The Pliocene (5.3–2.6 Ma) warm period, a possible analogue for future climate warming1,2, is characterized by a global cooling trend culminating in the extensive Northern Hemisphere glaciations of the Quaternary (2.6–0 Ma)3,4,5,6. Recent work suggests that monsoon circulation intensifies over the same period7,8,9. However, this contradicts evidence that enhanced monsoons increase heat transport to high latitudes10,11 as well as model predictions and geological records that show summer monsoons were generally weaker during colder periods10,12,13. Geochemical7,8,14, magnetic7,15 and biological16,17,18,19 proxies have been widely used to infer summer monsoon history. A major problem inherent in these proxies is that they are potentially affected by both precipitation and temperature. During the Quaternary period when high (low) temperature was generally coupled with high (low) summer monsoon intensity7,10 these proxies may be accurate in reflecting summer monsoon intensity. However, they cannot be assumed to show straightforward responses to monsoon intensity when temperature and precipitation potentially have different trends7. Therefore, significant debate7,14,16,17,20 exists about Pliocene monsoon history. Understanding Pliocene monsoon history and its relationship with the onset of intensive Northern Hemisphere glaciations requires separation of precipitation and temperature signals.

Rock magnetic parameters are widely used in reconstructing paleoclimate history from aeolian sediments7,15,21,22. Magnetic susceptibility measured at 470 Hz (χLF) of surface soils across the Chinese Loess Plateau is strongly correlated with annual mean precipitation (AMP)21,23. Ultrafine magnetic grains with very high χLF but no stable magnetic remanence are produced via intermittent wetting and drying cycles associated with monsoon climate during soil formation on the Chinese Loess Plateau and are responsible for high χLF in paleosol layers21,23,24. Fine stable single domain (SSD) and small pseudo- single domain (PSD) magnetic grains that have high anhysteretic remanent magnetization susceptibility (χARM) but low χLF are also formed during soil formation15. Inter-parameter proxies for magnetic grain-size variations, such as χARMLF and χARM/SIRM, where SIRM represents saturated isothermal remanent magnetization and measures the magnetic grain concentration with remanence, are also sensitive paleoclimate indicators15,24. However, no study explores the relationship between magnetic parameter ratios and specific, separate climatic parameters for surface soils on the Chinese Loess Plateau.

Here we demonstrate that soil magnetic parameters have different sensitivities to AMP and annual mean temperature (AMT), based on a new surface soil calibration across the Chinese Loess Plateau (Fig. 1). Then, we use magnetic parameter records from wind-blown Pliocene red clay (finer and more weathered than Quaternary loess) to discuss the evolution of paleotemperature and paleoprecipitation on the Chinese Loess Plateau. Finally, we tentatively present the first separate calibrated, quantitative reconstructions of Pliocene air temperature and East Asian summer monsoon precipitation history on the Chinese Loess Plateau through joint analysis of red clay magnetic parameters. Although these records have large uncertainties, this is the first quantified separation of these key aspects of Pliocene climate.

Figure 1
figure 1

Schematic map of the Chinese Loess Plateau and the location of the Chaona section (black circle), the Lingtai section (black circle), surface soil samples (solid dots) and meteorological stations (red circle) used in this paper.

The inset illustrates the location of the Chinese Loess Plateau relative to the Tibetan Plateau and modern Asian atmospheric circulation. Magnetic parameter values were averaged for dots within the gray rectangle before comparing with the meteorological data closest to them. Revised from Song et al. (2007)54.


Surface soil calibration

Fig. 2 shows the correlation patterns between Chinese Loess Plateau surface soil χARMLF (Figs. 2a and 2b), χARM/SIRM (Figs. 2c and 2d), χARM (Figs. 2e and 2f) and AMP and AMT from meteorological stations closest to these surface soils (China Meteorological Data Sharing Service System:, in Chinese). We find that χARM/SIRM is exponentially correlated with AMP and linearly correlated with AMT, as is the case for χARM. The correlations of these two magnetic parameters with AMP are stronger than with AMT (Fig. 2 and Table 1). The correlation between χARM and AMT is even weaker than the correlation between AMT and AMP, suggesting that χARM is primarily controlled by AMP variations. By contrast, χARMLF has similar R2 values to AMP and AMT (Fig. 2a, 2b and Table 1). The correlation between χARMLF and AMP is linear, whereas the correlation between χARMLF and AMT is more complex with a concave down regression fit (Fig. 2a,2b and Table 1).

Table 1 Correlations between magnetic and climatic parameters
Figure 2
figure 2

Correlation patterns between Chinese Loess Plateau surface soil magnetic and climatic parameters.

