Comparing the spatial patterns of climate change in the 9th and 5th millennia BP from TRACE-21 model simulations

The spatial patterns of global temperature and precipitation changes, as well as corresponding large-scale circulation patterns during the latter part of the 9th and 5th millennia BP (4800–4500 versus 4500–4000 BP and 9200– 8800 versus 8800–8000 BP) are compared through a group of transient simulations using the Community Climate System Model version 3 (CCSM3). Both periods are characterized by significant sea surface temperature (SST) decreases over the North Atlantic, south of Iceland. Temperatures were also colder across the Northern Hemisphere but warmer in the Southern Hemisphere. Significant precipitation decreases are seen over most of the Northern Hemisphere, especially over Eurasia and the Asian monsoon regions, indicating a weaker summer monsoon. Large precipitation anomalies over northern South America and adjacent ocean regions are related to a southward displacement of the Intertropical Convergence Zone (ITCZ) in that region. Climate changes in the late 9th millennium BP (the “8.2 ka event”) are widely considered to have been caused by a large freshwater discharge into the northern Atlantic, which is confirmed in a meltwater forcing sensitivity experiment, but this was not the cause of changes occurring between the early and latter halves of the 5th millennium BP. Model simulations suggest that a combination of factors, led by long-term changes in insolation, drove a steady decline in SSTs across the North Atlantic and a reduction in the North Atlantic Meridional Overturning Circulation (AMOC), over the past 4500 years, with associated teleconnections across the globe, leading to drought in some areas. Multi-century-scale fluctuations in SSTs and AMOC strength were superimposed on this decline. This helps explain the onset of neoglaciation around 5000–4500 BP, followed by a series of neoglacial advances and retreats during recent millennia. The “4.2 ka BP Event” appears to have been one of several late Holocene multi-century fluctuations that were embedded in the long-term, low-frequency change in climate that occurred after ∼ 4.8 ka. Whether these multicentury fluctuations were a response to internal centennialscale ocean–atmosphere variability or external forcing (such as explosive volcanic eruptions and associated feedbacks) or a combination of such conditions is not known and requires further study.


Introduction
It is well documented that the first-order driver of Holocene climate change was orbital forcing, with an overall decline in summer insolation in summer months, particularly at high latitudes. This led to a drop in temperatures at high latitudes and less rainfall throughout the monsoon regions of the Northern Hemisphere, as seen in many paleoclimatic records (Burns, 2011;Solomina et al., 2015). Shorter-term rainfall fluctuations superimposed on this long-term change in hydrological conditions are clearly seen in many speleothem and lacustrine sediment records (e.g., Wang et al., 2005;Kathayat et al., 2017). Abrupt hydrological changes around 4.2 ka have been documented for various regions of the world; it has been suggested that the major global monsoon and ocean-atmosphere circulation systems were deflected or weakened synchronously at this time, causing major centuryscale precipitation disruptions (severe megadroughts) over different regions (Weiss, 2017). Other studies Tan et al., 2018a) have also noted weakening of the Asian summer monsoon at around this time, resulting in drought over the northern part of eastern China and flooding over the southern part.
In recent years, a more comprehensive picture of the "4.2 ka BP Event" has been derived from analysis of new high-resolution proxy data from different regions, and the event has become the focus of symposia and research conferences (e.g., Weiss, 2015). This event is of particular interest as it is associated with societal collapse and regional abandonment in many different regions. For example, the collapse and abandonment of Akkadian imperial settlements in the Khabur Plains, and other communities in dry farming domains across the Aegean and west Asia, was in response to the abrupt nature with which the megadrought began (with its onset in less than 5 years), its magnitude (a precipitation reduction of 30 %-50 %), and its long duration (200-300 years) (Weiss, 2017).
