipitation and associated low-level monsoon circulation and reflects the fact that the GM is a forced response to the variation of solar insolation, with a phase delay of a few months that reflects the oceans' considerable thermal inertia. The regional monsoons (Fig. 2) summer occurs over India and is due to strong warming of the Eurasian landmass and the Tibetan Plateau heating effect (Yanai and Wu, 2006). The southern extreme in austral summer occurrs near Madagascar and is related to the impact of the strong southward cross equatorial flow originating from NH winter monsoon and associated Iranian High (Fig. 3b). Due to the large meridional scale of the flow (greater than 20 degrees of 10 latitudes), geostrophic effects divert the cross-equatorial flow eastward on the equatorward side of the summer hemisphere ITCZ; and the winds in between the northernmost and southern most locations of the ITCZ reverse direction annually (Fig. 3). This portion of the summer ITCZ is often associated with a low-pressure monsoon trough. For this reason, the oceanic convergence zones over the Southern Indian (40-100 • E), WNP 15 (110-150 • E), Southwest Pacific (150-180 • E), far eastern North Pacific (70-110 • W), and eastern Atlantic Oceans can be regarded as marine monsoon troughs that are characterized by a uni-modal annual variation of rainfall, with a significant peak in summer and development of monsoon westerlies on their equatorward side. Conversely, in the central Pacific Ocean, where the ITCZ annual displacement is less than 5 degrees 20 of latitude, the zonal wind (trade winds) between the ITCZ do not change direction between summer and winter, and therefore a monsoon is absent. This portion of the ITCZ can be referred to as "trade wind convergence zone" (B. Wang, 1994). This trade wind convergence zone has a persistent rainy season throughout the year and a bi-modal seasonal distribution of rainfall with maxima in the equinoctial seasons. 25 In summary, the ITCZs between the dateline and 110 • W in the Pacific and between 30 and 60 • W in Atlantic have minimal seasonal meridional migration and a bi-modal seasonal rainfall distribution, and are thus viewed as trade wind convergence zones. The bulk (about three quarters) of the ITCZ is embedded within the monsoon regions 2174 to the northwest edge of the WNP Subtropical High. The WNP monsoon tough and WNP Subtropical High are thus a dynamically coupled system (Tao and Chen, 1987), and changes in the tropical WNP monsoon can indirectly influence the EA subtropical monsoon through changing the WNP Subtropical High. In this sense, the EA summer monsoon is best viewed as part of the GM system.

Modern monsoon observation
Monitoring the GM entails observing winds, temperature, rainfall, clouds, and humidity, among other fields, on a planetary scale across a broad range of temporal scales from hourly to interdecadal. The challenges involved are therefore considerable. The era of coordinated planetary scale tropospheric observations adequate for this task com- 15 menced with the establishment of the northern hemisphere upper air network in the 1940's (e.g. Kalnay et al., 1996). Early monsoon studies used data from this network, primarily in a piecemeal fashion addressing phenomenology of regional monsoons. Building upon these data, atmospheric "reanalysis" enabled diagnosis of the monsoons on a global scale, with a more coordinated and holistic approach. Reanalyses provided 20 an unchanging data assimilation scheme and modeling framework in an effort to remove sources of spurious variability from operational analyses. Nonetheless, due to their variable input data streams, reanalyses do suffer from spurious temporal variability and a lack of sufficient observations results in reanalysis fields that are strongly model influenced and therefore of questionable realism. Still, given their detailed nature 25 and global scope, reanalysis products are among the mostly widely used data sources for diagnosing the GM and have proven to be instrumental in understanding it, when interpreted with appropriate care. Improvement in reanalyses also offers the prospect for 2175 enhanced realism and diagnostic capabilities over time, as products have evolved from 1st generation (e.g. Kalnay et al., 1996;Kistler et al., 2001), to 2nd generation (e.g. ERA-40, Uppala et al., 2005, JRA-25, Onogi et al., 2007, and 3rd generation products (ERA Interim, Dee et al., 2011, MERRA, Rienecker et al., 2011, incorporating more sophisticated data assimilation approaches (e.g. analysis increments and 4D-Var) and 5 models.
Satellite retrievals are also a fundamental source of data for understanding the GM, providing a multi-decadal record of rainfall, clouds, water vapor, temperature, and surface winds. The satellite record is particularly valuable given that the majority of monsoon domains lie over data-scarce ocean regions. As with reanalyses, satellite re-10 trievals are also being improved over time and new instruments offer new capabilities and potential for understanding the monsoons. For example, the recently launched Global Precipitation Measurement (GPM) instrument, has greater sensitivities to a range of precipitation intensities and greater spatial coverage than did its predecessor satellites. Combined with improvements in reanalyses, these sources of data offer con- 15 siderable promise for continued progress in understanding the GM. Counter-examples to these advances also exist, as other aspects of the observing system have degraded over time. For example, the density of surface rain gauges has decreased considerably in recent years (e.g. Becker et al., 2013) and observations of river discharge have been in decline for decades (e.g. Bjerklie et al., 2003). These deficiencies present a Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the Atlantic (Sarnthein et al., 1981), and of sapropels in the Mediterranean (Rossignol-Strick et al., 1982;Rossignol-Strick, 1983). At the same time, paleo-records of the Indian monsoon were discovered in the Arabian Sea on the basis of upwelling variations (Prell, 1984). All these studies focused on the latest Quaternary with a temporal resolution of about 1 kyr, using two groups of proxies: those related to wind, using 5 deep-sea eolian dust, upwelling-induced marine productivity, and those related to rainfall, such as lake level and flood-induced sapropels. Given the short duration of the analyzed sections and relatively coarse time resolution, there was a good correlation between the two groups of proxies, with both indicating a key role for orbital forcing at the precessional band. 10 The 1990s witnessed a boom in the development of paleo-monsoon research over several regions. The Ocean Drilling Program (ODP) Leg 117 in the western Arabian Sea in 1987 was the first drilling cruise specifically designed for retrieving proxies of the evolution and variability of the Indian monsoon, extending the monsoon history back to the late Pliocene and significantly improving our understanding of the driving 15 mechanism of monsoon variability (e.g., Clemens et al., 1991Clemens et al., , 1996. Another longterm sequence of paleo-monsoon records was obtained from the Loess Plateau of China, where thick eolian deposits were successfully cross validated with deep-sea sediments, yielding a history of the East Asian monsoon over the last 7-8Ma (e.g., An et al., 2001). For the African monsoon, the ODP Leg 108 in 1986 to the Equatorial Atlantic 20 recovered a Plio-Pleistocene sequence of Sahara dust that documented the period of hominids evolution in conjunction with variability in the African monsoon (deMenocal, 1995).
While the above-discussed monsoon sequences were primarily based on windrelated proxies (eolian dust, upwelling-induced productivity), others have been estab- 25 lished using precipitation-related proxies, including several from the Mediterranean and South China Seas. The ODP Leg 160 to the Eastern Mediterranean in 1995 (Emeis et al., 1996) and Leg 184 to the South China Sea in 1999 (P. Wang et al., 2000) recovered monsoon records back to the Neogene based on the rain-related proxies of chemical 2177 weathering rate, sapropel occurrence, and types of land vegetation. With the application of new techniques like X-ray fluorescence core-scanning, the time resolution of these reconstructions approach 2-3 hundreds years for the last 5 Ma (e.g., Tian et al., 2011).
Monsoon archives with much higher resolution have been recovered from marine 5 or lacustrine cores with lamination or ultra-high sedimentation rates for the late Quaternary. For example, reconstructions of monsoon sequences with decadal resolution were provided from the South China Sea (L. Higginson et al., 2003) and the Arabian Sea (Agnihotri et al., 2002;Gupta et al., 2003). Even very high resolution records (4-5 years) have been made available from the Cariaco Basin for the North 10 American monsoon (Haug et al., 2001) and elsewhere. However, sediment archives have their limitations in resolving absolute age, and paleoclimate records with annual or even seasonal resolution can be retrieved from other sorts of material such as tree rings, coral reefs, ice cores and stalagmites. Of particular significance are the latter two archives: ice cores and stalagmites. 15 Although ice core drilling began as early as in the 1960s, monsoon variations recorded in ice-core air-bubble have become available only since in the 1990s. Of particular importance are two parameters: methane concentration (Chappellaz et al., 1990;Blunier et al., 1995) and atmospheric oxygen isotope fraction (Bender et al., 1994). Since 2000, speleothem records have become available in several monsoon 20 regions and have become the focus of recent paleo-monsoon work. Given the annual banding and highly precise dating technique of thorium-230, speleothem's paleoclimate sequence can be resolved at annual resolution, improving upon many previous monsoon reconstructions. Over the last decade, numerous speleothem oxygen isotope sequences have been published, including those from South China (e.g., Y. Wang 25 et al., 2001Yuan et al., 2004), now extending back to over 380 ka , and to South America (X. Cruz et al., 2005). These speleothem records clearly demonstrate predominance of precession forcing of the 2178 Asian monsoon in orbital scale and its correlation with high-latitude variations at millennial scales.
The new findings resulting from speleothem records have attracted broad interest but have also raised new debates in the paleo-climate community. As in ice-core records, speleothem-derived monsoon sequences are dominated by a 23 ka precessional peri-5 odicity. This finding is in sharp contrast to the monsoon records from the Arabian Sea in which the obliquity forcing exceeds that of precession (Clemens and Prell, 2003). The two diverging views on monsoon variations differ in orbital-scale periodicity and phasing: with the former assuming a direct response to boreal summer insolation, while the latter infers an 8-ka delay in responding to precession, due to latent heat transfer from the Southern Hemisphere (Ruddiman, 2006). This divergence in opinion has evoked a hot debate as to which proxies are representative of the Asian monsoon: the marine records from the Arabian Sea or the speleothem records from the Asian land (e.g., Clemens and Prell, 2007;Clemens et al., 2010;Ziegler et al., 2010;Weber and Tuenter, 2011) which will be discussed in in 15 our follow on work. Here our goal is to note that the divergence of opinion is, at least partly, related to the different nature of the proxies used: with upwelling records based on wind being physically distinct from the speleothem records based on rain. Looking back at the evolution of paleo-monsoon research, it was initiated with both wind-and rain-based proxies, and the two kinds of sequences correlated fairly well at that stage. 20 Controversies appeared however with the introduction of new proxies over the last decade, in particular δ 18 O from the speleothem calcite and atmospheric oxygen. Both are proxies related to hydrological processes and the resultant monsoon sequences show small lags of 2-3 ka in response to insolation, in contrast to considerably longer lags of 5-8 ka of the sequences in the Arabian Sea. A small lag has been supported by assessing monsoon variations beyond instrumental records relies on the use of proxies, i.e. indirect measures of past climate features preserved in natural archives. Concerning the GM, two questions are to be addressed here: firstly, which proxies can be used for quantitative assessment of regional monsoons for a global synthesis and, secondly, can we find proxies to measure GM variations?

Monsoon proxies
In general, monsoon proxies can be divided into two groups according to the primary aspects of the monsoon that they address: proxies related to monsoon winds (direction, strength and persistence), and those associated with monsoon precipitation. Attached is a simplified table of frequently used proxies to indicate monsoon variations in geo-15 logical records (Table 1). For more detail the reader is referred to the SCOR/PAGES review of the Asian monsoon evolution and variability (Table 1 in P. .

