Comparison of the U 37 K ' , LDI , TEX 86 H and RI-OH temperature proxies in sediments from the northern shelf of the South China Sea

The temperature proxies U37 K' , LDI, TEX86 H and RI-OH are derived from lipid biomarkers, namely long-chain alkenones from coccolithophorids, long-chain diols ascribed tentatively to eustigmatophytes, as well as glycerol dialkyl 15 glycerol tetraethers (GDGTs) and OH-GDGTs produced by Archaea, respectively. The applicability of these proxies in the South China Sea (SCS) has been investigated previously. However, in each study only one or two of the proxies have been compared, and the recently updated calibrations or new calibrating methods such as BASYPAR and BAYSPLINE have not been applied. Here, we investigate four proxies in parallel in a set of surface sediment samples from the northern SCS shelf and relate them to local sea surface temperature (SST), which allows us to compare and assess similarities and differences 20

Coastal seas are an ideal place for how the organic temperature proxies are influenced by the various confounding factors due to the environmental and ecological seasonality in the transition zone from shallow to deep-sea settings. In this study, we analysed alkenones, LCDs and GDGTs in surface sediments from the northern SCS shelf and continental slope, with a number of samples retrieved from locations shallower than 30 m. All of the proxies U 37 K' , TEX 86 H , LDI and RI-OH have been 75 previously studied in the northern SCS, a (sub)tropical monsoon climate region (Chen et al., 2018;Ge et al., 2013;Jia et al., 2012Jia et al., , 2017Lü et al., 2015;Wei et al., 2011;Yang et al., 2018;Zhang et al., 2013;Zhu et al., 2018). However, in these previous studies, only one or two of the proxies were investigated in parallel. Moreover, the updated calibrations or new calibrating methods such as Bayesian calibration models Tingley, 2014, 2018) need to be applied and examined in such a shallow coastal environment, where hydrography and nutrient dynamics are distinctly different from those in the 80 open northern SCS (Wong et al., 2015) and their influences on proxies are still incompletely known. Here, we investigate all four of the above-mentioned temperature proxies in the same surface sediment samples from this area and compare how they correspond to local SSTs. This effort is intended to help improving regional multi-proxy seawater temperature reconstructions, which could be more comprehensive and objective than those based on any single ones (Eglinton and Eglinton, 2008). In addition, such a kind of investigation can shed light on the ecology of the related biomarker producers in 85 this region, which is not entirely understood at present.

Study area and sample collection
The Pearl River estuary (PRE) and the northern shelf of SCS lie in a (sub)tropical monsoon climate region, with two contrasting monsoon seasons: the East Asian summer monsoon (EASM) and winter monsoon (EAWM). The EASM 90 generally lasts from May to September (Feng et al., 2007;Wang et al., 2004) and the EAWM typically from December to next February (Koseki et al., 2013;Wu, 2016). During the EAWM periods, northeasterly winds are strong, cold and dry, leading to surface cooling, intensified vertical mixing, and hence an increase of upward nutrient supply to surface waters (Tseng et al., 2005). During the EASM periods, warm southwesterly winds rich in moisture, induce wet conditions, coastal upwelling (Jing et al., 2009) and enhanced freshwater input from the Pearl River, the second largest river in terms of 95 discharge in China. The outflow of the Pearl River delivers large amounts of nutrients to the coastal northern SCS (Yin et al., 2000).
The northern SCS is a typical oligotrophic sea, characterized by nitrogen-limited conditions like most open oceans (Chen and Chen, 2006), while the Pearl River water is rich in nutrients characterized by a nitrogen/phosphorus (N/P) ratio >100 (Xu et al., 2008;Yin and Harrison, 2008). Thus different nutrient regimes prevail in the northern SCS: primary production is 100 stimulated by discharge of the Pearl River on the inner shelf during the EASM (Chen and Chen, 2006); in the open basin and on the shelf during the EAWM, nutrients are supplied by vertical mixing (Wong et al., 2015).
In the study area, a total of 23 core top sediments (0-1 or 0-2 cm depths) were collected between 2011 and 2017 (Table S1) from the PRE and the northern SCS, from water depths (WD) ranging from 6.5 to 1307 m (Table S1). Most (n =15 of 23) of them were recovered from the inner shelf (WD <50 m), seven from the outer shelf  and one from the 105 continental slope (LD-21, WD = 1307 m) (Fig. 1a). The samples were collected using a gravity box corer or grab sampler and then stored frozen at −20 °C in the laboratory before treatment.

