A 18,000-year Record of Tropical Land Temperature, Convective Activity and 7 Rainfall Seasonality from The Maritime Continent

the onset of the Meghalayan period Abstract The maritime continent exports an enormous amount of heat and moisture to the rest of the globe 42 via deep atmospheric convection. How this export has changed through time during over de Late 43 Glacial period and through the Holocene is hardly known, yet critical for the understanding of 44 global climate dynamics. In this study, we present a very well dated, continuous paleoclimatic and 45 -environmental record from southern Thailand covering the last 18,000 years, including the first 46 land-based temperature reconstruction of the maritime continent. Confirming a recent climate 47 modelling study, we found evidence for a strongly seasonal climate for most of the late glacial 48 period, causing biomass burning and suppression of rainforest growth, despite rising CO 2 levels 49 and increasing mean humidity. Temperatures were ca. 5°C cooler than today during the last cold 50 stadial periods, and ca. 2°C warmer between 7000-2000 yr ago. We also found that tropical wet- 51 season insolation (WSI) is a primary driver of the strength of deep atmospheric convection, exerting 52 a strong influence on the both the Monsoon systems and the Walker circulation, and hence on global 53 climate dynamics. glacial cycles due to sea level change. Our results highlight the importance of the IPWP as the 'steam engine of the world' to global climate, and how it responds to orbital forcing and sea level change.

Two parallel sediment cores were retrieved in one-meter sections using a rod-operated Russian 119 corer from a small raft at the deepest part of the lake. After recovery, the sections were wrapped in 120 foil and secured and transported in PVC tubes to Stockholm University, where they were stored at 121 4°C until further description (Table S1) and analysis. Sub-samples were taken in contiguous 1-cm 122 increments and split to accommodate subsequent analyses. One half of the samples was utilized for 123 macrofossil and charcoal analysis and radiocarbon dating. The other half of the samples was freeze-124 dried and analysed for loss-on-ignition (LOI), bulk total organic carbon (TOC), nitrogen (TN) and 125 their bulk isotopes (Table S4), lipid biomarkers and compound-specific hydrogen and carbon 126 isotopes. For LOI, samples were dried overnight at 105°C, ground and then combusted at 550 °C 127 for 3h. LOI was calculated as a percentage of the dry sample weight to obtain an estimate of the 128 organic matter and carbonate content. In parallel, a sediment-water interface surface core covering 129 the last 150 years was retrieved and sampled on site in one cm slices (Yamoah et al., 2016). Approximately 380 samples were sieved under running water (mesh sizes 0.5 and 0.25 mm) to 133 recover plant macrofossils for radiocarbon dating. Plant remains were picked with tweezers under 134 a binocular microscope, described, and rinsed multiple times in deionized water, placed in pre-135 cleaned glass vials and dried overnight at 105 °C. 59 samples were dated at the 14Chrono Centre, 136 Queen's University Belfast, where pre-treatment and measurement followed the methodology 137 described in (Chawchai et al., 2015). Based on these, an age-model (Fig. S1) was constructed using 138 Bacon, a Bayesian statistics-based routine (Blaauw and Christen, 2011) that estimates the 139 accumulation rate for sediment segments based on the radiocarbon dates calibrated using the 140 intCal13 NH calibration curve (Reimer et al., 2013). Radiocarbon dates are given in Table S2.  Table S3. centrifugation. The process was repeated three times and supernatants were combined. Aliphatic 154 hydrocarbon fractions were isolated from the total lipid extract using silica gel columns (5% 155 deactivated) that were first eluted with pure hexane (F1) and subsequently with a mixture of DCM-156 MeOH (1:1 v/v) to obtain a polar fraction (F2). A saturated hydrocarbon fraction was obtained by 157 eluting the F1 fraction through 10% AgNO3 impregnated silica gel using pure hexane as eluent.