Wind-based proxies
As seen from Sect. 3.2.1 and Table 1, eolian dust and coastal upwelling are the most frequently used of the wind-based monsoon proxies. Eolian dust has been used for 20 estimating monsoon variations in various oceans (e.g., Sarnthein et al., 1981;Sirocko et al., 1993), as well as in loess-type deposits, particularly in China. In the 1990s, the Plio-Pleistocene history of the East Asian monsoon was diagnosed from the Loess Plateau in China, using the grain size of loess particles to indicate winter monsoon and magnetic susceptibility of the loess-paleosol sequences for summer monsoon (e.g., An paleo-monsoon proxies, such as stalagmites and ice-cores, are also rain-based. Since the modern monsoon is most often studied in the context of hydrological cycle, rainbased proxies offers the opportunity for an analogous assessment of the GM across time scales.
Over recent years there has thus been an increase in the number of paleo-monsoon 5 publications based on rain-based proxies. For example, high-resolution records of chemical weathering rate in marine sediment ascribed to monsoon precipitation have been used extensively. With introduction of the new techniques of nondestructive analyses such as X-ray fluorescence, some element ratios such as Ti / Al and K / Si have been used to explore variability in monsoon precipitation with much higher time reso-10 lution than previously available chemical analyses (e.g., Tian et al., 2011). River runoff is another index used in paleo-monsoon assessment, with the best example perhaps being the sapropel layers deposited in the Mediterranean produced by the extreme Nile flooding in response to the intensified African summer monsoon (e.g., Ziegler et al., 2010). A number of weathering-related chemical proxies have also been applied to 15 monsoon analysis at tectonic time scale (Clift et al., 2014). Table 1 shows a wide range of proxies indicative of humidity changes in marine and terrestrial records, including the loess sequences. While the pioneering works on paleo-monsoons in the Loess Plateau of China mostly used wind-induced proxies, newer studies have emphasized the proxies resulting from pedogenesis (e.g., Guo et al., 2000). 20 It is remarkable that almost all new, high-resolution archives of paleo-monsoon variability are associated with hydrological rather than wind processes. This includes proxies derived from tree rings, ice-cores and speleothemes, which each hold the potential for achieving annual resolution. Nonetheless, there are also caveats for using these rain-based proxies, including uncertainties arising from the use of tree-ring δD (Feng Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2009a), although, at mentioned above, the extent to which speleotheme δ 18 O acts as a strongly constrained indicator of summer monsoon intensity remains unclear.

Exploring proxies for global monsoon
The recognition of the GM as a global system poses a question: how can we collectively measure the GM intensity variations in the past? Since seasonality is an inherent 5 feature of the GM, its intensity can be measured explicitly, for example, as globally averaged seasonal range of monsoon precipitation (summer-minus-winter precipitation) (Wang and Ding, 2006), or merely from the global averaged local summer rainfall, which dominates the seasonal range (B. Wang et al., 2012). The local summer rainfall in the monsoon region dominates the annual total rainfall amount, hence the annual total rainfall, to a large degree, can be used as an approximate indication of overall monsoon strength. In this sense, proxies with annual resolution are sufficient. Because the monsoon signals from various regions are being mixed and intergrated by atmospehic and oceans circulations, the air-bubble and marine records are more promising for assessment of GM intensity than the terrestrial reocrds which mostly reflect the ef- 15 fects of the prevailing regional monsoon. Accordingly, the following discussions will be focused on ice-core air bubbles and on marine records.

Ice-core air bubble
Ice cores provide an invaluable record of climate changes with "fossil air" captured in their bubbles. Of the various proxies available from air bubbles, two have been sug-and Raymo, 2005). As boreal wetlands are also a main methane source, their extent is closely linked with the climate and ice conditions at the northern high-latitudes (Landais et al., 2010). As shown by a recent study, atmospheric CH 4 records integrate all wetland processes, including the important effects of changes in monsoonal circulations of both hemispheres and the non-monsoonal contributions mainly from boreal 15 wetlands. Approximately, it contains ∼ 60 % of the signals from the past changes of GM, and ∼ 40 % from the boreal wetlands . Therefore, the use of CH 4 as a GM proxy is likely to be limited by the major influence of boreal wetlands on the overall methane budget.

20
Another parameter available from ice cores clearly related to monsoon variability is atmospheric δ 18 O in air bubbles. A strong similarity exists between the δ 18 O atm and 65 • N summer insolation on orbital time scale (Fig. 5b;Petit et al., 1999;Landais et al., 2010)  relating to ice volume and monsoon-related orbital variations respectively (Shackleton, 2000). Consequently, the δ 18 O atm record, like that of CH 4 , is a somewhat convoluted but potentially useful proxy of the GM. A more viable basis for a GM proxy may be the Dole effect, i.e. the difference between the δ 18 O of atmospheric O 2 in air and the δ 18 O of contemporaneous seawater 5 (Fig. 5a;Bender et al., 1994;Landais et al., 2010). δ 18 O of atmospheric O 2 over glacial cycles responds to changes in the δ 18 O of seawater, and the Dole effect represents oxygen isotope fractionation during photosynthesis, respiration, and hydrologic processes (evaporation, precipitation, and evapotranspiration); all of these are processes related to the GM. The influence of low-latitude processes on isotopic fractionation governing the Dole effect is highlighted by the similar magnitude of its LGM and present values despite vast environmental differences between the two periods (Bender et al., 1994). According to Luz and Barkan (2011), the Dole effect is not primarily sensitive to past changes in the ratio of land-to-sea photosynthetic rates, but rather to changes in low-latitude hydrology. Therefore, the strong correlation between variability in the 15 Dole effect, Northern Hemisphere insolation, and monsoon records are unsurprising (Severinghaus et al., 2009). In general, the use of the Dole effect as a proxy of GM is supported by both theoretical analyses and observations. However, the current method of calculating the Dole effect (the difference between δ 18 O atm and δ 18 O sw ) is likely simplistic, because of the 20 complex process of isotopic fractionation in the hydrological cycle. At least three issues should therefore be considered when calculating the Dole effect: (1) the absence of a common timescale for marine and ice core records (Landais et al., 2010); (2) difficulties in obtaining pure values of δ 18 O sw (Rohling and Bigg, 1998;Waelbroeck et al., 2002)

Deep-sea sediments
Another potential archive of GM variability is deep-sea sediment. In addition to marine δ 18 O, which is used for Dole effect calculation, marine δ 13 C is a useful proxy at time scales longer than glacial cycles. Because of the long residence time of carbon in the oceanic reservoir beyond 100 ka, the carbon isotope record exhibits 400 ka eccen-5 tricity cycles over the last 5 Ma, with maximum values (δ 13 C max ) occurring at eccentricity minima (Fig. 6). The 400 ka periodicity is observed in both benthic and planktic δ 13 C records, as well as in carbonate preservation records, implying rhythmic fluctuations in the oceanic carbon reservoir (P. Wang et al., 2010). In high-resolution deep sea records, the 400 ka long eccentricity cycles can be traced back through the entire Cenozoic (see below Sect. "Eccentricity modulation"). However, it is still premature to use long-term records of marine δ 13 C as a GM proxy over this long eccentricity band. The main problem is that the long-eccentricity cycles in the oceanic carbon reservoir can be disturbed by changes in the oceanic circulation. The 400 ka periodicity so clearly visible in the Pliocene δ 13 C records is obscured in 15 the Pleistocene after ∼ 1.6 Ma, and since then, the δ 13 C max no longer correspond to eccentricity minima (P. . A similar effect also happened in the middle Miocene beginning ∼ 13.9 Ma (Holbourn et al., 2005(Holbourn et al., , 2007Tian et al., 2009). The 1.6 and 13.9 Ma breakdown of the 400 ka cycles in oceanic δ 13 C records probably derived from reorganization of the ocean circulation, as both cases correspond to times of 20 significant expansion of the polar ice-sheets accompanied by the establishment of new deep ocean circulations. It is suggested that ice-sheet growth may have disturbed the normal 400 kyr periodicity of the Earth's climate system (P. Wang et al., 2010Wang et al., , 2014. Meanwhile there are observations indicating the presence of long eccentricity cycles in monsoon records after 1.6 Ma. For instance, the δ 13 C max remained present at 1.2 25 and 0.8 Ma in the Mediterranean (Fig. 6b), and dust flux minima at 0.4 Ma due to a weaker monsoon have been found in the equatorial Atlantic (Tiedemann et al., 1994).
In the equatorial Indian Ocean, the 400 ka cycles are observed in primary productivity  Beaufort et al., 1997) whereas the δ 13 C and carbonate records from the same site display 500 ka cycles (Bassinot et al., 1994). It is likely that the influence of the 400 kyr cycle persisted in low-latitude processes in the climate system such as monsoons, but was dampened in the oceanic carbon reservoir . Thus, it remains unclear as to how the GM signal enters marine inorganic δ 13 C records. We 5 also do not know how the observed 500 ka "super-cycles" are related to GM variations. Nevertheless, the above discussion narrows the choice of GM proxies to two ideal candidates: the Dole effect at the precession frequency band represented by the δ 18 O atm record in ice-cores with ice-volume signals removed, and the marine inorganic δ 13 C at the long-eccentricity band. The two parameters are mutually connected 10 by the monsoon-driven low-latitude hydrological cycle. Although the Dole effect and marine δ 13 C are proposed as GM proxies, there is a long way to go before confidence in them is high, and a number of important questions have yet to be answered before the physical meaning of these and related parameters can be revealed. The Dole effect will also evolve in its definition and method of calculation, and the marine δ 13 C record, 15 with its long eccentricity period, needs to be further explored during periods of high ice-volume and especially during the Pleistocene.

Interannual variation
Quantification and understanding of present-day monsoon variability are crucial for 20 interpreting its past and predicting its future. Interannual variations can best be studied using the modern observed record and have been widely studied on regional rather than global scales due to their indigenous characteristics associated with specific landocean configuration and differing feedback processes. The principal regional domains include South Asia (e.g., Webster et al., 1998), East Asia (e.g., Tao and Chen, 1987), Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | (e.g., Higgins et al., 2003), and South America (e.g., Zhou and Lau, 1998). To what extent can the interannual variations of GM precipitation (GMP) be driven by internal feedback processes in the climate system? Interannal variability is commonly studied using the calendar year to delineate successive time periods. However, for the analysis of GMP, doing so is inadequate, be-5 cause, seasonal distribution of GMP has two peaks, one in July-August (due to NHSM) and one in January-February (due to SHSM), with a minimum in April. In order to depict GMP interannual variation, we therefore use the "global monsoon year", which begins on 1 May and extends through the following 30 April, which includes the NHSM from May to October followed by the SHSM from November to April next year. The monsoon year concept is also suitable for depicting ENSO evolution, because ENSO events normally start in late boreal spring or early summer, mature toward the end of the calendar year, and decay in the following spring.
Analysis of global observations over the past three decades reveals a coherent interannual variation across regional monsoons from one monsoon year to the next 15 (B. Wang et al., 2012). The spatial pattern of the interannual monsoon precipitation variability (Fig. 7a) is characterized by a coherent rainfall signature across a majority of monsoon regions except the southwestern Indian Ocean and the southern part of the South American monsoon. Notably, this coherence exists over nearly all continental monsoon regions. An overall drying pattern is associated with the warm phase of 20 El Niño-Southern Oscillation (ENSO) (Fig. 7b), while enhanced precipitation characterizes La Niña conditions. Thus, from one monsoon year (May to the next April) to the next, most monsoon regions, despite being separated by vast areas of arid trade winds and deserts, vary in a coherent manner, connected by ENSO and associated atmospheric teleconnections. 25 Despite the generally uniform pattern of the leading GMP mode (Fig. 7a), ENSO exerts a tighter control on the NHSM than on the SHSM. ENSO also tends to more strongly influence continental monsoon rainfall than the oceanic monsoons. total amount of the NHSM land rainfall is highly related to ENSO with correlation coefficient r = 0.73 for 1979-2008. In many respects, this ENSO influence can be thought of a governing influence on the competition between rainfall over land and ocean on a global scale, which influences the global-integrated storage of water over land and contributes significantly to 5 interannual variability of sea level (e.g. Boening et al., 2011). In regions where the atmospheric response to ENSO tends to produce a regional dipole pattern, such as over East Asia, southern Africa, and South America, regional characteristics of variability should be emphasized. In addition, the dominant spectral peaks in monsoon precipitation associated with ENSO vary by region as variations are 3-7 years over West Africa, 10 North America and Australia but only 2-3 years over South and East Asia and southern Africa (Zhou et al., 2008a).