Lipid extraction and separation
After freeze-drying and homogenizing, about 5 g of sediments were ultrasonically extracted three times with DCM: MeOH (9:1, v/v) for 15 min. Before extraction, known amounts of 2-nonadecanone, androstanol and C46 GDGT were added as 110 internal standards. Supernatants of each extraction were obtained by centrifugation. The total lipid extracts were combined and concentrated with rotary evaporation to ~1 mL, and saponified for 2 h at 80 °C with 1 mL of KOH (0.1 M) in MeOH: H2O (9:1, v/v). Saponification has been suggested as a crucial clean-up procedure for eliminating interferences from coeluting wax esters during instrumental analysis of alkenones (Villanueva et al., 1997). The sample pre-treatment we used were also used by some of the participants in the interlaboratory comparison of TEX86 analytical methods, where extraction 115 6 procedures were not found to exert significant and systematic effects on TEX86 results (Schouten et al., 2013b). The neutral fractions were extracted with n-hexane, and were further separated into alkane, alkenone and alcohol sub-fractions (the latter containing diols and GDGTs) by column chromatography on silica gel using n-hexane, DCM: n-hexane (2:1, v/v) and DCM: MeOH (1:1, v/v), respectively.

Alkenone analysis and U 37 K' index 120
Alkenones were analysed using a 7890A gas chromatograph (GC, Agilent Technologies) equipped with a cold on-column injection system, a DB-5MS fused silica capillary column (60 m, ID 250 µm, 0.25 µm film coupled to a 5 m, ID 530 µm deactivated fused silica precolumn) and a flame ionization detector (FID). Helium was used as carrier gas (constant flow, 1.5 mL/min) and the GC oven was heated using the following temperature program: 60 °C for 1 min, 20 °C/min to 150 °C, 6 °C/min to 320 °C and a final hold time of 35 min. Di-unsaturated (C37:2) and tri-unsaturated (C37:3) alkenones were 125 identified by comparison of the retention times with a reference sample composed of known compounds (Fig. S1a). Peak areas were determined by integrating the respective peaks, and concentrations were calculated using the response factor of the internal standard 2-nonadecanone.
The U 37 K' index was calculated using Eq. (1) after Prahl and Wakeham (1987 (1) 130 SST was estimated using the calibration of Müller et al. (1998) with an uncertainty of 1.5 °C : In addition, a Bayesian calibration (BAYSPLINE, Tierney and Tingley, 2018) was also applied, as our annual SSTs were above 24 °C . The analytical uncertainty of U 37 K' index (0.01) was determined from multiple extractions and analyses of a labinternal reference standard sediment, which was co-analysed with samples for half a year (n = 24). 135

Long chain diol analysis
One half of each alcohol fraction was silylated with N, O-bis(trimethylsilyl)-trifluoroacetamide (BSTFA)/1% trimethylchlorosilane (TMCS) and acetonitrile (30 µL each) and heated at 60 °C for 1 h. Diols were analysed by gas chromatography-mass spectrometry (GC/MS) on an Agilent 6850 GC coupled to an Agilent 5975C MSD operating in electron impact (EI) mode with an ionization energy of 70 eV. The GC was equipped with a fused silica capillary column 140 (Restek Rxi-1ms, length 30 m; 250 µm ID, film thickness 0.25 µm). Helium was used as carrier gas at a constant flow rate of 1.2 mL/min. Samples (1 µL) were injected in splitless mode in a split/splitless injector (S/SL) held at 280 °C. The GC temperature program was as follows: 60 °C start temperature, held for 3 min, increased to 150 °C at a rate of 20 °C/min, increased further to 320 °C at a rate of 4 °C/min and finally held at 320 °C for 15 min. The source temperature of the MS was set to 230 °C and the quadrupole to 150 °C. 145 For identification of the diols, the MS was operated in single-ion monitoring (SIM) mode with the following m/z: 313.3 (C28 1,13-diol, C30 1,15-diol), and 341.3 (C30 1,13-diol, C32 1,15-diol) (Fig. S1b;Versteegh et al., 1997;Rampen et al., 2012).
Fractional abundances of the diols were calculated from their integrated peak areas in the respective mass chromatograms.
The LDI was calculated and converted to SST using Eq. (3) and Eq. (4) from Rampen et al. (2012)   The %C32 1,15 index reflecting riverine input was calculated using Eq. (6) given by Lattaud et al. (2017)  During the time of analyses, there was no reference sample for diol measurement in our lab, so the analytical uncertainty of LDI is could not be determined.