Although a drought episode around 4.2 ka has been found in many proxy reconstructions, the mechanisms that brought this about are still unclear, though different hypotheses have been proposed. For example, Staubwasser and Weiss (2006) suggested that the abrupt climate change event at 4.2 ka, as well as other widespread droughts around 8.2 and 5.2 ka over the eastern Mediterranean, west Asia, and the Indian subcontinent, was caused by a change in subtropical upper-level flow over the eastern Mediterranean and Asia. Some studies have suggested that these large-scale circulation anomalies may reflect persistent modes of internal climate variability, though there is a wide range of other explanations. For example, Booth et al. (2005) indicated that the widespread midlatitude and subtropical drought around 4.2 ka was linked to a La Niña-like sea surface temperature (SST) pattern, possibly associated with amplification of this spatial mode by variations in solar irradiance or volcanism. On the other hand, Hong et al. (2005) analyzed a 12 000-year proxy record for the East Asian monsoon and concluded that such abnormal climate conditions could possibly result from frequent and severe El Niño activities. Using paired oxygen isotope records from North America, Liu et al. (2014b) indicated that there was a transition from a negative Pacific North American (PNA)-like pattern during the mid-Holocene to a positive PNA-like pattern during the late Holocene, which led to drier conditions in northwestern North America. A similar conclusion was reached by Finkenbinder et al. (2016) based on lake sediment records from Newfoundland. They argued that this transition took place around 4.3 ka, leading to wetter conditions across the Newfoundland region. In contrast, Bond et al. (2001) argued that North Atlantic SST anoma-lies around 4.2 ka were related to a negative North Atlantic Oscillation (NAO) pattern, linked to solar forcing. Deininger et al. (2017) also found that changes in the atmospheric circulation associated with northward and southward propagating westerlies (similar to the NAO but on a millennial instead of a decadal scale) could be a possible driver of coherency and cyclicity during the last 4.5 kyr, as seen in multiple speleothem δ 18 O records that span most of the European continent. Thus, although there have been many suggested mechanisms, the ultimate drivers for climatic anomalies at 4.2 ka remain unclear. Wang (2009a) reviewed studies of Holocene cold events, and concluded that the most severe Holocene cold event, at ∼ 8.2 ka, was brought about by an outburst flood from proglacial Lake Agassiz. This large volume of freshwater drained into the North Atlantic extremely rapidly, leading to a brief reorganization of the Atlantic Meridional Overturning Circulation (AMOC) and a southward displacement of the ITCZ, resulting in dry conditions over many regions (Barber et al., 1999;Bianchi and McCave, 1999;Risebrobakken et al., 2003;McManus et al., 2004;Clarke et al., 2004). Potential external forcing factors for the 4.2 ka BP Event include non-linear responses to Milankovitch forcing, solar irradiance variations, and explosive volcanic eruptions, all of which may have brought about variations in the oceanatmosphere system (Booth et al., 2005). Wang (2009a) concluded that solar irradiance minima were the main cause of cold events in the middle to late Holocene (including the 4.2 ka BP Event) and that internal oscillations within the climate system could possibly have intensified these cold events under certain circumstances (Wang, 2009b).
In summary, the 8.2 ka event and corresponding southward shift in the ITCZ were caused by glacial flooding of the North Atlantic and this can be reasonably simulated by coupled general circulation models (GCMs) with different boundary conditions and freshwater forcing (Alley and Agustsdottir, 2005;LeGrande et al., 2006). By contrast, the forcing mechanisms that brought about the 4.2 ka BP Event are currently uncertain. At 4.2 ka, the major global monsoon and oceanatmosphere circulation systems may have been deflected or weakened synchronously, causing major century-scale precipitation disruptions, with severe megadroughts over many different regions (Weiss, 2017). As GCM simulations of the 4.2 ka BP Event have not received much attention, in this study, the spatial patterns and corresponding mechanisms relevant to the 4.2 ka BP Event are examined and compared to those associated with the 8.2 ka event.