Interdecadal variation
On interdecadal timescales, numerous studies have investigated the linkage between regional monsoons and other major modes of climate variability. For instance, Indian 15 summer monsoon precipitation has been shown to exhibit a correlation with the North Atlantic Oscillation (Goswami et al., 2006), Northern China's rainfall iscorrelated with the Pacific Decadal Oscillation (Cheng and Zhou, 2013), and west African and North American monsoon variability is related to the Atlantic Multidecadal Oscillation (AMO) (Zhang and Delworth, 2006). A variety of decadal and interdecadal variations of re-20 gional monsoons has been identified, with differing periodicity and phase change points . However, the underlying causes of GM interdecadal variability has yet to be widely studied.
Studies of the GM change on interdecadal timescales requires long-term wellcalibrated observations. The scarcity of extended duration oceanic rainfall observa- interdecadal variability, with a decreasing trend mainly due to weakening of the West African and South Asian monsoons from 1950s to 1970s. However, since 1980 the global land monsoon rainfall has no significant trend, while global oceanic monsoon precipitation shows an increase (Wang and Ding, 2006;Zhou et al., 2008a;Fasullo, 2012). This long-term reduction in global land monsoon precipitation can be reason-5 ably well captured by atmospheric general circulation models forced by the observed sea surface temperature (SST) anomalies (Zhou et al., 2008b). It is notable however, that major observational uncertainties also exist as the gauge network has major spatial gaps and quality control issues and global land monsoon rainfall records from other sources (e.g. reanalyses) show significant discrepancies. Across reanalyses, mutual 10 discrepancies arise from changes in the assimilated data streams, both prior to and during the satellite era, suggesting that trends that in other studies have been taken as real are likely be spurious (Fasullo, 2012).
With the most updated GPCP data (Huffman et al., 2009 show that during the recent global warming of about 0.4 • C since the late 1970s the 15 GMP has intensified over the past three decades mainly due to the significant upward trend in NHSM while the SHSM has exhibited no significant trend. A recent intensification of the NHSM originates primarily from an enhanced east-west thermal contrast in the Pacific Ocean, coupled to intensification of the subtropical high in the eastern Pacific and decreasing pressure over the Indo-Pacific warm pool. This enhanced Pa-20 cific zonal thermal contrast is largely a result of natural variability (e.g. Merrifield, 2012) and tends to amplify both the NHSM and SHSM. On the other hand, the hemispherical thermal contrast may be enhanced due to anthropogenic forcing. The stronger (weaker) warming trend in the NH (SH) creates a hemispheric thermal contrast, which favors intensification of the NHSM but in turn weakens the SHSM. The combined ef- 25 fects of the two factors may help explain why the NHSM has intensified over recent decades while the SHSM has exhibited little trend.
The aforementioned studies of monsoonal trends in the modern record have yet to be able to definitively conclude whether these trends are attributable to anthropogenic forcing or arise from internal variability. Using a circulation index for the NHSM, it is possible to investigate the interdecadal variability of the GM as reanalysis data has become available for the past century (e.g., monthly circulation data taken from 20th Century Reanalysis for 1871, Compo et al., 2011. B. Wang et al. (2013) used the NHSM circulation index defined by the vertical shear of zonal winds between 850 and 5 200 hPa averaged in (0 • -20 • N, 120 • W-120 • E) (Fig. 8a). While NHSM circulation index is highly correlated with the NHSM rainfall intensity over the modern record (r = 0.85 for 1979-2011), this approach has the drawback of potentially being insensitive to changes in rainfall since it is based on dynamics alone. Nonetheless, a major finding of the work has been to demonstrate that the NHSM circulation has experienced large amplitude interdecadal fluctuations since 1871, which is primarily associated with what has been termed "Mega-ENSO", the SST difference between the western Pacific K-shape area and eastern Pacific triangle (Fig. 8b). The mega-ENSO has larger spatial scale then ENSO and a longer decadal-to-multidecadal time scale. The mega-ENSO index is an integrated measure of ENSO, the Pacific Decadal Oscillation (PDO, Mantua and Hare, 15 2002), and Interdecadal Pacific Oscillation (IPO, Power et al., 1999), and is defined by using the total anomaly field, rather than an EOF decomposition. On the decadal time scale, as measured by the 3 year running mean time series, the long-term variation of the NHSM circulation index has been shown to be strongly correlated to the Mega-ENSO index (r = 0.77, for 1958-2010) using ERA-40 reanalysis data (Uppala et al., 20 2005). This significant correlation is confirmed for the period 1871-2010 (r = 0.62) by using the 20th century reanalysis dataset (Fig. 8c). Physically, the eastern Pacific cooling and the western Pacific warming are consistent with a strengthening of the Pacific subtropical Highs in both the North and South Pacific and their associated trade winds, causing moisture to converge into the Asian and African monsoon regions and 25 contribute to the intensification of NHSM rainfall. In addition, Zhao et al. (2012) show that an amplified land-ocean thermal contrast between the Eurasian landmass and its adjacent oceans can be described by a positive phase of the Asian-Pacific Oscillation (APO), whose positive phase corresponds to a Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | stronger than normal NH summer monsoon and strengthened southerly or southwesterly monsoon winds over tropical Africa, South Asia, and East Asia. These circulation anomalies induce enhanced summer monsoon rainfall over all major NH land monsoon regions. Monsoon interdecadal variability includes not only interdecadal variations of the 5 mean monsoon but also variations of interannual variability itself (second moment), which also has considerable impacts. It is well known that the monsoon-ENSO relationship is non-stationary and it changes over the multi-decadal time scale (Webster et al., 1998). For instance, the ISM-ENSO relationship has weakened since late 1970's (Kumar et al., 1999), while the relationships between ENSO and the East Asia-western 10 North Pacific, and Indonesian monsoons have strengthened (B. . As such, both the year-to-year variability and interconnectedness of the monsoons often exhibits strong interdecadal variability, which can be attributed in part to changes in ENSO characteristics (amplitude, frequency, and onset characteristics). What drives such variability in ENSO remains an open science question. 15 There has been little study of the changes in the interannual variability of the global or hemispheric monsoons on the interdecadal time scales. As for interannual variability, decadal shifts in the GM are linked to changes in rainfall over land globally and are mirrored in the changing rates of sea level rise (Fig. 8). These shifts have the potential to extend an understanding of monsoon variability back in time through scrutiny of the 20 sea level gauge record.

Summary
As discussed in Sect. 4.2, ENSO has dominant influence on the interannual variation of the GM while interdecadal variations of the NHSM are likely driven by Mega-ENSO or Interdecadal Pacific Oscillation (IPO) influences (e.g. Power et al., 1999;Parker et 25 al., 2007). In addition, NHSM interdecadal variations are shown to be significantly associated with the Atlantic Multi-decadal Oscillation (B. Wang et al., 2013) and APO (Zhao et al., 2012). Thus, regional monsoons can be coordinated not only by changes 2192 Introduction in orbital forcing (Kutzbach and Otto-Bliesner, 1982;Liu et al., 2004) and variations on centennial-millennial time scales (Liu et al., 2009), but also by internal feedback processes on interannual and multidecadal time scale. These are likely to act dynamically, through global atmospheric teleconnections and related atmosphere-ocean interactions on a planetary scale.
It is likely therefore that the NH and SH summer monsoons are not only governed by a common set of processes, such as the enhanced east-west thermal contrast in the Indo-Pacific, but also by inter-hemispheric thermal asymmetries associated with their contrasting land extents that have potentially opposing effects on the NH and SH monsoons (B. Wang et al., 2012). Study of such interdecadal variation of GM is still in an early stage, however there are strong indications that the causes of the interdecadal variability of the SHSM and the GSM warrant further study. An improved understanding of the physical processes that control the GM and its interdecadal variability is therefore a top priority. 15 Climate variations on sub-orbital timescales in low-to middle latitude monsoon regions have been identified in high-resolution proxy records provided by ice-core, coral, tree ring, lacustrine, loess, marine deposit, and historical records (e.g., Thompson et al., 1986;Cobb et al., 2003;Ge et al., 2003;Cook et al., 2010;Bird et al., 2011;Sun et al., 2012;Bard et al., 2013;Ziegler et al., 2013;Deplazes et al., 2013). Over recent 20 decades, however, speleothems have attracted primary scientific attention and offered a new perspective on paleo-climate reconstructions (Henderson, 2006). This archive is particularly suitable for characterizing GM variability on sub-orbital timescales, because of the distinct advantage of its considerable spatial coverage in all major regional monsoon domains, and its high temporal resolution and precise chronology. Today, a Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | related coherent variability across the regional monsoons is due to the speleothem records.

Millennial scale variability
Millennial-scale climate oscillations or events were large in amplitude and well documented over the past few glacial periods, contrasting to the much smaller counterparts 5 during interglacial periods, for example the Holocene. In the 1970s, Greenland ice core records revealed a series of millennial events, which demonstrated that temperatures over Greenland have oscillated on millennial-scale during the last glacial time abruptly and significantly between two modes, extremely cold (stadials) and relatively mild (interstadials) intervals (e.g., Broecker et al., 1985). Such millennial oscillations In the 1980s, North Atlantic marine sediment records revealed several episodes of usually abundant ice rafted debris (i.e., Heinrich 1988;Bond and Lotti, 1995), referred as Heinrich events (H events) (Fig. 9). In addition, it was also found that colder temper- 15 atures over Greenland correlate with lower ocean temperatures inferred from larger percentages of the polar foraminifera species in marine sediment records from the North Atlantic Ocean (e.g., Bond et al., 1993). In contrast however, millennial events revealed from Antarctic ice cores have much smaller amplitudes. On the basis of CH 4 correlations (Blunier and Brook, 2001), a gradual Antarctic warming was found to co-20 incide with Greenland cold events. This pattern of opposing temperature changes in the North Atlantic and Antarctica was termed the "bipolar seesaw" (e.g., Stocker et al., 1992;Broecker, 1998;Stocker and Johnsen, 2003), a hypothesis that attributes changes in the Atlantic meridional overturning circulation in the Atlantic Ocean as a trigger and/or propagator of global millennial oscillations (e.g., Broecker and Denton, 1989;Broecker, 2003). Similar millennial-scale climate oscillations have also been clearly documented in low to middle latitude regions, particularly in monsoonal regions, including the East Asian 2194 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version
On the basis of this considerable body of proxy records, a comprehensive pattern of GM millennial variability has merged (e.g., Cheng et al., 2012a). While the NH summer monsoons (i.e., the North African, Asian including both Indian and East Asian, 20 and North American-Mesoamerican summer monsoons) weakened abruptly during millennial-scale cold events (such as the YD event, Greenland stadials or Chinese stadials), the Indo-Austrian, South American and South African monsoons intensified in SH during same episodes, and vice versa. This contrasting interhemispheric behavior is now evident across a wealth of proxy data (Fig. 9). In addition, records from Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | turn, winter precipitation increases (Asmerom et al., 2010). As a result, the millennialscale events in this region are apparently anti-phased, in terms of precipitation δ 18 O and possibly amount, with their counterparts in Cariaco basin/the tropical North Atlantic Ocean, Mesoamerican, and the Asian-North African monsoons (Fig. 9). To first order, anti-phased millennial monsoon events in both hemispheres are broadly similar 5 in terms of their amplitude and abruptness as inferred from proxy records. They also appear to be more comparable with Greenland than with Antarctic events.