GDGT analysis and indices (TEX86, , BIT, MI, RI, and RI-OH) 160
GDGTs were analysed by high performance liquid chromatography (HPLC) coupled via an atmospheric pressure chemical ionization (APCI) interface to a single quadrupole mass spectrometer (MS), with a method slightly modified from Hopmans et al. (2016). Analyses were performed on an Agilent 1200 series HPLC system and an Agilent 6120 MSD. Separation of the individual GDGTs including the 5-/6-methyl isomers of branched-GDGTs was achieved on two UPLC silica columns in series (Waters Acquity BEH HILIC, 2.1150 mm, 1.7 µm), with a 2.15 mm pre-column of the same material maintained at 165 30 °C. Mobile phase A and B consisted of n-hexane: chloroform (99:1, v/v) and n-hexane: 2-propanol: chloroform (89:10:1, v/v/v), respectively. After sample injection (20 µL) and 25 min isocratic elution with 18 % mobile phase B, the proportion of B was linearly increased to 50 % within 25 min, and thereafter to 100 % for the next 30 min. After another 5 min and prior to the analysis of the next sample, the column was re-equilibrated with 18 % B for 15 min. The flow rate was 0.22 mL/min and a maximum back pressure of 220 bar was obtained. The total run time was 100 min. 170 GDGTs were detected using positive ion APCI-MS and selective ion monitoring (SIM) of their (M+H) + ions  or abundant ion-source fragmentation products of OH-GDGTs (Liu et al., 2012). APCI spray-chamber conditions were as follows: nebulizer pressure 50 psi, vaporizer temperature 350 °C, N2 drying gas flow 5 L/min and 350 °C, capillary voltage (ion transfer tube) −4 kV and corona current +5 µA. The MS-detector was set for SIM of the following (M+H) + ions: Quantification of the individual GDGTs was achieved by integrating the respective peak areas. Compound contents were calculated using the response factor obtained from the C46 standard and by normalizing to the amount of extracted sediment. 180 Due to the lack of appropriate standards, individual relative response factors between the C46 standard and the different GDGTs could not be considered, the obtained concentrations should therefore be regarded as being only semi-quantitative. We also examined the Bayesian, spatially-varying calibration (BAYSPAR, Tierney and Tingley, 2014) for the TEX86 as well 190 as a regional winter calibration based on suspended particulate matter (SPM) in the SCS (calibration error of the latter: where Ia is the basic tetramethyl brGDGT; IIa and IIIa are 5-methyl brGDGTs; IIa' and IIIa' are 6-methyl brGDGTs (DeJonge et al., 2013).
The methane index (MI) was calculated using the Eq. (12) given by Zhang et al. (2011): The ring index (RI) was calculated using the Eq. (13)  200 The RI-OH index was calculated using Eq. (14) from Lü et al., (2015), with an uncertainty of 0.01 determined from a labinternal reference standard sediment, which was repeatedly extracted and co-analysed with samples for 3 months (n = 20). Lü et al. (2015) presented sedimentary OH-GDGTs data from the China marginal sea (CMS), including some from the 205 northern SCS. In their data, RI-OH correlated best with the summer SST (R 2 = 0.87). Besides, a recent observation in the ECS showed that OH-GDGTs abundance in surface water in summer were two times higher than that in winter (Lü et al., 2019), suggesting higher OH-GDGTs production in summer. Thus, the summer calibration (Eq. (15), calibration error: 0.9 °C) from Lü et al. (2015) was applied:

Climatological mean temperature data and temperature residuals of proxies
The sedimentation rates are not exactly known for each sampling site but sedimentation rates have been reported to vary spatially from 0.2 to 0.6 cm yr -1 (Ge et al., 2014;Liu et al., 2014) in the study region. Accordingly, the 0-2 cm surface sediments represent accumulation of more or less a decade. Considering the age uncertainties, we extracted mean annual and monthly SST data for each sampling site, as well as surface salinity in the study region, during an available decadal period 215 (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) from the NOAA World Ocean Atlas 2018 (WOA18) on a 0.25° grid resolution (https://www.nodc.noaa.gov/OC5/woa18/woa18data.html). Even though a linear trend (0.031 °C yr -1 ) of SST warming has been reported for the SCS (Yu et al., 2019), a different choice of a reference interval would not result in significantly different mean values. The grid resolution of 0.25° in the database is sufficient to define the climatology of the study region, as the distances between 19 out of 23 sampling sites are >0.25° (Fig. 1a). As the regional climate feature are dominated by 220 the seasonally reversing monsoon winds and the transitions between the two contrasting seasons, i.e. from October to November and from March to April, respectively, are relatively short, the SST data were re-analysed and averaged for the two dominant seasons, i.e. EASM (May to September) and EAWM (December to February). Besides, monthly satellite Chlorophyll-a (Chl-a) L3 data were obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) between 2005 and 2017, and average Chl-a values in the EASM and EAWM seasons were calculated according to above definition of 225 seasons.
In our study region, SSTs varied spatially within a small range of about 5.5 °C and 3.5 °C during the EAWM and EASM periods, respectively. Together with non-thermal impacts on SST-proxies, such a narrow range usually leads to poor SSTproxy correlations. Thus, we did not use correlation as a criterion to investigate the preferred season of growth of the biomarker producing organisms. Instead, we considered temperature residuals between calculated temperatures from 230 established calibrations and WOA18-derived SSTs, calculated as: 3 Results

Hydrological and Chl-a distributions
The annual mean SSTs of sampling sites from WOA18 dataset ranged between 24.  (Table S1). The mean SSTs displayed an increasing trend offshore with the largest difference of ca. 5.5 °C in the EAWM season between inshore and offshore ( Fig. 2a).
Surface salinities were generally high at ~34 and uniformly distributed in the study area during the EAWM season, with 240 slightly lower values of ~33.5 along the coastline. While during the EASM season, surface waters freshened due to high precipitation and elevated freshwater discharge from the Pearl River, causing a salinity gradient offshore with the lowest values (<32) in the PRE and the highest (~34) to the east of Hainan island (Fig. 3b). Surface Chl-a levels were clearly higher on the inner shelf than on the outer shelf during both the EAWM and EASM seasons (Fig. 3c, 3d). Chl-a concentration also exhibited a seasonal contrast, which, however, was different between the inner shelf and the outer shelf: higher Chl-a 245 occurring in the EASM season on the inner shelf, and in the EAWM season on the outer shelf ( Fig. 3c, 3d).
Fractional abundances of each LCDs from different samples were set as variables, the strength and direction of association that exists between two variables is determined as the PCC, denoted as r. The results showed that C28 and C30 1,13-diols and C32 1,15-diol were significantly correlated with each other (r: 0.56-0.83, p <0.005, Table 2). In contrast, these three diols 265 were negatively correlated with C30 1,15-diol (r: −0.68 to −0.90, p <0.005, Table 2), with the latter exhibiting an opposite distribution pattern and showing an overall increasing trend towards the offshore (Fig. 4c).
with high values at shelf and coastal sites (WD ≤186 m) (Fig. 5l), but a low value at the slope site (LD-21). In contrast, the relative abundance of OH-GDGT-0 ([OH-0]) remained low at shelf and coastal sites, but was elevated at the slope site (Fig.   5j).

