Early Pleistocene climate in western arid central Asia inferred from loess-palaeosol sequences

Arid central Asia (ACA) is one of the most arid regions in the mid-latitudes and one of the main potential dust sources for the northern hemisphere. The lack of in situ early Pleistocene loess/dust records from ACA hinders our comprehensive understanding of the spatio-temporal record of aeolian loess accumulation and long term climatic changes in Asia as a whole. Here, we report the results of sedimentological, chronological and climatic studies of early Pleistocene loess-palaeosol sequences (LPS) from the northeastern Iranian Golestan Province (NIGP) in the western part of ACA. Our results reveal that: 1) Accumulation of loess on the NIGP commenced at ~2.4–1.8 Ma, making it the oldest loess known so far in western ACA; 2) the climate during the early Pleistocene in the NIGP was semi-arid, but wetter, warmer, and less windy than during the late Pleistocene and present interglacial; 3) orbital-scale palaeoclimatic changes in ACA during the early Pleistoceneare in-phase with those of monsoonal Asia, a relationship which was probably related to the growth and decay of northern hemisphere ice sheets.

The reported middle-upper Pleistocene loess-palaeosol profiles in northern Iran are distributed on the INGP (Agh Band section) and along the northern foothills of the Alborz mountains between the cities of Sari and Minodasht S9-S12 (Fig. 1b). These loess profiles consist of an alternation of homogenous dull, yellowish-coloured (2.5 Y 5/3) loess layers and well-developed brown (7.5 YR 4/3) palaeosol beds, deposited in glacial and interglacial periods, respectively S9, S11, S12 . Infrared optically stimulated luminescence (IRSL) dating suggested that the basal age of the loess deposits in the Agh Band section from the INGP was around 145±14 ka S12 , representing the oldest loess known so far in the INGP.
Red-coloured non-marine sediments are widely distributed in the INGP (Fig. 1b,   Fig. S1). They unconformably underlie the upper Pleistocene loess successions S12 and conformably overlie the late Cenozoic limestone sequences that include abundant mollusc shells (Fig. 2a). The sedimentology of three well-exposed sections (Fig. S1), namely AB1 (37°41'20" N, 55°9'30" E), AB2 (37°38'10" N, 55°12'53" E), and KB (37°36'1" N, 55°25'19" E), was investigated and the most complete AB1 section was selected for magnetostratigraphic study.  (Fig. S2). These species are indicative of shallow marine environments and confirm the Akchagylian age of the deposits S13-S14 . The basal age of the Akchagylian stage is 3.2 Ma based on bio-magnetostratigraphic dating with constraints provided by fission track dating of volcanic ashes. However, the age of the top is poorly constrained, with estimates ranging from ~2.4 to 1.8 Ma S15 . In any case, the limestone and the overlying red beds in the AB1 section must be younger than 3.2 Ma.

Magnetostratigrapy of the reddish loess-palaeosol sequence in the AB1 section
Magnetic hysteresis loops of representative samples from the AB1 section are narrow and are closed at around 400 mT (Fig. S3), suggesting that the magnetic carriers are mainly magnetically 'soft', with a low coercivity S16 . The thermomagnetic curves exhibit a major drop in magnetization near 580 °C, followed by a progressive decrease until 680 °C (Fig. S3), indicating the presence of both magnetite and hematite S16 . On orthogonal vector diagrams S17 most of the samples exhibit two magnetic components (Fig. S4). The low-stability component is generally unblocked at low temperatures ranging from room temperature to 200 °C and its direction is broadly consistent with the present day field; it is thus interpreted as a secondary viscous remanent magnetization (VRM). The second component has relatively stable directions and decays linearly towards the origin (Fig. S4). This high-stability component is interpreted as the characteristic remanent magnetization (ChRM), the mean direction of which was calculated using principal component analysis (PCA) of a least-squares fit S18 . Generally, the samples with maximum angular deviations (MAD) larger than 15° were considered to fail the significance test in directional analysis and were rejected for further analysis. Parallel samples were measured in these cases and they also did not pass the significance test. In total, 167 reliable ChRM directions were obtained from the 187 measured data (89.3 %). After tilt correction, the mean normal and reverse polarity directions are: D/I = 357.0°/50.5° (K=3.58, N=16), and D/I = 192.0°/−51.9° (K = 22.53, N = 151) respectively. The ChRM directions passed the reversal test for palaeomagnetic stability S19 , suggesting that the ChRMs obtained are the primary NRMs.
The palaeomagnetic polarity of the red-coloured strata in the AB1 section is defined by the virtual geomagnetic poles (VGP) calculated from the ChRM declination and inclination data (Fig. S5). The polarity record consists of two normal and two reversal intervals, designated N1, R1, N2, and R2, from the top to the bottom of the section. In addition, a stable and reliable normal point was obtained at 10.7 m, which may record a geomagnetic excursion (Fig. S5).
With the aforementioned biostratigraphic age constraints, the magnetostratigraphy for the red beds in the AB1 section is easily correlated to the Matuyama reversal polarity zone (C1r.1r to C2r.2r) of the geomagnetic polarity time scale S20 . There are two possibilities for additional correlations. The first possibility is that N1 and N2 correlate to the Olduvai event (C2n) and the Reunion event (C2n.1n), and that the normal event at 10.7 m correlates to the geomagnetic excursion occurring at 2.19 Ma (Fig. S5). Based on the resulting age model, the sediment redness record exhibits a similar pattern of variation to the redness record of the Lingtai section from the CLP S21 , and the LR04 stacked benthic δ 18 O record S22 (Fig. 4). An alternative option is that N1 and N2 correlate to the Jaramillo event (C1r.1n) and the Cobb Mountain event (C1r.2n), respectively (Fig. S5). However, this correlation is probably incorrect for the following reasons: 1) Using this age model most of the non-marine sediments were deposited during an interval equivalent to the middle Apsheron transgression of the Caspian Sea, and the estimated sea level was about 100 m higher than at present S23 .
Since the studied sections are located in a former marine environment (Fig. 1a), deposition of non-marine sediments at this time is in conflict with the regional geological history. 2) The normal event at 10.7 m cannot be correlated to any known geomagnetic excursions S20 . 3) The redness record exhibits no correlation with the deep-sea δ 18 O record S22 , conflicting with the middle Pleistocene loess records from mid-latitude Asia S24-S25 .
Based on the most likely magnetostratigraphic correlation, the sediment accumulation rates within the R1 and N2 intervals are ca. 3.03 cm/kyr and 2.29 cm/kyr, respectively. By using the sediment accumulation rate from the neighboring stratum, the basal and top ages of the red beds from the AB1 section are inferred to be ~2.383 and ~1.815 Ma, respectively.