Abrupt vs. gradual changes: the Holocene history
The Holocene (the past 11.5 ka) climate in monsoon regions is generally stable in comparison to during the last glacial period. It generally exhibits a long-term trend broadly tracking the summer insolation. A warm and wet period in NH monsoon regions occurred approximately between 9-5 ka BP during the so-called the Holocene Climatic Optimum. In contrast, monsoon variations in the SH during the Holocene, such as for example in South America, relate strongly to austral summer insolation and thus are approximately anti-phased with the NH monsoons (see next section for details). In ad- 15 dition to gradual changes in insolation, non-linear response/feedback mechanisms may result in a series of abrupt monsoonal variations. For example, Morrill et al. (2003) emphasized two major abrupt climate events, likely occurred synchronously in the Asian monsoon domain at ∼ 11.5 ka and 5.0-4.5 ka BP, respectively. The first abrupt intensification of the Asian monsoon occurred at ∼ 11.5 ka BP and corresponds to the onset 20 of the Holocene. This monsoon jump also marks the end of the YD event, which is extremely abrupt with a major shift that occurred over ∼ 30 years (Alley et al., 1993;Ma et al., 2012). The onset of the Holocene at 11.5 ka BP is a very dramatic event with global effects, including intensification of the North African (deMenocal et al., 2000), Asian  While it is clear that the onset of the Holocene was extremely abrupt and exhibited global influence, another abrupt change in the Asian monsoon domain between 5.0-4.5 ka BP as described by Morrill et al. (2003) is still unclear in terms of its spatiotemporal scope and mechanism. The event occurred at a time between Bond events 3 and 4 (see the following section), when the Asian monsoon was waning along a summer 5 insolation decrease trend at the precession band. Actually, we can see similar abrupt changes often occurred under the same insolation condition when looking deep into the Asian monsoon history as documented in sepleothem records. For example, at the end of MIS 5e is an exceedingly rapid shift (e.g., Yuan et al., 2004;Kelly et al., 2006;. However, new high-resolution and absolute-dated speleothem 10 (e.g., Y. Cai et al., 2010b) and synthesized lake (Zhang et al., 2011) records do not corroborate this event. One possibility in explaining this disagreement is that the climate events in the Asian monsoon domain around this time are not representative of a single synchronous event. For instance, a recent new high-resolution record of the Indian monsoon from NE India reveals an abrupt event ca. 4.0 ka BP and, 15 in contrast, there are no abrupt events observed between 5.0-4.5 ka BP in the same record (Berkelhammer et al., 2012). Given the fact that Asian summer monsoon weakened progressively, and its fringe likely retreated southwards, during the transition from middle to late Holocene, it is unclear whether climate variations in the region are really well characterized as a single coherent event across the monsoons' geographical 20 extent.
In Africa, it has become clear that summer insolation provides an overarching control on the North African monsoon on orbital timescales, including for example during the Holocene (e.g., Kutzbach and Street-Perrott, 1985;deMenocal and Rind, 1993;Kutzbach et al., 1996). During the early and mid-Holocene epoch (∼ 11.5-5 ka BP), 25 paleohydrological data suggest that both North and East Africa experienced wetter conditions relative to today (e.g., Gasse, 2000). In North Africa, this Holocene pluvial period is often referred to as the "African Humid Period" (deMenocal et al., 2000), and during the period, the Sahara became green with lakes and a rich vertebrate fauna in a region that is now desert. Pollen evidence indicates that the Saharan desert at this time was transformed into an open grass savannah, dotted with shrub and tree species that today grow hundreds of kilometers to the South (e.g., Ritchie et al., 1985;Adams 1997;Jolly et al., 1998;Kröpelin et al., 2008) (Fig. 10). Nevertheless, the suddenness of the end of the African Humid Period nearby 5 ka BP is still a matter of debate (deMenocal 5 et al., 2000;Adkins et al., 2006;Renssen et al., 2006;Liu et al., , 2007Kröpelin et al., 2008;Brovkin and Claussen 2008;McGee et al., 2013). The marine sediment record (Hole 658C) off northwestern Africa provides a classic record that documents an abrupt increase in terrigenous (eolian) concentration at ∼ 5.5 ka BP (deMenocal et al., 2000). This event has been corroborated recently by 10 additional marine records along the northwest African margin with a revised age of 4.9 ka BP (McGee et al., 2013). Together, these records suggest that the end of the African Humid Period was rather abrupt and potentially linked to the collapse of the Saharan savannah (Claussen et al., 1999;deMenocal et al., 2000), although it remains unclear whether the suddenness was simply a direct response to an abrupt event in 15 the monsoon precipitation or a nonlinear response to a regional vegetation threshold which ultimately was reached due to the gradual waning of the monsoon.
The precipitation record from the marine sediment core in the Gulf of Guinea (MD03-270), an indicator of relative changes in the outflow of the Niger and Sanaga rivers, shows only a minor event at about 5 ka BP (Weldeab et al., 2007). The results from 20 examination of North Atlantic marine sediments also indicate a gradual change in the coastal upwelling and accompanying SST around the time (Adkins et al., 2006). There is currently no clear evidence of an abrupt collapse in large-scale North African monsoon precipitation around 5 ka BP, suggesting an ecosystem threshold mechanism: a sudden transition from grassland to desert despite a relatively gradual change in rain- 25 fall. Furthermore, on the basis of paleo-environmental reconstructions from Lake Yoa in northern Chad, Kröpelin et al. (2008) argue that the end of the African Humid Period is a gradual shift, rather than an abrupt termination, which is contradictory to the abrupt dust flux increase documented in marine records along the northwest African margin. It thus appears that at least one of proxy records, the Atlantic marine sediment core or the Yoa record, may not be representative of the hydrological history throughout broader North Africa. Notwithstanding this ambiguity, it is likely that the summer monsoon northern fringes in North African and Asia retreated southward hundreds of kilometers from the mid-5 early to late Holocene (Winkler and Wang 1993;Adams 1997;Jiang and Liu et al., 2007) (Fig. 10). In other words, a substantial part of the land in the region that is currently arid or semi-arid was once "greener" during the early to mid-Holocene and these previously monsoonal fringe belts experienced a transition from relatively wet conditions to today's arid state. Still, it remains unclear if this transition is characterized by a gradual progressive shift or by an abrupt jump, and what the associated impacts may have been in regions where the modern summer monsoon prevails. A recent cave record, likely from the northern fringe region where the Asian monsoon withdrew, is the Kesang record from western China, which remarkably reveals a Holocene hydroclimate pattern very similar to records established from the modern Asian monsoon do-15 main (Fig. 10) and might potentially provide new insights into the transition history from the middle to late Holocene (Cheng et al., 2012b). However, additional high-resolution and absolutely-dated paleohydrological records are critically needed to expand their spatiotemporal coverage in order to definitively establish the monsoon zonal migration transition time in broader Holocene. 20
In monsoon regions, the most prominent and widespread climate event during the Holocene is perhaps the 8.2 ka event (e.g., Cheng et al., 2009b;. The timing and structure of the event has been well documented in Asian (i.e., Hoti and Qunf records, Oman and Heshang-Dongge records, China) and South American monsoon records (i.e., Paixão, Padre Lapa Grande records) (X. Cheng 15 et al., 2009b;Stríkis et al., 2011) (Fig. 11). Similar to the millennial-scale events during the last glacial period, the 8.2 ka event manifests as a weak Asian monsoon and strong South American monsoon event. Many other events, such as for example the Bond events, are likely to have had spatial extents larger than previously believed in the monsoon regions (e.g., Gupta et al., 2003;Stríkis et al., 2011;20 Cheng et al., 2012a). However, the amplitudes of the events are most likely smaller than the 8.2 ka event and, thus, potential noise associated with climatic proxies may be relatively high in comparison with the climate signals from the events themself. As such, additional high-resolution, accurately dated climate proxy records are required to characterize and understand such events. For instance, a high-resolution stalagmite Introduction  Bianchi and McCave, 1999;Bond et al., 1997Bond et al., , 2001Wanner et al., , 2011. This periodicity is similar to the observed millennial oscillation during the last glacial period (Rahmstorf, 2003). However, the mechanisms underlying this periodic change at an interval of about 1500 years remain unclear.
Although the spatial patterns of the centennial monsoon variability are likely to be 5 complex, a number of events during the past two millennia appear to have exhibited an apparent anti-phased relationship across hemispheres. For example, the LIA has been shown to be drier in the Asian monsoon region and wetter in the South American and southern Asian-Australian monsoon regions (e.g., Newton et al., 2006;Zhang et al., 2008;Reuter et al., 2009;Vuille et al., 2012;Cheng et al., 2012a). As noticed in Cheng and Edwards (2012), some of these events may have had considerable socio-economic consequences (Fig. 12). For example, in the late 9th Century, the Mayans and Tang Chinese societies both faced more arid conditions (Yancheva et al., 2007). The Vikings settled Greenland in the 980's, during the Europe's MCA, while the Northern Song Chinese prospered, with expanded rice cultivation and a large growth in population. Also, 15 at this time, the newly established Guge Kingdom thrived. Meanwhile, in contrast, the Tiwanaku, of tropical South America faced drought (Ortloff and Kolata, 1993), consistent with the hemispheric contrast in rainfall suggested by data. If one aspect of this climatic phenomenon is a meridional shifts in the low latitude (to mid-latitude in the case of the Asian monsoon) rainbelt, as has been inferred for other times (Cheng et 20 al., 2009b;Reuter et al, 2009), then dry conditions faced by the Tiwanaku may well be part of the same phenomenon that brought abundant rainfall to the Guge Kingdom and the Northern Song Chinese: a broad northward shift of the ITCZ, with additional rain at the northern fringes of the belt, but with less rainfall on its southern fringes. Another example of this mechanism comes from the 4.2 ka event, when an arid event occurred in 25 the NH. This event is believed to be one of the most severe Holocene droughts in terms of instigating cultural upheavals. The event is likely to have initiated the southeastward habitat tracking of the Indus Valley Civilization as the Indian monsoon was waning during the event (Giosan et al., 2012). At the same time, the drought associated with the Introduction  (Wu and Liu, 2004). It is likely that the event had considerable reach across the Asian monsoon domain (Berkelhammer et al., 2012), and there is also an indication that the Southern Hemisphere experienced wet conditions during the event (Fig. 11). However, additional data and analysis will likely be needed to fur-5 ther confirm the link between climate events, human cultural shifts, and the meridional shifts of monsoonal rainfall on centennial to decadal timescales. In summary, the largest amplitude and most abrupt millennial oscillations appear to have been centered near Greenland and the North Atlantic Ocean. In contrast, small, gradual, and approximately out of phase variability is apparent in Antarctic 10 records. In the context of available data, it appears likely that low-latitude monsoon variability on a millennial timescale is manifested primarily by a broad anti-phased relationship between the hemispheres. It does not appear impossible that centennialmultidecadal monsoon variations, at least some of major events, concurred during the Holocene with anti-phased relation between two Hemispheres as well. These obser- 15 vations clearly demonstrate that the dominant variability of regional monsoons on suborbital timescales is coherent on a planetary scale.
6 Global monsoon at orbital time scale

Orbital forcing and insolation
Orbital forcing exerts a major influence on the GM. The monsoon itself is a conse-20 quence of the Earth's orbit and tilt, and as Earth's orbit has varied throughout geological history low frequency changes in the GM have occurred. Indeed orbital variability was the primary motivating factor behind many of earliest paleo-monsoon studies (e.g. Kutzbach, 1981).
The solar insolation received at the Earth's surface is determined by three orbital pa- and eccentricity (100 000 and 400 000 years). According to geological records and numerical modeling, low-latitude monsoon variations are mainly caused by changes in precession (Molfino and McIntyre, 1990;McIntyre and Molfino, 1996;Short et al., 1991), but tropical solar insolation varies not only on the 20 ka cycle of precession, but also strongly on the 100 and 400 ka eccentricity and 10 ka semi-precession cy-5 cles (Berger and Loutre, 1997;Berger et al., 2006). The precessional cycle does not change insolation's annual total, but rather influences its seasonality and, in turn, the GMs, significantly. Since seasonal variations are out of phase between hemispheres, the insolation changes associated with precession are also out of phase, resulting in hemispheric contrasts in paleomonsoon records at the precessional time scale.
Increasingly, proxy evidence has revealed that orbital cycles have influenced the GM throughout the Phanerozoic. Orbital cycles in monsoon proxy records have been widely reported from the Paleozoic and Mesozoic deposits (e.g., Armstrong et al., 2009;Wilde et al., 1991;De Vleeschouwer et al., 2012;Vollmer et al., 2008;Floegel et al., 2005). Although a global assessment of the Pre-Quaternary paleomonsoon is hampered by 15 a scarcity of data, the orbital forcing in monsoon variability is global in nature. Of particular significance is the cyclic occurrence of African monsoon induced sapropel layers in the Mediterranean Sea. The precession-paced sapropel and carbonate cycles have been used to construct astronomical time scale for the Neogene period in the global geochronology (Lourens et al., 2004). Apparently the most significant advances 20 in the past two decades have been in understanding the GM during the Quaternary. Thanks to extensive works on loess-soil sequences (Guo et al., 1996(Guo et al., , 2000, stalagmites Cheng et al., 2012b;Yuan et al., 2004), lacustrine (e.g. An et al., 2011), and marine records (e.g. Caley et al., 2011;Clemens et al., 1991;Clemens and Prell, 2003), a particularly complete record of the paleo-monsoon in Asia is available, as summarized by the SCOR/IMAGES Working Group (P. . Here, variability in all regional monsoon systems is addressed.