Seasonality of the U 37 K' proxy
Although the relationship between U 37 K' and SST is robust and well supported by culture studies (Conte et al., 1998;Prahl and Wakeham, 1987;Prahl et al., 1988;Sawada et al., 1996;Volkman et al., 1995), the U 37 K' response to SST has been found to be attenuated in warm environments (>24 °C), with the slope of the regression decreasing by nearly 50 % as U 37 K' approaches 310 unity (e.g., Conte et al., 2006, Sonzogni et al., 1997, Tierney and Tingley, 2018. In the northern SCS, annual SSTs are generally >24 °C ; however, non-linear calibrations for U 37 K' have not been applied in previous studies. The BAYSPLINE (Tierney and Tingley, 2018) is the latest non-linear calibration, the application of which in this study showed that it yielded temperatures similar (within 0.7 °C ) to the linear calibration by Müller et al. (1998). Considering the errors of the linear calibration (±1.5 °C) and the BAYSPLINE calibration (up to ±2 .5 °C , 1σ), there is no difference between the two sets of 315 SST estimates (Fig. 2a). Also, Pelejero and Grimalt (1997) analysed a series of core-top sediments in the SCS basin and 13 found good linear correlations between U 37 K' and averaged SSTs of various depths (0, 10, 20, and 30 m) and seasons, indicating that the linear relationship between U 37 K' and SST is still maintained in such a warm environment. This supports the above finding that there is insignificant difference between SST estimates of linear and non-linear calibrations. Nonetheless, most U 37 K' -derived temperatures were slightly higher than annual mean SSTs, suggesting a seasonal bias to the EASM season 320 (Fig. 2a), especially for samples recovered from WD ≤100 m. Based on a study of a inshore-offshore transect between 33 m and 102 m WD, Zhang et al. (2013) also proposed U 37 K' -SST to be spring-and summer-biased (April-August) in this region.
In surface waters of the SCS outer shelf, the coccolithophore E. huxleyi, a major alkenone producer, has been shown to be most abundant in the monsoon transition periods, such as in October (46 × 10 3 cells L −1 ) and March (19 × 10 3 cells L −1 ), somewhat less abundant in July dominated by EASM (4 × 10 3 cells L −1 ) and least abundant in January dominated by EAWM 325 (2 × 10 3 cells L −1 ) (Chen et al., 2007). The lowest abundance of coccolithophores in winter, when Chl-a is elevated (Fig. 3c) due to enhanced mixing, likely results from their competitive disadvantage relative to diatoms (Chen et all., 2007). As the SSTs in the monsoon transition periods are close to the annual mean SSTs, the above seasonal changes in the abundance of E.
huxleyi support our view that U 37 K' reflects annual mean SST with a slight bias toward the warm season. Nonetheless, we note that the bias is not significant. 330 However, on the SCS shelf, spatial and temporal distributions of alkenone producers have not been carefully investigated, especially on the inner shelf, where high surface Chl-a levels occur in the EASM season (Fig. 3d). During this period, surface water salinities are relatively low, mainly due to the discharge of the Pearl River. The river water is enriched in nutrients, the impact of which on primary production, however, is largely limited to the areas within the PRE and along the coast (Fig. 3d). In addition, the nutrient distribution in the river is characterized by high N:P ratios of up to ~100:1 (Dai et al., 335 2008;Lu and Gan, 2015;Xu et al., 2008;Zhang et al., 2013). We surmise that such an input of an unbalanced nutrient ratio could stimulate the growth, even though not prominent blooms, of alkenone-producing haptophytes, e.g., E. huxleyi, in the oligotrophic shelf waters during the EASM period, since both in situ investigations and experiments have reported that E.
huxleyi have a competitive advantage over other phytoplankton at high N:P ratios (Riegman et al., 1992, Tyrrell andTaylor, 1996). This phenomenon is likely because the species has a great activity of the enzyme alkaline phosphatase, facilitating 340 assimilation of dissolved organic phosphates (Bijma et al., 2001).