(a), χARMLF and AMP; (b), χARMLF and AMT; (c), χARM/SIRM and AMP; (d), χARM/SIRM and AMT; (e), χARM and AMP; (f), χARM and AMT; (g), AMP and AMT of meteorological stations closest to these surface soils; (h), χARMLF and AMP × AMT. Meteorological data are from the China Meteorological Data Sharing Service System (, in Chinese).

Paleoclimate reconstruction

The χARM, χARMLF and χARM/SIRM records from Chaona9 (35° 06′ N, 107° 12′ E) and Lingtai25 (35° 04′ N, 107° 39′E), central Chinese Loess Plateau, covary with each other (Fig. 3). The three magnetic records co-vary from 6 to ~4.8 Ma. However, from ~4.3 to 2.7 Ma, the χARM and χARM/SIRM records show increasing trends while the χARMLF record shows a decreasing trend.

Figure 3
figure 3

Paleoclimatic and paleoceanographic data for the time interval 6–2.6 Ma.

(a), (b), (c), the Chaona (black) and Lingtai (blue) χARM, χARM/SIRM and χARMLF records, respectively. (d), (e), East Equatorial Pacific Ocean Drilling Project (ODP) site 846 benthic oxygen isotope26 and sea surface temperature6 records respectively. (f), (g), reconstructed Chaona (black) and Lingtai (blue) annual mean temperature (AMT) and annual mean precipitation (AMP) records based on the correlation regression fits in Figs. 2f and 2h respectively. (h), Chaona (black) and Lingtai (blue) sedimentation rate records. For the interval 3.6–2.6 Ma, the Chaona and Lingtai age models were tuned to the monsoon stack55; for the interval 6–3.6 Ma, the age models of the two sites are based on paleomagnetic dating9,25. EASM: East Asian summer monsoon.

Based on regression fits in Fig. 2f and 2h we tentatively reconstruct the first quantified temperature and precipitation history between 6 and 2.6 Ma on the Chinese Loess Plateau (Fig. 3f and 3g). χARM is not sensitive to AMT variations and can therefore be used to estimate AMP history (Fig. 2f). Based on the correlation fit between χARMLF and AMT × AMP (Fig. 2h) and the reconstructed AMP record (Fig. 2f) we then tentatively also reconstruct AMT history on the Chinese Loess Plateau for the interval 6–2.6 Ma. We also use χARM/SIRM to estimate AMP (Fig. 2d) and then solve for AMT based on the correlation relationship between χARMLF and AMT × AMP (Fig. 2h) and the reconstructed AMP record (Fig. 2d). The resultant AMP and AMT values between these two independent methods are closely aligned (Fig. 4), supporting the reconstructions.

Figure 4
figure 4

A comparison of reconstructed annual mean precipitation (AMP) and annual mean temperature (AMT).

(a), AMP based on the empirical relationship in Fig. 2d; (b), AMT based on the empirical relationships in Figs. 2d and 2h; (c), AMP based on the empirical relationship in Fig. 2f; (d), AMT based on the empirical relationships in Figs. 2f and 2h. We note that the AMT and AMP records based on two methods have similar trends. The dashed line represents the 2 σ uncertainty (standard error) of each record. The uncertainties in the paleo data were propagated from the uncertainties of surface soils in Fig. 2.


This concave down shaped regression fit between χARMLF and AMT indicates that χARMLF is more sensitive to temperature variations when temperature is lower. By contrast, the concave up shaped regression fits between χARM/SIRM or χARM and AMP indicates that χARMLF and χARM are more sensitive to precipitation variations when precipitation is higher (Table 1 and supplementary materials).

These relationships suggest that in the scenario of high (low) precipitation coupled with high (low) temperature, χARM, χARMLF and χARM/SIRM should all have similar trends, with higher (lower) magnetic parameter values corresponding to higher (lower) precipitation and temperature. This relationship has been shown to be the case in the Quaternary when precipitation and temperature are coupled15. However, χARMLF and χARM/SIRM should show different trends in the scenario of increasing precipitation and decreasing temperature trends, as has been speculated for the Pliocene7. In this case, χARM/SIRM and χARM will have an increasing trend, controlled dominantly by the effects of increasing precipitation (Fig. 2d, 2f and Table 1), while χARMLF will have a decreasing trend, attributed to the dominant effects of decreasing temperature (Fig. 2a and Table 1). However, any decreasing trend of χARMLF should be modest because of the effects of increasing precipitation, which will tend to pull the decreasing χARMLF trend in the opposite direction. For similar reasons, an increasing trend of χARM/SIRM should also be modest because of the effects of decreasing temperature, which will tend to pull the increasing χARM/SIRM trend in the opposite direction. In the scenario of increasing temperature and decreasing precipitation trends, χARM/SIRM will have an increasing trend or stay at a constant low value, attributed to effects of decreasing precipitation; and χARMLF will have a decreasing trend, attributed to effects of decreasing precipitation. However, the trends will again be modest because of the opposite effects of temperature and precipitation.