Precession and inter-hemispheric contrasts
The precession forcing of the GM is most evident in proxy records of the Holocene, given the richness of geological data archives and particularly the high-resolution cave and deep-sea records. Stalagmites from East Asia exhibit a long-term trend which is broadly similar to changes in summer insolation, with a general warm/wet period in NH 5 monsoon regions from approximately 9.0-5.0 ka BP, during the so-called Holocene Climatic Optimum (HCO) (Fig. 13a). Speleothem records from Qunf cave, southern Oman (Fleitmann et al., 2003;Fig. 13b), Timta cave, northern India (Sinha et al., 2005) and Tianmen cave, Tibet (Cai et al., 2012), suggest that the Indian monsoon has varied in concert with the East Asian monsoon during the Holocene . The basic structures of variability for the East Asian and Indian monsoons are essentially similar, characterized by a broadly coherent pattern with a rather gradual long-term change that follows NH summer insolation (e.g., Yuan et al. 2004;Cheng et al., 2009bCheng et al., , 2012bCai et al., 2010aCai et al., , 2012Zhang et al., 2011). A similar pattern has also been observed in marine proxy records of the North African monsoon 15 (Weldeab et al., 2007;deMenocal et al., 2000;Fig. 13e, f).
Recently, a number of speleothem records from tropical-subtropical South America have demonstrated that Holocene variations in the South American monsoon also visually track changes in SH summer insolation (Cruz et al., 2005(Cruz et al., , 2007; X. Wang et al., , 2007avan Breukelen et al., 2008;Cheng et al., 2013) and thus, exhibit 20 approximately an interhemispheric anti-phasing relationship with the Asian monsoon ( Fig. 13g-i).
The Australian (aka. Austral-Indonesian) monsoon system is sometimes considered a part of the Asian monsoon (Beaufort et al., 2010) due to its link with the modern Asian monsoon system (Trenberth et al., 2000;B. Wang et al., 2003). Long-term continuous 25 geological records for the Australian summer monsoon however are scarse. Recently, multi-proxy analysis of a core retrieved in the Eastern Banda Sea has provided some insight into the Australian monsoon history for the past 150 ka (Fig. 14c; Beaufort et Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | al., 2010). These coccolith and pollen assemblages show that the primary production in the Banda Sea and length of the dry season in northern Australia and southeastern Indonesia, which primarily influenced by the winter monsoon, vary on precessional frequencies and are generally correlated with Asian summer monsoon records (Fig. 14a).
The monsoon systems are a key component of warm season precipitation in North 5 and South America (Vera et al., 2006), yet monsoon research in the Americas generally began much later than for the Asian and African monsoons. Although the climate impact of modern day ITCZ shifts in the Americas has been a topic of research for decades, paleo-monsoon studies began only in recent years. In this work, it has been shown through the analyses of lacustrine deposits in New Mexico that periods of en-  thermal maximum followed by a trend toward drier conditions since ∼ 5 ka has also been inferred from marine sediments in the Cariaco Basin (Haug et al., 2001) and Gulf of Mexico (Poore et al., 2003). Better studied is the late Quaternary history of the South American monsoon whose changes have been documented in ice cores (Thompson et al., 1998), pollen 20 (Pessenda et al., 2004), lake sedimentary and hydrologic data (Rowe et al., 2003). A δ 18 O time series of calcite from Lake Junin with a well-constrained chronology has provided a record of hydrologic variability that spans the last glacial-interglacial transition in the southern tropics (Seltzer et al., 2000).  Cheng et al., 2013) and thus, exhibiting a marked interhemispheric out of phase relationship with the Asian monsoon (Fig. 15). Africa is presently the only continent that is divided by the equator into two nearly equal parts, resulting in distinct monsoon systems in each hemisphere. Rich geological archives of the North Africa monsoon has been recovered from the East African Rift 5 lakes (e.g., Gasse et al., 2008), the Mediterranean Sea (e.g., Ziegler et al., 2010), the North Atlantic (e.g., Pokras and Mix, 1987;Weldeab et al., 2007), as well as caves (Bar-Matthews et al., 2003). As a result, the history of the North African monsoon is better resolved by proxy data than its southern counterpart. As mentioned above, sapropel layers formed in the Mediterranean as a response to the Nile River flood induced by a strengthened North African monsoon, and the cyclic occurrences of sapropel suggest a strong role for precessional forcing of the monsoon (e.g., Rossignol-Strick, 1983;Rossignol-Strick et al., 1998;Ziegler et al., 2010). A recent record based on relative iron content from the Mediterranean corroborates the suggested precessional control (Revel et al., 2010). 15 In the North Atlantic, an extended marine record shows that the North African monsoon is marked by strong signals of the ∼ 20 ka frequency of precession (Fig. 13f;DeMenocal, 1995DeMenocal, , 2000. Strong ∼ 20 ka precessional signals were also found in the Ba / Ca ratio sequence of planktonic foraminifera from the Gulf of Guinea, indicative of West African monsoon hydrology of the last glacial-interglacial cycle (Fig. 13e;20 Weldeab et al., 2007). Speleothem δ 18 O records from Israel and Lebanon have also been linked to the North African monsoon change, as they, to first order, track surface seawater δ 18 O changes in the Eastern Mediterranean Sea (Bar- Matthews et al., 2003), and the latter is influenced strongly by river discharge from the North African monsoon region. These speleothem records show a predominant Holocene pattern similar 25 to others observed in the North African monsoon region (Bar- Matthews et al., 2003;Verheyden et al., 2008). Recently, a correlation has been revealed between the Eastern Mediterranean record (color reflectance from ODP 968 core) and Asian monsoon record from the Masoko maar (southern Tanzania) lake reveals that the variability of deposition is strongly influenced by the precessional cycle and its harmonics (Garcin et al., 2006). Notable is the 200 kyr record from a crater lake on the interior plateau of South Africa. The summer rainfall proxy shows ∼ 20 ka signals linked with the precessional band adjacent to the North African record (Fig. 16;Partridge et al., 1997;Gasse, 2000). Recently, geochemical records offshore SE Africa have been used to document a Holocene precipitation pattern out of phase with its North African counterparts (Ziegler et al., 2013). In combination with the North African monsoon records, these data support the strong influence of low-latitude interhemispheric contrasts in insolation on the monsoon circulations that are broadly characterized by an anti-phase 15 relationships between the subtropical South and North African monsoons.

Eccentricity modulation
The recent availability of high-resolution long-term records spanning over one million years has enabled the identification of longer orbital periodicities in monsoon variatiability. As mentioned above (Sect. 3.3.2), the 400 ka long eccentricity is of particular im-20 portance for paleo-monsoon studies. Eccentricity influences the climate system mainly through its modulation of the amplitude of climate precession, however unlike for precession, eccentricity forcing exhibits no interhemispheric contrast. The long eccentricity cycle far exceeds the glacial cycle in duration and modulates the monsoon variations on a 10 5 years scale. By controlling the weathering rate and other processes, these 25 low frequency monsoon cycles also lead to periodic changes in the oceanic carbon reservoir. Since the residence time of carbon in the ocean is much longer than 100 ka (Kump, 1991;Katz et al., 2005), the 400 ka period of the monsoon is most evident 2207 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | in the inorganic δ 13 C and carbonate reservoirs, representing a key mode of monsoon variability at orbital time scales, as supported by recent modeling experiments (Russon et al., 2010;Ma et al., 2011). In spite of the Earth currently experiencing its long eccentricity minimum, the 400 kyr rhythm has been largely overlooked in Quaternary paleoclimatology, because of its 5 obscuring since 1.6 Ma (P. Wang et al., 2010Wang et al., , 2014 and the inadequate length of most available proxy records. Nevertheless, long eccentricity of 400 ka is the most stable orbital parameter throughout the geological history (Berger et al., 1992;Matthews and Froelich, 2002), and recognition of their influence has increased remarkably over the last decade. Now, these long eccentricity cycles have been extensively documented in geochemical and lithological analyses of clay lake deposits revel precession, obliquity and eccentricity cycles in Triassic monsoon precipitation (Vollmer et al., 2008); and Cretaceous limestone and marlstone couplets have been shown to indicate precessional rhythms of monsoon-related hydrological cycling (Floegel et al., 2005). Although a global vision of the Pre-Quaternary paleomonsoon is hampered in most cases by a 5 scarcity of data, the orbital forcing in monsoon variations is widely documented.

Other controlling factors
Thus far, we have described the precessional cycles and hemispheric asymmetries that characterize orbital forcing of the GM, demonstrating the global nature of orbital periodicities and confirming the ubiquity of responses across the regional monsoons. 10 However, the variations of regional monsoons are not dictated by solar insolation alone, but also depend strongly on geographical boundary conditions. The following is an introduction to three major controlling factors that act to modulate the influence of orbital forcing including inter-hemispheric, high-latitude, and oceanic factors. The inter-hemispheric factor is most significant in equatorial regions where cross-15 equatorial exchanges are strong. A prime example of such a region is North Africa, where lacustrine deposit sequences south of the Equator nonetheless are coherent with insolation variations in the NH. It was discovered in Lake Tanganyika (3.5-9 • S) 25 years ago, that monsoon-driven lake level fluctuations during the past 26 kyr were in phase with those north of the Equator at orbital timescales (Gasse et al., 1989). 20 Temperature and precipitation proxy records spanning the last 60 kyr from the same lake have been found to be influenced primarily by changes in Indian Ocean SST and the winter Indian monsoon rather than by ITCZ migration (Fig. 17;Tierney et al., 2008). The High-latitude factor is largely related to the influence of the boreal ice-sheet. It is perhaps expected that the existence of huge polar ice-sheets impacted low-latitude 25 climate in the Late Quaternary. As seen from Figs. 16 and 14b, the North and South African monsoons varied in response to precessional forcing, but the 20 kyr periodicity is blurred after ca. 50 ka, perhaps due to the growth of the Arctic ice-sheet (Gasse, 2209(Gasse, 2000. The influence of glacial cycles is documented even clearer in the loess-paleosoil sequences of China, where pedogenic intensity and chemical weathering of paleo-soil are largely dependent on summer rainfall . The chemical weathering indexes (such as the Fed/Fet) display clear signals of periodicity related to all orbit parameters, but the record shows a board similarity to the marine δ 18 O record, the proxy of global 5 ice-volume changes, with generally strong/weak monsoons correspondent to interglacial/glacial periods (Guo et al., 2000(Guo et al., , 1998. This pattern is consistent with the lowlatitude vegetation changes that are closely related to the monsoons (Zheng and Lei, 1999). An Oceanic factor influencing monsoon variability is perhaps also expected because 10 monsoon is generated in most instances by a land-sea thermal contrast. Modeling studies show that the responses of the monsoons to insolation variability and oceanic feedbacks differ substantially by region, with an oceanic influence that is particularly strong for the Australian monsoon Wyrwoll et al., 2007). Consequently, both sea level and solar intensity in the monsoon regions are important for the intensity and 15 southward extent of the Australian monsoon (Marshall and Lynch, 2008). For example, monsoon precipitation inferred from stalagmites in southeast Indonesia (∼ 10 • S) basically follows the precession forced SH insolation, but it unexpectedly increased from 11 000 to 7000 years ago, when the Indonesian continental shelf was flooded by global sea-level rise (Griffiths et al., 2009). As mentioned earlier, oceanic factor also impact 20 the long eccentricity cycles of GM (see Sect. 3.2.2). In reality, however, these three factors may collectively interact to broaden the spectrum of monsoon changes from an otherwise regular 20 ka rhythm. An example of records exhibiting this complex spectrum is the lake-level history of Lake Eyre, central Australia (Magee et al., 2004). A continuous record of the Australian summer monsoon 25 for the past 150 ka was reconstructed by the lake-level history of Lake Eyre with a wellcalibrated chronology (Magee et al., 2004). Lake level changes at the precession band have been shown to sometimes match southern summer insolation changes, as is consistent with the modeling results (Wyrwoll et al., 2012). However, lake levels were also significantly higher during interglacials than during glacials. These clear signals in the Australian summer monsoon are also attributable to the cross-equatorial influence of the Asian winter monsoon that reinforces the Australian summer monsoon and is also consistent with the impact of sea-level changes (Griffiths et al., 2009). Longer geological records, although discontinuous, also show monsoon-induced lake level maxima 5 during interglacial periods throughout the past 300 ka (Bowler et al., 2001).