Comparison of the redness (a*) record of the AB1 section and the stacked benthic δ 18 O (‰) record
We used the redness (a*) record as a high resolution proxy climate index for correlation to the LR04 stacked benthic δ 18 O (‰) record S22 for the following reasons: 1) The attitudes of the red-coloured strata in the red beds are the same and the alternations of palaeosol and loess-like horizons are highly comparable among the AB1, AB2 and KB sections (Fig. S1). This suggests that the red-coloured sediments were deposited continuously without significant hiatuses, at least on an orbital time scale. 2) Colour variations are the most readily visualized property of the red-coloured sediments in the studied sections (Fig. 2). Both strongly-developed and weakly-developed palaeosols have higher redness (a*) values than the loess-like sediments, suggesting that redness can be used as a high resolution index for stratigraphic correlations. 3) Numerous studies have demonstrated that variations in the redness of loess-palaeosol sequences are a reliable proxy for moisture and/or temperature history in mid-latitude of Asia S24-S27 . 4

) The reported middle-upper
Pleistocene loess records from northern Iran S11-S12 and southern Tajikistan S24 in ACA indicate that the dull yellowish-coloured loess layers (lower a* values) were deposited in glacial periods, whereas the brown palaeosol horizons (higher a* values) developed during interglacials, yielding a direct genetic linkage between Northern Hemisphere ice volume and sedimentary colours in mid-latitude Asia on orbital time-scales since the middle Pleistocene S24-S25, S28-29 .
A time series of redness (a*) of the AB1 section was established by linearly interpolating between the ages of geomagnetic polarity boundaries. Cross-spectral analysis demonstrates that the amplitudes of the peaks in spectral density of redness closely match those of the marine oxygen isotope record at the 41 ka periodicity (Fig.   S6). To facilitate correlation, the 41-kyr periodicity components of the redness record from the AB1 section, the redness record from the Lingtai Section S21 in monsoonal Asia, and the LR04 stacked benthic δ 18 O records S22  The well-constrained 'Reunion event' occurring at 2.128-2.148 Ma is used as a 'marker layer' (dark grey bar in Fig. 4) for stratigraphic correlation. It is apparent that a weakly-developed palaeosol (higher redness) formed during an interglacial period (lower δ 18 O), which is consistent with the loess records from the CLP S32 . The corresponding peaks in redness in both the INGP and CLP are slightly lagged relative to the peak in the δ 18 O record (Fig. 4). These lags may be attributed to different post-depositional remanent magnetization lock-in depths in loess and marine sediments and/or to bioturbation within the surface mixed layer S32 . Based on the primary palaeomagnetic time scale, variations of redness from the INGP and CLP yield similar patterns, and exhibit an in-phase relationship with global ice volume during the early Pleistocene. That is, higher redness (higher temperature and humidity) corresponds to lower δ 18 O (less ice and higher temperature), and vice versa (Fig. 4). It should be noted that the redness time scales for the AB1 and Lingtai sections S21 were obtained by linearly interpolating between the ages of geomagnetic polarity boundaries. This method leads to dating errors within polarities, which may result in the differences in redness between the two sections during certain intervals (e.g.

2.3-2.4 Ma).
According to the correlation of marker layers (Fig. 4, dark grey bar) and middle-late Pleistocene records from ACA and monsoonal Asia S24 , we believe these 'shifts' arise from uncertainties in palaeomagnetic dating rather than from an out-of phase relationship between ACA and monsoonal Asia.     The numbers in italics beyond the GPTS 2004 indicate the numerical age of geomagnetic excursions that occurred during the Matuyama reversed interval. Figure S6. Cross spectral analysis of the redness record from the AB1 section, using the primary palaeomagnetic time scale, and a stacked δ 18 O record S22 over the time interval 1.8-2.5 Ma. Spectral density is normalized and plotted on a log scale. The solid straight line denotes the 80% significance level of coherence. Shaded bars indicate the dominant spectral peaks for these records. Note the high coherence within the 41 kyr periodicity band.