Data and methodology
Simulations of the last 21 kyr (TRACE-21) were used in this study (He, 2011;He et al., 2013;Wen et al., 2016). These transient simulations have been completed using version 3 of the Community Climate System Model (CCSM3), which is a coupled ocean-atmosphere general circulation model developed by the National Center for Atmospheric Research (NCAR). The atmosphere model in the CCSM3 is the Community Atmospheric Model 3 (CAM3) with a horizontal resolution of ∼ 3.75 • (T31), and the ocean model is the Parallel Ocean Program (POP) with a longitudinal resolution of 3.6 • and variable latitudinal resolution.
The "full-forcing" TRACE-21 simulation includes changes in orbital parameters, greenhouse gases, ice extent (based on the ICE 5G-VM2 configurations), and meltwater fluxes from the Northern Hemisphere and Antarctic ice sheets. The orbital forcing is based on transient variations of orbital configuration (Berger, 1978). The concentrations of greenhouse gases were adopted from Joos and Spahni (2008). The ice sheet data were modified from the reconstruction of Peltier (2004) and the meltwater scheme was adopted from Liu et al. (2009).
Simulations in which only one of these factors was included have also been carried out and are available in the TRACE-21 archive Wen et al., 2016). These simulations can reproduce the timing and magnitude of many aspects of climate evolution during the last 21 ka, such as changes in SST (He et al., 2013). However, there are significant differences between the rate of temperature change in the model during the early Holocene and many paleoclimatic records (Liu et al., 2014a;Marcott et al., 2013;Marsicek et al., 2018). In this study, we do not address this enigma but use the transient model data to compare intervals within the Holocene when abrupt changes in climate are known to have occurred in some regions (∼ 8.2 and ∼ 4.2 ka). These times were recently adopted by the International Commission on Stratigraphy as the chronological boundaries of the early, middle, and late Holocene (Walker et al., 2012(Walker et al., , 2018. We examine mean annual surface temperature, annual precipitation, and SSTs from the full-forcing experiment, and also AMOC strength, defined as the maximum Atlantic stream function between 20 and 50 • N between 500 and 5000 m depth (Ottera et al., 2010) from the full-forcing and orbital-forcing experiments.

Results
First, we assess Holocene climate variability as simulated in the full-forcing experiment. Figure 1 shows the time series of surface temperature and precipitation over the last 13 kyr. It shows cooling associated with the Younger Dryas, followed by Holocene warming, but also a brief cooling episode from ∼ 8500 to 8000 BP. Thereafter, the record exhibits strong multi-century-scale variability. Temperature and precipitation are positively correlated at this global scale. It is tempting to associate the colder episodes with those identified by Wanner et al. (2011) or Bond et al. (2001) but only a few of these are coincident in time. The period 4.5-4.0 ka was chosen for analysis, by subtracting the mean annual 2 m air temperatures, SSTs, and precipitation of the period 4500-4000 BP from the preceding period (4800-4500 BP). The spatial distribution of air temperature (Fig. 2a) shows that temperatures were significantly colder over most of the extratropical Northern Hemisphere but generally warmer in the tropics and in the Southern Hemisphere. The main exceptions are northern South America, which was cooler, and northern India and Pakistan, which were significantly warmer. Precipitation decreased over almost all of the Northern Hemisphere, particularly in the tropics where the ITCZ shifted southward, mainly over South America and adjacent ocean regions, resulting in higher rainfall in the 0-20 • S zonal band from 4500 to 4000 BP (Fig. 2b). There was less precipitation over the northern part of China but more precipitation over southern China, consistent with paleoclimate reconstructions that indicate a weaker East Asian monsoon Tan et al., 2018a). This pattern is also similar to the situation during the Little Ice Age (LIA) in China and some of the  megadroughts that have happened in recent centuries (Cook et al., 2010;Tan et al., 2018b). Over other Asian monsoon regions, such as India, there were also significant precipitation reductions during the second half of the 5th millennium BP, consistent with speleothem records that show a decline in Indian summer monsoon rainfall over this period (Kathayat et al., 2017). Over Central America and the northern edge of South America, conditions were also drier in the later period, but over the rest of South America, and adjacent ocean regions, precipitation was higher, due to a southward displacement of the ITCZ; this pattern is supported by speleothem records of rainfall in Mexico and Brazil (Lachniet et al., 2013;Bernal et al., 2016). The SST pattern shows significantly cooler temperatures in the period 4500-4000 BP over the North Atlantic. This cooling is centered around 50 • N (south of Iceland) and extends into the subtropics on the eastern side of the subtropical gyre. Slightly cooler temperatures are also found over the North Pacific (Fig. 2c). By contrast, for most of the Southern Hemisphere, there was a positive change in temperature. Rotated empirical orthogonal function (EOF) analysis on the global SST field shows the primary feature (in EOFs 1 and 2) to be the cooler SSTs over the North Atlantic, with a shift around 4.5 ka from a predominantly positive to a generally negative pattern (Fig. 3). This is similar to an AMOC-like pattern over the northern Atlantic that has been identified in both instrumental and paleoclimatic records (Delworth and Mann, 2000;Knudsen et al., 2011).