Summary
Some basic features shared by the major regional monsoon systems emerge from the above review including: (1) ∼ 20 ka signals related to the orbital precession are common across almost all the monsoon geological records despite their different geographical 10 locations; (2) monsoon changes at the ∼ 20 ka precession-related band are generally out of phase between the hemispheres (within the accuracy of their chronology) but the precession forcing is modulated by 400 ka long eccentricity which shows no hemispheric contrast; and (3) the regional monsoons are influenced by factors other than precession forcing. Many geological records show the existence of both 100 and 40 ka 15 signals that are essentially synchronous in the Southern and Northern Hemispheres, with stronger monsoons during the interglacials than for the glacial periods.
Our understanding of the orbital forcing of monsoon climate, however, is based on records that are heavily biased towards the Quaternary and to the orbital cycles at 10 4 years timescale due to the scarcity of monsoon proxy sequences longer than 1 Ma.

20
The longest records of monsoon history are largely restricted to the Asian sector. These limitations of paleomonsoon reconstructions hampers a broader understanding the influence of orbital variations in different climates, such as for example in a world devoid of large ice-sheets or with high greenhouse gas concentrations. As a consequence, little is known about the low frequency processes that modulate 10 4 years cycles. 25 One efficient way to extend monsoon records deeper in time is to conduct highresolution analyses of deep-sea sediments. By using techniques such as magnetic measurements and X-ray fluorescence core scanning, a broad spectrum of 2211 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | periodicities has been found in Neogene records of African and Asian monsoon, ranging from semi-precession to long eccentricity timescales (Larrasoaña et al., 2003;Tian et al., 2011). Most promising are records from deep-seas where sedimentation rate is high, such as around large river estuaries. Long-term South American monsoon proxies, for example, may be reconstructed by analyzing the marine deposits outside the 5 Amazon River estuary.

Pre-Quaternary monsoon
The above discussions deal with modern and Quaternary monsoon variations spanning inter-annual to orbital timescales. Tectonic processes usually measure 10 6 years or longer, and monsoon history at this time scale traces back beyond the Quaternary. Pre-Quaternary paleoclimatology has advanced rapidly over recent decades, and paleomonsoon research now addressed variability spanning nearly the entire Phanerozoic. For the early Paleozoic, for example, Late Ordovician monsoon climate has been inferred from geological evidence of ITCZ migrations (Armstrong et al., 2009). Aspects 15 of the Silurian monsoon have been inferred from paleoceanographic reconstructions (Wilde et al., 1991), and the variability of Middle Devoian monsoon climate has been reconstructed using magnetic susceptibility records (De Vleeschouwer et al., 2012). For the late Mesozoic, the monsoon-driven reversal of surface tropical currents in the Cretaceous Tethys Seaway has been simulated in models (Bush, 1997), and mid-20 dle Cretaceous limestone-marlstone couplets have been interpreted as arising from precession-forced variability in monsoon precipitation (Floegel et al., 2005). Despite of the modest number of papers devoted to Pre-Quaternary monsoon, two time intervals have attracted greater attention in paleo-monsoon studies: the late Paleozoic to early Mesozoic, and the late Cenozoic. The formation of the super-continent Pangea persisting from the Permian to early Jurassic periods. It is likely that with the evolution of the land distribution into multiple continents, this unique GM system collapsed into regional subsystems. After reorganization of the land-sea distribution, the modern monsoon sub-systems were established in the late Cenozoic, as suggested by emerging evidence, discussed below. Accordingly, these two intervals will be the focus 5 of the following review of paleo-monsoon at tectonic time scale.

Super-continent and mega-monsoon
The GM global monsoon system would be uniquely simple if the Earth had only a single continent. Webster (1981) hypothesized that if all the continents were gathered around the northern pole, it would be a continental cap north of 14 • N with a nearly complete 10 aridity inland, and with monsoonal precipitation in summer along the coastal region. In fact, there were geological times when all the continents on the Earth assembled into a single continent, the last of which was termed "Pangaea", a result of the so-called "Wilson cycle" which governs the redistribution of landmasses globally across tectonic timescales. This Mega-continent generated Mega-monsoon (Kutzbach and Gallimore, 15 1989;Parrish, 1993), the most intensive monsoon system in geological history. Since enhanced monsoon circulations can drive enhanced aridity in non-monsoon regions (Rodwell and Hoskins, 1996), it is expected that a "Mega-desert" region should accompany the "mega-monsoon".
From the late Permian to Early Jurassic (∼ 250-180 Ma) all continents assembled 20 into two major landmasses, Laurasia and Gondwanaland, converged near the equator into the super-continent Pangaea, culminating at the early Triassic. Modeling results show Pangaea resulted in a Mega-monsoon of global scale with a reversal of surface winds between summer and winter (Fig. 18c, d)  with wide annual range of temperature (50 • C) in its interior (Rodwell and Hoskins, 1996). The idea of a mega-monsoon first appeared in 1973 when Pamela Robinson drew the hypothetical position of the ITCZ for the late Permian and early Triassic with a 40degree range of seasonal migration and a single extremely intense monsoon system 5 (Lamb, 1977). Although the cross-equator mega-continent provided tectonic settings for the development of the mega-monsoon, the circulation itself could penetrate into the continental interior only with the presence of a pronounced highland (Parrish, 1993). The Appalachians in America and Variscan range in Europe reached a mean altitude of 4500 and 2000-3000 m, respectively, during this period (Fluteau et al., 2001). Such 10 altitudes are lower than in Tibet today, but the presence of these mountain ranges at equatorial latitudes then were nonetheless of great importance for the monsoon development (Fig. 19;Tabor and Montañez, 2002). The Triassic was distinguished by both an intense monsoon and also by extreme aridity, and the Pangaea was the time of maximal accumulation of evaporites and eolian deposits (Gordon, 1975). In the south- 15 western United States, wind-deposited sands accumulated to a thickness of 2500 m during the 160 million years when Pangaea straddled the Equator. These strata are the thickest and most widespread aeolian dune deposits known from the entire global sedimentary record (Loope et al., 2001), and are accompanied by loess accumulation in semi-arid regions (Soreghan et al., 2008). 20 The Pangaea mega-monsoon is likely to have experienced significant variations in its intensity due to orbital forcing, as manifested in the orbital cycles in the Triassic playas and lakes. In the Mid-German Basin, Late Triassic dolomite/red mudstone beds depict strong periodicity of the playa system. These periodic changes are believed to have been associated with monsoon variability in the northern low latitudes of the supercon-25 tinent (Vollmer et al., 2008). In northeast US, the micro-lamination in the lacustrine deposits of the Newark super-group recorded the alternation of dry/humid conditions and related lake level fluctuations in the Pangaea from the Late Triassic to Early Jurassic (Olsen, 1986). Detailed studies using the 6700 m-long section have revealed a broad Introduction range of periodicities in monsoon climate over the past 33 Ma: varves with 0.2-03 mmthin couplets of alternating light (dry winter) and dark (rainy summer) layers from which seasonal contrast of monsoon climate can be inferred (Fig. 20b); thicker sediment variations representing the 20 ka precession cycles (4 m on average), 100 ka (20-25 m) and 400 ka (90 m) eccentricity cycles (Fig. 20c, d, Olsen and Kent, 1996). 5 In the Late Jurassic, the "mega-monsoon" collapsed with the break-up of Pangaea; but the spatial distribution of precipitation remains largely monsoonal throughout the Late Jurassic, replaced by a predominatly zonally symmetric distribution in the Cretaceous (Weissert and Mohr, 1996). 10 As defined in Sect. 2, the GM can be regarded as an integrated system of six regional monsoon sub-systems (Fig. 2). In the recent decades, substantial amounts of proxy data have been derived for the onset of the Asian monsoon-dominated climate (referred hereafter as to Asian monsoon climate) for both temporal (Guo et al., 2002(Guo et al., , 2008 and spatial (Sun and Wang, 2005) perspectives, while the onsets of the other monsoonal 15 sub-systems are more poorly known, mainly because of the lack of pertinent geological evidence with well-constrained chronology.

Establishment of the Asian monsoon system
Two prominent features characterize the modern Asian environment: the moist southern regions, which are primarily under the influence of the southwest (South Asian) Introduction inated climate in Asia. The evidence for this conclusion was mainly derived from the southern side of the Himalayas. A record of planktonic foraminifera from the Arabian Sea revealed strong wind-induced upwelling since the late Miocene at ∼ 8 Ma and this has been interpreted as an indication of the onset or strengthening of the Indian Ocean (South Asian) monsoon (Kroon et al., 1991;France-Lanord and Derry, 1994;Singh 5 and Gupta, 2004). The expansion of plants that use C4 photosynthesis at ∼ 8 Ma in South Asia is also suggestive of a strengthening of South Asian monsoon (Quade et al., 1989). Because climate models link the intensification of the Asian monsoon with the tectonic uplift of Tibetan Plateau (Kutzbach et al., 1989(Kutzbach et al., , 1993Ruddiman and Kutzbach, 1989;Prell and Kutzbach, 1997), the 8 Ma view was also been supported by a number of tectonic studies (Harrison et al., 1992) showing some prominent tectonic changes at the southern margins of the Himalayan-Tibetan complex. However, geological records acquired from the northern side of the Tibetan Plateau over the last two decades have led to major improvements in our understanding of the precise timing of these changes.