During ~4.8–6 Ma, the χARM, χARMLF and χARM/SIRM records from Chaona and Lingtai co-vary with each other (Fig. 3a, b and c) and with ice volume26. This relationship is consistent with the concept that monsoon climate dominated the Chinese Loess Plateau and high (low) temperature was coupled with high (low) precipitation7. However, during ~4.8–2.7 Ma, χARMLF decreases but both the χARM and χARM/SIRM records increase (Fig. 3a, b and c). Based on the above surface calibration results (Fig. 2 and Table 1), we conclude that the Chinese Loess Plateau experienced a cooling trend coeval with increasing precipitation from ~4.8 to 2.7 Ma. The quantitative reconstructions support the conclusion that the Chinese Loess Plateau experienced a cooling trend concomitant with increasing precipitation from ~4.8 to 2.7 Ma (Figs. 3f and g).

In contrast to the above, some biological and geochemical records14,27 on the Chinese Loess Plateau have been used to suggest drying climate from ~4.5 to 2.7 Ma on the Chinese Loess Plateau. However, we argue that biological and geochemical proxies are influenced by both temperature and precipitation. Furthermore, pollen records from the Chaona section19,28 are consistent with our inferences based on magnetic parameters. According to these studies, during the late Pliocene, Cupressaceae and Juniperus along with Ulmus dominated the Loess Plateau, revealing that the environment showed a marked change characterized by hot and rainy summers and cold and dry winters19,28. This inconsistency between biological records on the Loess Plateau19,27 underscores the complexity of using biological evidence to infer monsoon intensity. Geochemical monsoon records, such as the widely used Chemical Index of Alternation, suggest that chemical weathering was weaker on the Loess Plateau during the late Pliocene in comparison with the early Pliocene. However, geochemical parameters are affected not only by precipitation, but also by temperature (controlling reaction rates) and materials available to weather (determined by sediment accumulation rate)29. It is widely reported that sediment accumulation rate is higher for the late Pliocene on the Loess Plateau14,30 (Fig. 3h), leading to more materials available for weathering. Thus, weaker chemical weathering and alteration is not necessarily linked with weaker monsoon precipitation, but can be attributed to cooler climate and the availability of sediment for weathering14 (Fig. 3h). Our magnetic evidence does not suffer from these ambiguities and demonstrates increased precipitation over this period.

While the general qualitative trends in our data are clear and similar trends are seen in the quantitative data, our quantitative reconstructions potentially have large uncertainties. First, although magnetic parameters have different sensitivities to AMP and AMT, each of them is potentially affected in part by both AMP and AMT. Thus, the influence of another component cannot be entirely removed, even if the majority of variation can be explained by one climate parameter alone. Second, the reconstructed Pliocene paleotemperature exceeds modern temperatures on the Chinese Loess Plateau. Although it is well known that the Pliocene was warmer than the Quaternary31, these higher temperatures increase the uncertainties associated with the quantitative paleotemperature reconstructions based on our modern climofunction during the Pliocene. Third, the decoupled temperature and precipitation trends on the Chinese Loess Plateau seem to contrast with proposed monsoonal climate in the region, potentially meaning that modern monsoonal analogues cannot be directly applied to Pliocene red clay. However, although our records show that Pliocene temperature and precipitation on the Loess Plateau have opposite trends over tectonic timescales, it is clear from our records (Fig. 3) that at orbital timescales, temperature and precipitation are in phase, consistent with features of a monsoonal climate32. Thus, we maintain that modern calibrations can still be used to understand Pliocene magnetic paleoclimatic records.

This method used here is readily applicable to loess or red clay where magnetic enhancement occurred during interglacial periods associated with increases in abundance of ultrafine magnetic grains produced via soil-formation processes33,34. However, caution needs to be exercised in situations where magnetic minerals tend to be destroyed during soil-formation processes, such as in Siberian or Argentine loess35,36,37.