158
The saturated hydrocarbon fractions were analyzed by gas chromatography -mass spectrometry 159 for identification and quantification, using a Shimadzu GCMS-QP2010 Ultra. C21 to C33 n-alkanes 160 were identified based on mass spectra from the literature and retention times. The concentrations 161 of individual compounds were determined using a calibration curve made using mixtures of C21-162 C40 alkanes of known concentration and used to optimize the concentrations for compound-specific 163 isotope analysis. The hydrogen isotopic composition of n-alkanes (expressed in delta notation in ‰ against 167 VSMOW) was analyzed by gas chromatography-isotope ratio monitoring-mass spectrometry (GC-168 IRMS) using a Thermo Finnigan Delta V mass spectrometer interfaced with a Thermo Trace GC 169 2000 using the HTC reactor of a GC Isolink II and Conflo IV system. Helium was used as a carrier   following published protocols (Rattray and Smittenberg, 2020). Analysis was done using a Thermo-187 Dionex HPLC connected to a Thermo Scientific TSQ quantum access triple quadrupole mass 188 spectrometer, using an APCI interface. Chromatographic separation was achieved on a Kinetex 189 C18-XB reverse phase column using a gradient of mobile phase A: MeOH with 0.04% formic acid 190 and mobile phase B: propan-2-ol with 0.04% formic acid. GDGTs were detected in SIM mode at 191 m/z 1020 (scan width 7, 0.2s), 1034 (width 7, 0.2s), 1048 (width 7, 0.2s), 1296 (width 17.5, 0.5s).

192
Quantification was performed using Excalibur software, using the (M+) and (M+1) ions of the 193 GDGTs. More details can be found elsewhere (Rattray and Smittenberg, 2020   However, for high temperatures as is the case for our site, the main response to temperature is a 211 shift between tetra-and pentamethylated GDGTs, which makes the differentiation between 5-and 212 6-methyl GDGTs less relevant compared to colder environments. The relative abundance of tetra-, penta-and hexamethylated GDGTs plot in the same region as 217 datasets produced with the HILIC method from east African lakes and from global soils and peats 218 ( Fig S4). This strengthens the confidence that the brGDGTs we measured can be used as a The 18,000 year-long lake NTP sequence consist of organic rich gyttja with TOC contents ranging 285 between 10-40% (Fig. S2). TOC contents vary stepwise between 10 and 40% during the Late 286 Glacial part of the core, high TOC contents between 9.5-4.2 ka BP, turning to somewhat lower and 287 more variable contents over the last few millennia. Besides some variation caused by changes in 288 minerogenic input, we interpret the TOC changes as mainly caused by alternations between 289 meromictic conditions with permanent bottom water anoxia -leading to preservation of organic 290 matter, and monomictic conditions -resulting in greater organic matter oxidation within the 291 sediments. Stratification in tropical lakes is sensitive to small changes in the lake water level 292 between wet and dry seasons, heat budgets and climate (e.g., wind stress), and other limnological 293 or even ecological feedbacks (Lewis Jr, 1996). Given this multitude of factors, we do not attempt 294 to interpret the TOC content. Notably, there is no correlation between the variable TOC content 295 and the lipid biomarker proxies presented further below. This indicates that lake stratification and 296 preservation of organic matter did not influence the primary climatic signal of our proxy records.

297
The continuous occurrence of seeds of the aquatic plant taxon Najas ( Fig. 3f; SI Table 4   To further investigate past precipitation changes, we analyzed dDwax, with higher resolution 344 between 17-10 ka BP, to discern trends during deglaciation (Fig. 3a). dDwax was corrected for the  The unusual d 13 C excursion that starts at 16.0 ka BP suggests a renewed contribution of C4 362 vegetation to the carbon pool in this interval, even though the excursion is coincident with continued 363 warming and its onset correlates with a change in the rate of increase in atmospheric pCO2 (Fig. 3).    time. This long-term coupling between dD and MAATRC at orbital to millennial scales is opposite 465 to that of higher frequency relationships at annual to decadal scales (Fig. 2), where the total 466 insolation is distributed between latent and sensible heat. Orbital-scale changes in the seasonal 467 distribution of insolation apparently steer MAATRC and convective strength in the same direction.

468
The precessional cycle has indeed long been identified as the dominant component of orbital forcing provide further evidence for the influence of the precessional cycle on the isotopic composition of 490 regional precipitation, via the combined mechanisms of regional convective activity and associated 491 amount of precipitation. This is exacerbated by secondary effects of seasonality, which also affects 492 the distribution between latent and sensible heat. In the tropics there is a clear correlation between 493 insolation and rainfall amount (Fig. S7), with at present lowest values in June and July (Fig. 5) 494 . Over the course of a precessional cycle, the shift in seasonal distribution of 495 solar energy can be as much as 15%, which must be causing a large effect on seasonality. At and 496 near the equator, the 'dry' season may even have shifted from NH summer to SH summer (Fig. 5), 497 and the wettest season more towards or away from the March and September annual maximums, 498 depending on the orbital phase. Because of this we did not assign a wet season insolation curve to 499 the Tangga Cave record at Sumatra (Fig. 4). At our site lake Nong Thale Prong at 8°N, the present-day annual insolation curve exhibits two 516 highs: one in April and one in August/September (Fig. 6), when the sun's altitude is 90° at noon.