Source of LCDs in the surface sediments
The unusually low LDI-derived SST estimates relative to the WOA18-derived SSTs were observed close to the river mouth and on the inner shelf (Fig. 4e) suggest that LDI may be influenced by terrestrial/freshwater sources other than marine 345 producers. Similar findings were reported from the Iberian margin (de Bar et al., 2016), the Gulf of Lion, the Berau Delta, the Kara Sea (Lattaud et al., 2017) and the East China Sea (He et al., 2020), suggestive of terrestrial influence on LCDs compositions. Culture studies show that marine eustigmatophyte algae mainly produce 1,13 and 1,15-diols (Rampen et al., 14 2007(Rampen et al., 14 , 2014aVolkman et al., 1999). In freshwater environments, eustigmatophyte algae primarily produce C32 1,15-diol, especially in stagnant waters during dry seasons, when rivers have low-stands (Hä ggi et al., 2019;Lattaud et al., 2017;350 Rampen et al., 2014b). However, C30 1,15-diol is generally found to be dominant both in the marine water column and sediments and are likely produced by marine eustigmatophyte algae (Balzano et al., 2018).
In this study, the co-occurrence of high abundance of C28 and C30 1,13-diols and C32 1,15-diol in the PRE and on the inner shelf rather than in the offshore area (Fig. 4a, 4b, 4d) is consistent with the PCC analysis (Table 2), further suggesting a terrestrial/freshwater source of these diols. Such a spatial distribution pattern becomes more apparent when diol 355 compositions in SPM and sediments are illustrated from the PRE to the offshore (Fig. 4g). In contrast, the negative correlation of C30 1,15-diol with three other diols could be attributed to their different main sources, i.e. marine vs. terrestrial.

Influence of riverine LCDs
It has been pointed out that LCDs delivered by rivers can substantially affect LDI temperature estimates in coastal regions close to river mouths (e.g., Lattaud et al., 2017;He et al., 2020). Lattaud et al. (2017) pointed out that %C32 1,15 in the 360 typical marine sediments generally does not exceed a value of 20 %, which may be used as a cut-off for the reliable reconstruction of LDI-SST, and %C32 1,15 >20 % implies an increased contribution of riverine LCDs. In our samples, LDIderived temperature estimates from two calibrations were similar to the measured annual SSTs at most sites (Fig. 2b), with 6 exceptions at shallow sites (<26 m) in the PRE and on the inner shelf showing temperature values underestimated by as much as −11.0 ± 2.0 °C (Fig. 2b, 4e). We found that the greater underestimations corresponded to %C32 1,15 values that 365 are >20 % and 4 times higher than those of the other samples, and the samples with %C32 1,15 <20 % had smaller annual residuals ranging between −0.2 ± 2.0 °C and 1.2 ± 2.0 °C (Fig. 4e, Table S3). Besides, the %C32 1,15 values correlated positively (R 2 = 0.66, p <0.001) with the BIT index that is often used to indicate terrestrial input in the coastal area (Fig. 4f).
Thereby %C32 1,15 is also effective to indicate the river input in this region. After removal of data points (n = 6) with %C32 1,15 >20%, indicating significant influence of riverine LCDs, the LDI-SST of the reduced data set yields a mean annual 370 residuals of 0.3 ± 0.4 °C, much lower than those (1.3 ± 3.3 °C) of the full data set.

Seasonality of LDI index
Our results indicate that LDI-SSTs at sites with minimal river influences may reflect annual SSTs (Fig. 2b, Table S3), suggesting unbiased seasonal production of the source organisms of LDI in this study area. Similar results have been reported by Zhu et al. (2014), who found that LDI-SSTs in downcore sediments match well with the local annual SSTs in the 375 northern SCS. Rampen et al. (2007) found comparable annual flux of 1,15-diols at different stations in the Arabian Sea, and suggested that the biological producers of 1,15-diols do not require a high level of nutrients as needed, e.g., by Proboscia diatoms producing 1,14-diols. Thus, LDI may reflect annual SST, with low seasonal abundance variations of marine eustigmatophytes in spite of nutrient variations in an annual cycle on the northern SCS shelf. Nonetheless, since regional annual SSTs are indistinguishable from the monsoon transition periods in spring and/or autumn, we cannot rule out the 380 possibility of prominent occurrences of marine LCD producers during these transition periods.