The interval during ~4.8–4.3 Ma has been argued to be anomalous on the Chinese Loess Plateau20. Abundant clay coatings and a high free iron/total iron ratio in red clay sediments suggest that this interval experienced high monsoon precipitation20. However, by contrast this interval shows low χLF values7,20. Thus, it has been argued that χLF is not able to indicate East Asian Summer Monsoon intensity for this time interval20. However, we note that, in contrast to any other time interval during 6–2.6 Ma, ~4.8–4.3 Ma is dominated by low sedimentation rate (Fig. 3h) giving a significantly longer time for soils to develop clay coatings and experience chemical weathering. In contrast, magnetic enhancement of Chinese loess is not a function of pedogenic duration21 and magnetic parameters will thus better reflect climate conditions during this time period. Lower dust sedimentation rates can therefore explain the inconsistent monsoon proxies during ~4.8–4.3 Ma on the Chinese Loess Plateau. This time interval is synchronous with the point when the gradual closure of the Panama Seaway started to significantly affect surface seawater exchange between the Equatorial Pacific Ocean and the Caribbean Sea38 and signals initiation of Northern Hemisphere climate reorganization (Fig. 3).

There are two potential ways to explain the Pliocene wetting trend on the Chinese Loess Plateau: intensified East Asian summer monsoon precipitation or intensified westerly precipitation. However, oceanic moisture sources for westerly flow are far away from the Chinese Loess Plateau, limiting the contribution from westerly precipitation. We have previously attributed an apparently intensified East Asian Summer Monsoon simultaneous with increasing ice volume from 4.7 to 2.6 Ma to the combined effects of the closure of the Panama Seaway and Tibetan plateau uplift9. However, evidence supporting late Pliocene uplift of the Tibetan plateau is controversial39 and attempts to separate temperature and precipitation trends have not been performed. Here we propose an alternative mechanism to explain concurrent Pliocene climate cooling and monsoon intensification, demonstrated from the magnetic parameter records having different sensitivities to AMT and AMP. Paleoceanographic data38 show that as a result of gradual closure of the Panama Seaway, East Equatorial Pacific surface water freshens from ~4.8 Ma due to easterly trade wind transportation of moisture from the central Atlantic Ocean and the Caribbean Sea to the tropical Pacific Ocean. Modern day salinity differences between the East Equatorial Pacific Ocean and the Caribbean Sea38 were established by ~4.2 Ma. Fresher seawater will then have been transported to the North Pacific Ocean via ocean currents surrounding the North Pacific gyre, causing freshening of the North Pacific surface water. This freshening drove enhanced sea ice formation on the surface North Pacific40. Model simulation41 demonstrates that sea ice formation would strengthen the high pressure center over the North Pacific, enhancing Southerly and Southeasterly winds, which in turn would intensify East Asian Summer Monsoon precipitation and meridional moisture transport41 (Supplementary Fig. 1). Critically, we propose that intensified meridional moisture transport would cause further middle and high latitude precipitation (Fig. 3) and consequent further freshening in the Pacific and the Arctic Oceans. This change would drive additional intensification of the North Pacific high pressure center and East Asian Summer Monsoon precipitation41. We note that this meridionally-transported moisture would likely also have been transported to the North American continent2, providing a moisture source for North American ice sheets to grow. This previously unknown positive feedback explains for the first time the paradox of concurrent Pliocene cooling and East Asian Summer Monsoon precipitation intensification. Our hypothesis also explains previous paleoceanographic data6 that reveal that East Equatorial Pacific surface water experienced a cooling trend from ~4.3 to 2.6 Ma (Fig. 4e). No corresponding forcing was identified but under our proposed feedback an intensified North Pacific high pressure system at the same time would drive Equatorial Pacific water flow from east to west, intensifying upwelling and providing a reasonable explanation for the observed East Equatorial cooling trend. This previously overlooked positive feedback was initiated in the run up to the onset of extensive Northern Hemisphere glaciation. As it forces gradual cooling of Pacific surface water while at the same time providing both a moisture source for ice sheets and a freshwater driver for sea ice growth, we argue that this Pacific atmosphere-ocean feedback is a critical but hitherto unrecognized factor in the initiation of the Northern Hemisphere glaciation of the Quaternary Ice Age.

The model simulation of reference 41 is an important basis for our hypothesis. However, one important difference exists between the two studies. Reference 41 presents climate during boundary condition snapshots: completely closed and open (with sill depth of 2559 m) Panama Seaway. By contrast, our hypothesis treats the seaway closure as a continuous process, decreasing sill depth42 from probably less than 200 to 0 m. Using a sill depth close to that of the Pliocene (370 m), a study43 found that closure of the Panama Seaway would intensify precipitation in Northern Hemisphere high latitudes but that this closure plays a limited role in initiation of the Northern Hemisphere glaciations. However, reference 43 did not include the new positive feedback involving monsoon intensification proposed in this paper.