The same evaluation of changes in the 9th millennium BP was made by subtracting the mean annual 2 m air temperatures, SSTs, and precipitation from 8800 to 8000 BP from the preceding period (9200-8800 BP), since an abrupt change in temperature in the model occurred around 8.8 ka (Fig. 1a). Air temperatures were significantly lower in the second period over most of the Northern Hemisphere; only a zone from northern South America across to sub-Saharan Africa and In-   dia was warmer in the second period (Fig. 4a). Almost the entire Southern Hemisphere was warmer. Precipitation was lower in the second period across all of the Northern Hemisphere, especially along the ITCZ, which was displaced to the south. This resulted in increased rainfall in a belt south of the Equator, across almost all of the tropics (Fig. 4b). The rest of the Southern Hemisphere was also slightly wetter. SSTs show a strong pattern of cooling over the North Pacific, and the eastern North Atlantic, south of Iceland, extending around the Atlantic subtropical gyre into the tropical Atlantic and Caribbean (Fig. 4c). Rotated EOFs show that the anomalies in the North Atlantic and North Pacific dominate the first three EOFs (Fig. 5).
The spatial patterns of temperature changes, precipitation changes, and SST changes were remarkably similar in the late 9th millennium BP and in the period leading up to the late 5th millennium BP (Fig. 6). The major difference (Fig. 6a) is that SST changes over the subtropical Atlantic were greater, and the related changes across the Northern Hemisphere in the 9th millennium BP were larger than in the late 5th millennium BP. Similarly, the major changes in precipitation patterns were comparable but less pronounced from 4500-4000 BP. These similarities are somewhat puzzling as the meltwater forcing sensitivity experiment clearly shows that the 8.2 ka event was induced by a massive freshwater flux into the Atlantic, whereas (as far as we know) no comparable meltwater event occurred in the late Holocene, so it seems unlikely that such forcing was a factor driving the changes seen in the model output for 4500-4000 BP.

Discussion and conclusions
Paleoclimate records have shown that unusually dry conditions persisted for several centuries around 4.2 ka over many regions, and in some areas these had devastating societal impacts. In this study, the spatial patterns of temperature, precipitation, and corresponding circulation anomalies during the latter part of the 9th and 5th millennia BP (4800-4500 versus 4500-4000 BP and 9200-8800 versus 8800-8000 BP) were compared based on model simulations. The changes  in climate during both periods were similar and characterized by significant temperature and precipitation decreases over most of the Northern Hemisphere, whereas the Southern Hemisphere was slightly warmer and wetter. In particular, the ITCZ was displaced to the south across much of the globe, and monsoon regions of the Northern Hemisphere were generally drier. On a regional scale, there was less precipitation over the northern part of China but more precipitation over southern China, indicating a reduced eastern Asian summer monsoon.