Sources of iGDGTs in the surface sediments
In marine sediments, the iGDGT composition may sometimes be impacted, or even controlled by non-thermal factors, e.g., sources of iGDGTs other than Thaumarchaeota . Several indices, e.g., the MI (Zhang et al., 2011), BIT 385 (Hopmans et al., 2004), the [2]/[Cren] ratio , and the RI  have been developed to assess these impacts. Relatively low MI values (≤0.25) were observed at most sites in our study accompanied by low  (Fig. 5g) is similar to data reported by Zhang et al. (2012). As the BIT index in soils generally tends to be >0.9 (Hopmans et al., 2004), the highest BIT value at the site likely indicates a significant input of soil-derived GDGTs. However, the ability of the BIT index to indicate soil input in this region has been recently discounted 395 by the finding that branched GDGTs may be produced in-situ in aquatic systems . Nevertheless, considering that the sample PRE-A8 is located at the upper river mouth and shows the highest %C32 1,15 values as discussed above, we believe that iGDGTs at this site may be impacted to some extent by terrestrial input.
The [0]/[Cren] ratio was also high at the site PRE-A8. This is likely associated with river input, as the [0]/[Cren] ratio has been found to be high (>2) in soils and river sediments likely due to in-situ methanogenic archaea or imported soil-derived 400 methanogens Zhu et al., 2011). Slightly different from other iGDGTs, [Cren'] increased with increasing water depth, with the lowest value of 0.8 % found in the PRE (PRE-A8) and ~1.0 % close to the PRE, while it amounted to 5.4 % at the deepest site (LD-21) (Fig. 5f), in agreement with findings from Jia et al. (2017), who report [Cren'] of >4 % in deep-sea sediments in the SCS. This pattern was unlikely caused by input of soil iGDGTs, as [Cren'] in the soils in the catchments of the Pearl River is ~3 % . [Cren'] as low as 0.2-0.7 %, with a mean of 0.4 %, was observed 405 in the SPM of the lower Pearl River, which was attributed to the predominance of Euryarchaeota . This suggests that iGDGTs close to and within the PRE could also be impacted by the input from aquatic archaea other than Thaumarchaeota.

Thaumarchaeota. The occurrence of low [2]/[3] ratios and low [Cren'] fractional abundances for most of our study sites is in
agreement with the shallow water depths of these sites, as the depth boundary to separate the deep and shallow Thaumarchaeota, although not exactly determined, is likely 200-300 m (Jia et al., 2017;Kim et al., 2015Kim et al., , 2016. Theoretically, if planktonic archaea are the dominant GDGT producers, the RI values calculated using fractional abundances of all iGDGTs reflect a response to growth temperatures similar to TEX86. This results in a positive correlation between the 420 two indices. Accordingly, Zhang et al. (2016) presented the TEX86-RI relationship of the global core top data set, which they proposed to be used as a criterion to evaluate whether the TEX86 value of a given sample is influenced by non-thermal factors. We found that most of our sediment data show a good correlation between TEX86 and RI (Fig. 6a); however, they lie outside of the 95 % prediction band using the global TEX86-RI relationship (Fig. 6b, Zhang et al., 2016), but, with the exception of two samples, within the 95 % prediction of a "shallow-water" TEX86-RI relationship (Fig. 6b, Jia et al., 2017). 425 The two exceptional samples (LD-21 and PRE-A8) are thus likely influenced by other factors than temperature as discussed above. We suggest that this "shallow-water" TEX86-RI relationship that is different than that of the global core-top data set is a robust feature. Our study sites receive predominantly shallow Thaumarchaeota input as demonstrated above, and the shallow Thaumarchaeota likely responds to ambient temperature differently from the deep dwelling communities (Jia et al., 2017;Kim et al., 2015Kim et al., , 2016Taylor et al., 2013;Villanueva et al., 2015;Zhu et al., 2016). Similarly, the TEX86 and RI 430 values from an incubation study of marine Thaumarchaeota  are correlated but lie outside of the 95 % prediction band of the global relationship, likely due to differences in the archaeal community between the incubation experiment and natural marine settings . Together, this indicates that TEX86 is suitable for temperature estimation in our study area and TEX86 in most of our sediments likely indicate regional seawater temperatures.