Reference 41 produces a weakened East Asian winter monsoon associated with closure of the Panama Seaway and development of a high pressure system over the North Pacific Ocean. This contrasts with evidence demonstrating that the East Asian winter monsoon became stronger during the late Pliocene7. We note that although reference 41 showed geographical distribution of surface air pressure and wind during February which includes the information of Siberian-Mongolian high pressure system, it did not consider the likely intensification of the Siberian-Mongolian high pressure system from the early to the late Pliocene under global cooling44,45. As this system has more direct control over the winter climate on the Chinese Loess Plateau, due to its proximity, this explains the inconsistency between the model simulation results of reference 41 and the geological observations7.

A recent study46 also reported that different models produced variable sea surface salinity changes associated with closure of the Panama Seaway. However, only three of the 12 models (EC(415 m); CCMS3(1475 M); UVIC 6sh(130 m)) show dominantly increasing sea surface salinity associated with closure of the Panama Seaway. More importantly, of the 12 models examined, only the HacCM3 model by reference 43 used boundary conditions representative of part of the Pliocene. Thus, the results of reference 43, which demonstrated that the North Pacific sea surface salinity decreased significantly associated with closure of the Panama Seaway, are more convincing as accurate representations of Pliocene salinity.

While the closure of the Panama seaway is well constrained in the Pliocene, whether the Tibetan Plateau experienced a phase of intense uplift during the Pliocene is a topic of long-standing debate44,47,48,49,50. No clear phase of intensive uplift is known for 4.6 Ma and although some model simulations demonstrate the potential climatic influence of the plateau7,51, our reconstructions demonstrate the close association of Pliocene climate change with closure of the Panama seaway. A permanent Pliocene El Niño condition has been hypothesized to exist in the Pacific Ocean, possibly associated with closure of the Indonesian Seaway, while the end of this condition around 2.7 Ma may have contributed to initiation of the intensive Northern Hemisphere glaciations52. However, a recent study based on a coupled atmosphere-ocean general circulation model demonstrates that were the Pliocene dominated by a permanent El Niño-like condition it is unlikely that this would provide a major contribution to global warmth and its termination cannot contribute significantly to the onset of the intensive Northern Hemisphere glaciations53. Indeed, this same simulation found that there was no permanent El Niño-like condition during the Pliocene53.

As such, the Panama seaway closure initiated positive feedback involving monsoon circulation provides the best explanation of the data here. This paper demonstrates the fundamental importance of a hitherto unknown oceanographic-atmospheric feedback in driving late Cenozoic cooling, monsoon intensification and the onset of Northern Hemisphere glaciation, initiated by tectonic forcing. Our explanation of Pliocene climate trends emphasizes the Pacific as central to global climate. Model simulations1,2 of Pliocene and future climate need to take account of this hitherto unknown mechanism.


Surface soil samples were taken ~1 cm below the surface from A-horizons at locations across the Chinese Loess Plateau (Fig. 1). Removing the top 1 cm removes the potential effects of pollution to these soils. In selecting sites, we sought level, stable land surfaces that showed no obvious evidence of surface erosion or extensive human disturbance.

For all samples, magnetic susceptibility was measured using a Bartington MS2 susceptometer at frequencies of 470 Hz (i.e., χLF) and 4700 Hz (i.e., χHF). For surface soils and Chaona samples ARM was imparted using a 100 mT peak AF and a 0.05 mT constant biasing field, while for Lingtai samples ARM was imparted using a 100 mT peak AF and a 0.1 mT constant biasing field. This parameter is also expressed as χARM after normalization by the 0.05/0.1 mT direct bias field. SIRM was imparted in all samples at 1 T using a pulse magnetizer and measured using a 2 G cryogenic magnetometer on Chaona and surface samples and using a spinner magnetometer for Lingtai samples. Following these measurements, χARMLF and χARM/SIRM were calculated.

Because of the different bias field used for the Chaona and Lingtai samples when acquiring ARM, it is not feasible to compare the absolute magnetic parameter values directly. The highest χARM values at Lingtai and Chaona between 6 and 2.6 Myr are 16.753 and 13.164 (×10−6 m3 kg−1) respectively, both occurring at ~2.76 Myr. As such, we divide the Lingtai χARM, χARMLF and χARM/SIRM records by 16.753/13.164 ( = 1.273) to correct the bias field effects and to allow direct comparison of magnetic records from both sections. The corrected Lingtai magnetic records correlate well with the records from Chaona (Fig. 3a, b and c), demonstrating that the above correction is valid.