517
The annual movement of the ITCZ and the Monsoon system behaves in an attenuated fashion (Fig.   518   S5). From January onwards, temperatures rise (Fig. S6)  because the ITCZ remains south. Dry conditions with low cloud cover cause low albedo, resulting 520 in highest surface temperatures in April (Fig. S6). The ITCZ passes over quickly going northwards 521 during May and June, to merge with the Asian Summer Monsoon system during the NH summer 522 (Fig. S5) or is used to generate latent heat, leading to reduced surface temperatures (Fig. S6).

526
Between 6-4 ka BP, perihelium (the moment the earth is closest to the sun during its elliptical orbit) 527 occurred in September-October, causing 5% greater insolation in September compared to today 528 (Fig. 6). This stronger WSI for the SE Asian Monsoon and the northern IPWP will have caused between 6-4 kyr BP for the wettest period SON (See Fig. S6). The insolation curves have the same shape for higher 549 latitudes, but have different absolute values. The mainland SE Asian summer monsoon peaks in JAS, with highest 550 insolation between 10-8 kyr BP and very low insolation at the present. Note that the age axis is reverse compared to 551 proxy records. 552 553 Looking further back at 20 kyr BP, the seasonal pattern of insolation is similar as today (Fig. 7), Lateglacial conditions, resulting in an El-Niño-like mean state with extended dry seasons. They 625 attribute this mainly to orbital forcing, combined with a still much colder NH hemisphere causing 626 a much large temperature gradient between the tropics and the higher latitudes. The same factors 627 that lowered dDwax at our site (more rainout and more land-derived moisture from Sundaland, and 628 greater seasonality), must also have applied further inland. Remote processes upstream of the SE 629 Asian Monsoon, such as the presence / inundation of Sundaland, precession-forced changes in WSI 630 in the lower tropics, and seasonality, need to be considered when interpretating SE Asian water 631 isotope records in sediments and speleothems. Experiments with isotope-enabled general 632 circulation models are needed to gain further insight.

655
A main conclusion that can be drawn from our multi-proxy record is that seasonality is a major 656 factor that needs to be taken into consideration when interpreting climate and vegetation proxies 657 like d 13 Cwax and dDwax. Our data show that over the Late Glacial period the aerially exposed 658 Sundaland experienced a more continental and especially more seasonal climate than today with 659 biomass burning during dry winters, favoring C4 (Savannah) vegetation. This feature is most 660 apparent during the Bølling period that saw a rapid warming and strong increase in seasonal 661 precipitation conditions. A key turning point in a tug-of-war between pCO2, temperature and 662 seasonality as the three driving factors determining the ratio of C3 and C4 vegetation was the Older 663 Dryas event at 13.8 ka BP, after which climate evolved towards that of the year-round humid 664 climate known from the present day. Our Holocene record shows a clear mid-Holocene optimum 665 of deep convection in the northern IPWP, indicating that the 'steam engine of the world' was at full 666 power exporting greater amounts of (latent) heat, i.e. moisture, to the Northern Hemisphere during 667 this time. This declined over the last 5000 years, with dramatic effects starting in the Meghalayan 668 age at 4.2 ka BP where we find some evidence of severe droughts. Inferred from our own and from 669 other records, we argue that 'wet season' insolation (WSI), following the precessional cycle,   Tables S1-S8 are available at the Bolin Center for Climate Research database:   Figure S1. Age model of Lake Nong Thale Prong. Depth is expressed in meter below lake level. 983 984 985 Figure S2. Proxy records of lake Nong Thale Prong, with elements of Fig. 3 in the main paper, extended 986 with TOC content, bulk d 13 C, and carbonate content based on loss-on-ignition. 987 988 989 Figure S3. Comparison of the d 13 Cwax records of lake NTP (this study) and lake Towuti (Russell et al., 2014) and 990 atmospheric CO2 levels