Transition from the zonal to monsoonal climate pattern
An examination of the spatial distribution of geological indicators in China has revealed the transformation of the dry areas in the Cenozoic from a zonally symmetric distribution across China to a region restricted to northwestern China (P. Wang, 1990). This shift is confirmed by a more detailed mapping of geological data (Liu and Guo,20 1997), which shows the transition from a roughly zonal symmetric pattern of aridity during most of the Paleogene, to the modern monsoonal pattern, probably during the Oligocene or Miocene. In carefully examining the chronologies and climate significance of the Cenozoic paleo-botanical and geological evidence, Sun and Wang (2005)  interpretation, and another based on lithological data. Both lines of evidence define consistent environmental pattern changes in China, indicating clearly that the climate patterns in Asia during the Paleogene were predominantly zonal while during the Neogene they were highly similar to the modern era. More detailed mapping specifically addressing the time slices within the Oligocene and Miocene (Guo et al., 2008) corroborates these observations (Fig. 22). The data also indicated that the transition from the zonal pattern to the modern monsoonal pattern of climate. The Paleogene climate pattern in Asia is actually most similar to the modern-day configuration of the African monsoon region (Fig. 21), with strong subtropical subsidence to the north of the monsoonal zone in the Sahara desert and a dry climate comparable to the modern Mediterranean regime in regions immediately to the north of the desert region (Guo, 2010). Especially, an interior sea, referred to as the Paratethys, was still in existence during the Paleogene in the far western part of China (Dercourt et al., 1993), and presumably this provided a significant amount of moisture to the eastern half of the continent, a suggestion that has recently been confirmed by numerical stimulations 15 on the early Eocene climate  showing a Mediterranean-like climate for Central Asia in the Eocene.
The climate patterns for the Neogene are radically different from those in the Paleogene, but highly similar in shape to those of the modern monsoon-dominated era. The Paleogene dry belt in southern China disappeared and was replaced by moister 20 conditions. In contrast to the Paleogene, arid conditions are now primarily observed for northwest China while the middle reaches of the Yellow River, including the Loess Plateau, were dominated by semi-arid conditions. The patterns for the early, mid-and late-Miocene are essentially similar, suggesting that they have existed at least since the early Miocene. The Pliocene pattern is essentially similar to that of the Miocene monsoon at 8 Ma. However, much longer aeolian sequences, back to 22 Ma, have been found and dated over the last decade (Guo et al., 2002;Guo et al., 2008;Hao and Guo, 2007). The aeolian origin of these deposits has been well documented by various lines of evidence (Guo et al., 2002, Hao and Guo 2004Li et al., 2006;Liang et al., 2009;Liu et al., 2005Oldfield and Bloemendal, 2011;Qiao et 10 al., 2006). Discontinuous aeolian portions likely dating to 24- 25 Ma (Sun et al., 2010;Qiang et al., 2011) have also been reported. These new sequences, combined with the previously reported Quaternary loess-soil sequences and the underlying Red Clay, offer a unique aeolian continental record of paleoclimate.
Because eolian deflation only occurs in areas with poor vegetation cover (Pye, 1995;15 Tsoar and Pye, 1987;Sima et al., 2009), the thick and widespread Miocene aeolian deposits of China firmly attest to the existence of sizeable deserts in the Asian inlands by the early Miocene as dust sources (Guo et al., 2002(Guo et al., , 2008. The existence of these deserts is also supported by the Quartz morphology  and geochemistry signatures (Liang et al., 2009). The near-continuous record of aeolian sequences in 20 northern China, from the early Miocene to the Holocene, implies that inland deserts have been constantly maintained at least for the past 22 Ma (Guo et al., 2008), and that a northerly winter circulation, which is a key indicator of the Asian monsoon system (Liu and Yin, 2002), was already established by the early Miocene. The Neogene aeolian deposits contain more than 400 visually definable paleosols 25 (Guo et al., 2002;Hao and Guo, 2004). They are mostly luvisol (FAO-Unesco 1974) formed under humid forest environments (Guo et al., 2008), requiring a substantial amount of rainfall (Fedoroff and Goldberg, 1982). These Miocene paleosols imply the Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | existence of another circulation bringing moisture from the ocean. Clearly, the dust carrying winds and moisture carrying winds must be both independent and seasonal. If so, these are strongly suggestive of a monsoonal circulation regime. Pedological, sedimentological and geochemical approaches also demonstrate that the paleosols in the Miocene aeolian deposits are accretionary soils, resulting from the 5 interactions between the summer and winter monsoon. In summer, monsoonal rainfall associated with high temperatures favors pedogenesis, but eolian dust continues to be added to the soil surface in winter and early spring, although at lower intensities (Guo et al 2008). These soils significantly differ from the paleosols in loess of non-monsoon regions where soil largely represents a sedimentary hiatus (Cremaschi et al., 1990;Fe-10 doroff and Goldberg, 1982). Consequently, the accretionary properties of paleosols in the early Miocene loess can be regarded as strong evidence of a seasonally alternating circulations, and hence a monsoonal climate.
The aeolian deposits in China and paleo-environmental mappings thus provide consistent evidence with regards to the establishment of the Asian monsoon climate. 15 These depict a new understanding of Cenozoic climate changes in Asia that contrast with the 8 Ma view. A number of other geological records corroborate aspects of this new understanding near the Oligocene-Miocene boundary. For example, a slight decrease in the content of xerophytes at ∼ 23 Ma in a core from the Qaidam basin  is believed to be due to the influence of the summer monsoon. 20 The earliest high δ 13 C peaks appeared ∼ 20 Ma ago in a carbon isotope record of terrestrial black carbon have also been interpreted as an indication of early monsoon initiation (Jia et al., 2003). A prominent change in the mammalian and floristic regions in China have also been found to have occurred in the early Miocene (Jia et al., 2003;Qiu and Li, 2005;Song et al., 1983) and sudden increases in aeolian dust accumulation 25 rates have been inferred at ∼ 25 Ma (Rea, 1994). A comprehensive geochemical analysis also shows a major increase in the delivery of Asian dust material since ∼ 20 Ma in central Pacific , which has been interpreted as indicating the development of East Asian monsoon and formation of Asian loess. More recently, a Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | study of Arabian Sea proxies showed no significant changes in wind-driving upwelling at 8 Ma, and thus does not support the initiation of an enhanced summer monsoon in the late Miocene (Huang et al., 2007). Weathering records in marine sediments also traced the existence of the Asian monsoon back to the early Miocene followed by a gradual weakening since ∼ 10 Ma (Clift et al., 2010). gards to this issue. They showed that the initially arid southwest and southeast China regions in the Paleogene were both transformed into much more humid climates in the early Miocene, a shift which supports a notion of synchronous onset/strengthening of the two Asian summer monsoon systems. The Neogene monsoonal climate pattern in Asia differs radically from the other mon-15 soon regions. The geographical reach of the other regional monsoon systems, except the Australian one, tends to be restricted by the southern and northern fronts of ITCZ, as is particularly true for the African monsoon. In contrast, the front of the East Asian monsoon can penetrate deep into Asia, and in summer extend to regions where northern hemispheric westerlies prevail in a mid-latitude zone well beyond the ITCZ (Fig. 22). 20 This suggests that the East Asian summer monsoon circulation is able to break through the subtropical high-pressure belt, which usually acts as a barrier to moisture. This feature helps to explain the desert distribution pattern (Fig. 11) whereas in other continents deserts are located at the sub-tropical latitudes, while in the Asian interior they exist at much higher latitudes. The South American monsoon and Australian monsoons exhibit 25 similar features, but with a lesser extent. However, the Paleogene climate pattern in Asia is quite similar to today's African climates. The zonally oriented aridity belt is similar to that in the modern Sahara desert 2220 Introduction while the regions south of the Sahara are mainly under the influence of the African monsoon. The low latitude modern African monsoon is similar to southern-most China in the Paleogene where a tropical monsoon was present but did not extend far into the continent. In the region north of the aridity belt, a Mediterranean-like climate likely exists (Guo, 2010;Zhang et al., 2012), as is currently is the case for the North African 5 continent.
To date, little is known about the origin of the African monsoon mainly because of a lack of relevant long-term geological records. The longest records in the African monsoon region have been derived from eolian dust and sapropel deposits in the ocean cores, but the monsoon climate likely predates the earliest of these records. The oldest 10 monsoon-induced dust was dated to 11 Ma (De Menocal, 1995), and Mediterranean sapropel extends back only to the middle Miocene when the Mediterranean took on much of its present configuration (Kidd et al., 1978;Cramp and O'Sallivan, 1999). Although the oldest sapropel reported so far has dated to 13.8 Ma (Mourik et al., 2010), deposits showing humidity variations with a clear precessional periodicity have been 15 traced back to 15Ma and beyond .
Climate models suggest that that the African monsoon and its associated dry regions may have existed in the Oligocene (Fluteau et al., 1999). However, the age of the Sahara has been traced back only to the late Miocene at ∼ 7 Ma (Schuster et al., 2006). Confirmation of the model simulations is therefore hamped by the brevity of the 20 geological record.
Similar model-data differences also exist for the Australian region. Climate model suggested that a weaker-than-present Australian monsoon was present during the Miocene (Herold et al., 2011a). However, the modern day aridity of Australia has been traced back only as far as the Pliocene (Fujioka and Chappell, 2010), before which 25 time geological records are lacking. It should be mentioned that relatively arid conditions do not necessarily lead to the formation of deserts, which occur when some aridity threshold is exceeded, which itself can be modulated by other factors. For example, increases in the northern high-latitude ice sheets during the mid-Pliocene may have led Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | to enhanced aridity in the Asian interior and been instrumental in the desertification of the region, independent of the monsoon that existed on the continent's southern fringe (Guo et al., 2004).
In summary, our understanding of origin of the Asian monsoon system has benefited greatly from a wealth of bio-geochemical evidence sampling both the spatial (P. Wang, 5 1990;Liu and Guo, 1997;Sun and Wang, 2005;Guo et al., 2008) and temporal (Guo et al., 2002(Guo et al., , 2008 structure of the region's climate. Climate models (e.g., Herold et al., 2011) have also added important insight. In contrast, the proxy record and associated modeling analysis of the origin of other present-day monsoon systems remains scarce. Progress will require additional collaborative efforts aimed at documenting and 10 simulating the historical evolution of these regional monsoon climates. However, our discussions in this section reinforce several lines of insights about the Cenozoic longterm changes of GM.
1. The geographical reach of the various regional monsoons are quite different, ranging between the extremes of far-reaching the East Asian monsoon and the rela- 15 tively limited reach of the African monsoon. The first dominates a wide region at the mid-latitudes, well beyond the ITCZ, while the second primarily affects the lowlatitudes within the seasonal oscillation ranges of the ITCZ. The other monsoon sub-systems could be approximately regarded as the intermediates of these two end members, though the south Asian monsoon has considerably zonal influence 20 across the tropics. These distinctions are also clear for the orbital-scale aspects of the regional monsoons, as also discussed in Sect. 6.
2. The above features imply contrasting links between the monsoons and the broader atmospheric circulation, and proxy data support the possibility of differing times of initiation for the different regional monsoons. These contrasts are integral 25 considerations in addressing the GM history and related forcing mechanisms. It was hypothesized that the African monsoon and aridity might be traced back to Introduction very earlier history of the Earth, depending on the timing when the African continent drifted to subtropical latitudes (Guo et al., 2008).

The importance and complexity of monsoon projection
Foremost amongst the impacts of a changing climate are changes in rainfall, drought, 5 and associated climate extremes in highly populated and agriculturally productive regions. Anticipated trends in the monsoon domains are complex, varying as a function of season and region (Lee and Wang, 2014), and in instances, varying in a potentially nonlinear fashion with respect to global mean temperature (e.g. Cook et al., 2010). The challenge in projecting future impacts also relates in part to the compensating nature 10 of future changes, particularly over land, and thus the net future change is often the residual of larger competing influences including increases in both evaporation and precipitation.
The GM concept is likely to be useful in this context, given the consistency of monsoon responses to past external forcing (Y. , and the relevance of 15 fundamental constraints across monsoon domains, such as land-ocean portioning of moisture (Christensen et al., 2007;Fasullo, 2012;, and interhemispheric contrasts (Lee and Wang, 2014).
Considerable insight into future monsoon shifts in a warming world can also be gained from consideration of past climate. A particularly useful analogue for the future 20 monsoon is the Paleocene-Eocene Thermal Maximum event about 56 Ma ago, when 1500-4500 gigatons of carbon was released within < 20 ka, resulting in rapid global warming of 5-8 • C (Bowen et al., 2006;McInerney and Wing, 2011). This global warming was accompanied by drastic changes in hydrological cycling, with enhanced monsoon rain and increased seasonality in precipitation (Foreman et al., 2012). Although 25 our understanding of the mechanisms behind these changes is insufficient to defini-2223 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tively project future changes, similar past greenhouse events may provide a glimpse of the future (Zachos et al., 2008).