It is clear that the earlier period was strongly influenced by freshwater forcing in the North Atlantic, which drastically reduced the AMOC. The similarity in anomaly patterns between the 8.2 ka event and the late 5th millennium BP suggests that there was also disruption to the AMOC in the later period. However, as there was no comparable freshwater forcing in the 5th millennium BP, we must therefore consider what other factors might have played a role in reducing AMOC strength. There were no major solar irradiance changes at that time, so we can rule that out as a forcing factor. However, there was a major eruption of the Icelandic vol- cano Hekla at ∼ 4200 BP, and it is possible that such an event could have brought about regional cooling, leading to more extensive, thick sea ice and attendant freshwater effects on the AMOC (see Moreno-Chamarro et al., 2017). This mechanism deserves further scrutiny.
In the all-forcing TRACE-21 simulation, AMOC strength declined slightly during the late Holocene and underwent multi-century fluctuations (Fig. 7a), which were strongly correlated with SSTs in the region of the North Atlantic where cooling was so prominent from 4.5 to 4.0 ka (Fig. 8). Mean SSTs in this region over the last 4500 years of the model simulation stayed below the 4.8-4.5 ka average for ∼ 69 % of the time (Fig. 8), and AMOC strength was similarly below the 4.8-4.5 ka mean for 63 % of the time (Fig. 7a). One of these fluctuations was associated with an AMOC minima around 4.2 ka. In the TRACE-21 model simulation with only orbital forcing, AMOC strength reached its Holocene maximum around 4.8 ka, then slightly weakened (by ∼ 10 %) over the late Holocene, staying below the 4.8-4.5 ka mean for 87 % of the time, with minor multi-century variations superimposed on the long-term downward trend (Fig. 7b). This suggests that a combination of factors, led by long-term changes in insolation, drove a steady decline in SSTs across the North Atlantic and a reduction in the AMOC, with associated teleconnections across the globe (including drought in some regions). Minor fluctuations around this declining trend were the dominant pattern for most of the last 4500 years. This interpretation helps explain widespread paleoclimatic evidence for the onset of neoglaciation around 5000-4500 BP, followed by a series of neoglacial advances and retreats during recent millennia (Porter, 2000;Barclay et al., 2009;Solom-ina et al., 2015;Bradley and Bakke, 2019). Since the onset of neoglaciation early in the 5th millennium BP, mountain glaciers fluctuated in extent but did not entirely disappear, indicating that a distinctly different climate state prevailed compared to the period prior to ∼ 5 ka, when many mountain regions were ice-free.
We therefore conclude from the model simulations that the 4.2 ka BP Event was one of several late Holocene multicentury fluctuations that were embedded in a long-term, lowfrequency change in climate that occurred after ∼ 4.8 ka. Worldwide climatic anomalies during these fluctuations were driven by changes in the strength of the AMOC and related teleconnections. Whether such multi-century fluctuations were a response to internal centennial-scale oceanatmosphere variability (see Yan et al., 2018), external forcing (such as explosive volcanic eruptions and associated feedbacks), or a combination of such conditions is not known. Further studies of the role of both external forcing and internal variability are needed to provide a better understanding of such mechanisms (see Ottera et al., 2010;Moreno-Chamarro et al., 2017;Gupta and Marshall, 2018).
Data availability. The TraCE-21ka data used in this study can be accessed through the website of Earth System Grid (https://www. earthsystemgrid.org/project/trace.html, last access: 1 June 2018).
Author contributions. LN, RSB, and MY designed the experiments and carried them out. LN and MY performed the formal analyses. LN, RSB, and MY performed the investigation. LN and JL contributed to funding acquisition. JL and RSB provided supervision. LN and RSB prepared the paper with contributions from all co-authors.
Competing interests. The authors declare that they have no conflict of interest.

Special issue statement.
This article is part of the special issue "The 4.2 ka BP climatic event". It is a result of "The 4.2 ka BP Event: An International Workshop", Pisa, Italy, 10-12 January 2018.