Projected changes in the global monsoon using GCMs
A complementary approach for projecting future monsoon changes is with coupled ocean-atmosphere models and newly available Earth System Models (ESMs), which 5 also include interactive chemistry, an active carbon cycle, and biogeochemistry. In recent years, the array of global models produced by individual modeling centers has been aggregated into multi-model archives (e.g. Meehl et al., 2007a;Taylor et al., 2012) in order to facilitate their inter-comparison and combined consideration of historical simulations and future projections.

10
One approach for evaluating future monsoon simulations is to select only those projections from models that best depict monsoons in the current climate. From the 20 coupled models that participated in the phase five of Coupled Model Intercomparison Project (CMIP5), the historical run for 1850-2005 and the Representative Concentration Pathway (RCP) 4.5 run for 2006-2100 have been used to assess current fidelity 15 and future conditions (Lee and Wang, 2014). Metrics for evaluating model simulations were designed to document model performance for 1980-2005 and, based on these metrics, the four best models' multi-model ensemble (B4MME) was selected for its evaluation of future conditions, projecting the following changes in the twenty-first century. (1) Changes in monsoon precipitation exhibit huge differences between the NH and SH. The NH monsoon precipitation is expected to increase significantly due to a contrast in warming between the NH and SH, significant enhancement of the Hadley circulation during boreal summer, and atmospheric moistening, which together overcome an increase in tropospheric stability arising from greenhouse gases. There is a slight weakening of the Walker circulation in the B4MME but it is not statistically 25 significant, given the inter-model spread.
(2) The annual range of GM precipitation and the percentage of local summer rainfall will increase over most of the GM region, both over land and over ocean. (3) There will be a prominent east-west asymmetry 2224 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | in future changes with moistening of the eastern hemisphere monsoons and drying of the western hemisphere monsoon. (4) The NH monsoon onset will be advanced and its withdrawal will be delayed. (5) The land monsoon domain over Asia is projected to expand westward. There are important differences between an assessment of an earlier generation of models and these CMIP5 results. While some differences likely 5 relate to contrasts in the forcings used in future conditions for each experiment, other differences are likely to be related to advances in model physics and particularly in the inclusion of cloud-aerosol interactions in CMIP5.

Heat waves, droughts, and floods in the global monsoon environment
Chief amongst the robustly simulated changes arising from anthropogenic forcing is 10 a warming and moistening of the troposphere on a planetary scale (e.g. Meehl et al., 2007b). Warming of the globe brings with it a highly probable warming of the tropics. There is therefore a strong expectation that the frequency of heat waves will increase in the GM region (Kumar et al., 2011). Other effects, such as an increase in aridity over land in the pre-monsoon environment (Seth et al., 2011;Fasullo, 2012;Dirmeyer 15 et al., 2013), are likely to delay monsoon onset in some regions, thereby exacerbating the warming of the base state and intensifying heat waves, which tend to occur prior to monsoon onset. Increases in total atmospheric water vapor are tied strongly to warming and proportionate changes in extreme monsoon rainfall events are both anticipated (e.g. Tren-20 berth, 2011;Kumar et al., 2011) and observed (Goswami et al., 2006b;Chang et al., 2012), though regional structure in such changes is also likely (e.g. Chang et al., 2012). Relevant to coastal flooding and impacts in the monsoon zones, increases in sea level arising from warming oceans and melting of snow and ice sheets are also robustly projected (Vermeer and Rahmstorf, 2009). Potentially augmented by increases in storm 25 intensity (Unnikrishnan et al., 2011;Murakami et al., 2012), sea level rise is therefore very likely to lead to considerable societal and environmental impacts, particularly 2225 given the exceptional population densities and susceptibilities in coastal monsoon environments.
Changes in the tropical circulation arising from warming are also likely. From an energetic standpoint, the net radiative imbalance at the surface increases less quickly than atmospheric water vapor, which is governed instead by the Clausius -Clapey-5 ron relationship (∼ 2 % K −1 vs. 7 % K −1 , respectively). As a result, total evaporation, and by extension rainfall, increases more slowly than does water vapor amount, and the residence time of atmospheric moisture therefore increases (e.g. Held and Soden, 2006;Seager et al., 2010) suggesting a decrease in the intensity of the divergent atmospheric circulation (Tanaka et al., 2005;Vecchi and Soden, 2007). Increases in tropical tropospheric stability act to suppress the divergent circulation and rainfall (Chou et al., 2001;Neelin et al., 2003). Yet downscaling these changes in the global hydrologic cycle to regional monsoon environments is non-trivial and projections show substantial variability across individual monsoon environments, some of which arises from internal variability (Deser et al., 2012). Another key source of variability across models is their 15 contrasting structural representation of key processes. Understanding and constraining projections of future monsoon variability to account for these differences remains an active research topic.

The global monsoon as a climate feedback
While most studies on monsoon projection have focused on the impact of changes 20 in the global environment on the monsoon, it can equally be asked what role the monsoons have in influencing global climate. One consequence of global warming on precipitation is the "rich-gets-richer" and "poor-gets-poorer" pattern identified in multimodel projections of climate change (Neelin et al., 2006). Over the last 30 years of global warming, a similar pattern has been observed (Fig. 23a). While the precipitation 25 response to global warming is complex (Held and Soden, 2006) and not fully understood, increased atmospheric moisture and a relatively steady circulation are consistent with the observed trend and local summer monsoon precipitation bears a close 2226 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | resemblance to the trend (Fig. 23b). This arises from an increase in summer monsoon rainfall across the majority of the regional monsoon regions, except the South America. In contrast, most arid and semiarid desert or trade wind regimes, located to the west and poleward of each monsoon region, exhibit drying. Understanding the individual contributions of internal variability and forced change to this trend remains an active 5 research topic. Defined as the equilibrium response in global mean surface temperature to a doubling of carbon dioxide concentrations, climate sensitivity is a canonical measure of a changing climate. A persistent and important uncertainty surrounding the climate response to greenhouse gases is our inability to constrain climate sensitivity beyond the range of approximately 1.5 to 4.5 K. Large contributions to this uncertainty arise from inconsistencies in the modeling of the shortwave cloud feedback across models, particularly in the subtropics. Fasullo and Trenberth (2012) explore observational constraints on models and find the intensity of seasonal variations in subtropical subsidence and the related relative humidity (RH) of the free troposphere in models relate strongly 15 to their simulated climate sensitivity. These so-called "dry zones", whose origin and maintenance is explicitly linked to the global monsoon, have been proposed to be of fundamental importance in determining the shortwave cloud feedback under climate change, as they expand in a warming environment and erode the cloud field. Despite this importance, most models are systematically biased in representing them, with dry 20 zones that are too moist arising from an overturning circulation that is too weak. The suggestion therefore is that improving the representation of the global monsoon and the effects of its overturning circulation is essential to narrowing the uncertainty of the global effects of forced climate change.
9 Concluding remarks 25 In this paper, recent progress in modern and paleo-monsoon studies is reviewed in an attempt to answer the question as to whether the regional monsoons constitute a global 2227 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | system. We pool together available observational data to explore where there is coherent variability of regional monsoons at various time scales. The following conclusions can be drawn: 1. There is a certain coherence in monsoon variability across regions. Although independent variability exists, the regional monsoons exhibit mutual coherence across 5 various timescales, ranging from interannual, interdecadal, centennial and millennial, up to orbital and tectonic timescales. The available data, both from the modern observations and geological archives, support the utility of the GM concept.
2. Within the GM system, each subsystem has its own features depending on its geographic and topographic conditions, and recognition of the GM system does 10 not negate the value of regional monsoon studies. On the contrary, the GM concept helps to enhance our understanding and to improve future projection of the regional monsoons, while discriminating between the global and regional components of their variability.
3. Two groups of proxies are used for assessing monsoon variations beyond instru-15 mental records: proxies based on wind and on rain. In view of the better preservation of the precipitation signal in geological records, and the increasing role of hydrological cycle in modern climatology, the rain-based proxies offer a more promising opportunity for diagnosing monsoon variability across many timescales.
4. Historically, paleoclimatology originated from the Quaternary ice ages, and up 20 to now, paleoclimatology in general remains ice-sheet centered, in contrast with modern climatology, which focuses mainly on low-latitude processes. Growing evidence shows that the GM exists at least throughout the 600-Ma Phanerozoic Eon. A systematical study of the GM history will shed light on the role of lowlatitude processes in the climate evolution and variability.
5. With a rapidly growing number of publications dealing with paleo-monsoon, caution is required to avoid the misuse of monsoon proxies. Given the remarkable 2228 contrasts that exist in terms of the "best use" of monsoon proxies and the lack of mature proxies for the GM, it is crucial to strengthen exchanges and collaborations between the modern-and paleo-communities to better develop and calibrate monsoon proxies on the basis of modern observations.
In conclusion, the GM is a complex system of interacting processes that governs 5 climate variability across multiple timescales. The goal of this paper is restricted to demonstrate the applicability of the GM concept for understanding variability across timescales. It does not attempt to disentangle complex physical interactions within the climate system. An in-depth discussion on driving mechanisms and outstanding issues in the GM studies will therefore be the subject of a follow-on companion paper "The   Possible link between climate event and human culture (adapted from Cheng and Edwards, 2012). The Asian summer monsoon (green, Zhang et al., 2008) tracks the Alpine glacial advance and retreat (blue, Holzhauser et al., 2005), demonstrating that when temperatures were colder in Western Europe, conditions were drier in the monsoonal regions of China. The grey bars show some climate events that likely had influence on human cultural history over the past 2000 years.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Fig. 13. Holocene monsoon records from different monsoon regions. (A) The East Asian monsoon record from Dongge cave . (B) The Indian monsoon record from Qunf Cave (Fleitmann et al., 2003). (C) The Kesang record, eastern central Asia. Although it is outside of the modern Asian monsoon domain, it may be influenced by the summer monsoon incursions during the mid-Holocene (Cheng et al., 2012b;Kutzbach et al., 2014). (D) Eastern Mediterranean records from Soreq cave (brown, Bar-Matthews et al., 2003) and Jeita Cave Lebanon (blue, Verheyden et al., 2008). (E) The seawater δ 18 O record from the marine sediment core (MD03-270) (Weldeab et al., 2007). (F) The terrigenous concentration (%) record from the northwest African margin sediment (core Hole 658C, deMenocal et al., 2000). (G) The South American monsoon record from Cueva del Tigre Perdido cave, northern Peru (van Breukelen et al., 2008). (H) South American monsoon records from Botuverá cave (pink, Cruz et al. 2005; blue, X. . (I) The speleothem record from Cold Air cave, the Makapansgat Valley, South Africa (Holmgren et al., 2003). The records from Southern Hemisphere monsoon regions (G-I) are plotted inversely relative to their Northern Hemisphere counterparts (A-F). Some of locations of above records are shown in Fig. 5-2. Summer insolation (grey curves) at 65 • N (JJA, in A) and 30 • S (DJF, in G) are also plotted inversely for comparison (Berger, 1978). It appears that the abrupt fall of the North African monsoon (the grey bar) correlates with low δ 18 O values in Eastern Mediterranean records (D) and the latter is causally linked to weakening of the North African monsoon.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |   in (a) global annual precipitation and (b) precipitation in summer monsoon season (i.e., MJJAS for NH and NDJFM for SH) in units of mm day −1 decade −1 The significance of the linear trends was tested using the trend-to-noise ratio. Areas passing 90 % confidence level were stippled. The GPCP data were used. In (a) the climotological annual mean precipitation rate was shown by contours (1 (red), 2 (blue), 4, and 8 (purple) mm day −1 ). In (b) the monsoon and desert regions as defined in Fig. 2 were delineated by blue and red, respectively.