Seasonality of TEX 86 H index 435
Based on the above discussion on iGDGTs indices, only two samples, one in the PRE (PRE-A8) and the other on the slope (LD-21), are markedly different from the remaining samples that appear minimally influenced by soil/freshwater-derived archaea and deep-dwelling Thaumarchaeota or methane-cycling archaea. We therefore exclude these two samples from the following examination of temperature signal recorded by the TEX 86 H index.
Our TEX 86 H -SST estimates were 1.0 ± 2.6 to 8.8 ± 2.6 °C lower than annual SST using the calibration of Kim et al. (2010), 440 similar to previous studies. The temperature estimates were even lower than the coldest monthly SSTs in the shelf area between 10-100 m WD (Fig. 6c, Table S4). The BAYSPAR estimates yielded slightly higher SSTs, with annual residuals being reduced by ~1.0 °C , however, they are still lower than the coldest monthly SSTs (Fig. 6c, Table S4). When compared with the mean EAWM SSTs, the residuals of both calibrations ranged from −5.4 ± 2.6 °C to 1.9 ± 2.6 °C (Kim's calibration) and from −3.7 ± 2.3 °C to 1.6 ± 1.8 °C (BAYSPAR), respectively (Fig. 6c, Table S4). As these residuals are not much larger 445 than the calibration error, it may be inferred that TEX86 proxy on the northern SCS shelf reflects SST during the coldest season. Similar conclusions have been drawn in several previous studies of TEX86 in the northern SCS (Ge et al., 2013;Wei et al., 2011;Zhang et al., 2012;Zhou et al., 2014). Support for this inference comes from a recent observation of iGDGTs abundance in surface waters of the SCS shelf, which in winter were three times higher than in summer (Jia et al., 2017). Furthermore, we noted that different from the global dataset utilized to establish the TEX 86 H -SST or TEX86-SST, which 450 include a large number of deep-sea sediment samples, our data here were exclusively from shallow sediments receiving iGDGTs predominantly from shallow dwelling Thaumarchaeota. The global calibrations might not be suitable for temperature estimation in our study, as indicated by the different TEX86-RI relationship of our data from the global relationship as discussed above (Fig. 6b) and the fact that TEX86-derived temperatures are even lower than observed SSTs in the coldest month. Therefore, a local "shallow-water" calibration could be more appropriate for temperature reconstruction. 455 Accordingly, the calibration established from winter SPM (i.e. Eq. (10), Jia et al. 2017) in surface waters of the SCS was applied here. This calibration indeed yielded temperatures closer to the EAWM SSTs ( Fig. 2c) with reduced residuals and calibration errors (−2.8 ± 1.3 °C to 1.7 ± 1.3 °C) (Fig. 6c, Table S4). But it is obvious that some of temperature estimates are still slightly below SSTs in the coldest month. This occurrence has been observed around the PRE and was attributed to the minor contributions of iGDGTs 1 to 4 from MG-II Euryarchaeota . However, it is still in debate 460 whether MG-II Euryarchaeota can produce iGDGTs or not (e.g., Lincoln et al., 2014;Schouten et al., 2014;Besseling et al., 2020;Ma et al., 2020) due mainly to lack of cultured representatives of MG-II Euryarchaeota presently.
The relatively closer association of TEX86 temperature estimates with EAWM SSTs than EASM and annual SSTs suggests that conditions during the EAWM period may be favourable for the bloom of the autotrophic ammonia oxidizing Thaumarchaeota, the activity of which is enhanced at low light availability and high ammonia concentrations (Horak et al., 465 2018). At present data on seasonal variations of seawater ammonia in the study region are not available. Water column light levels in the EAWM season are generally low due to the reduced solar irradiation, which may foster a preferential occurrence of Thaumarchaeota during the EAWM season, and hence lead to a bias of TEX86 temperatures toward EAWM SSTs.

Source of OH-GDGTs and their influences on RI-OH-SST estimates
A few studies have detected OH-GDGTs in marine, river, lacustrine and soil environments, indicating ubiquitous and multiple sources of OH-GDGTs (Chen et al., 2016;Huguet et al., 2013;Kang et al., 2017;Liu et al., 2012b;Park et al., 2019;Wang et al., 2012). Kang et al. (2017) noted that [OH-0] (OH-GDGT-0) dominates in marine and estuarine environments (56 ± 10 %), but [OH-2] (OH-GDGT-2) is abundant in lake, river and soil environments, which may lead to overestimated RI-475 OH-SSTs in case of substantial terrestrial input. Consistently, we found higher RI-OH-SST than EASM SST in the PRE (Fig.  2d, Table S5), where terrestrial input is significant. Besides, at site PRE-A8, its iGDGT composition has been found to be influenced by terrestrial input (see section 4.3.1), which also appears to have an impact on OH-GDGT composition.
In addition, like the Thaumarchaeota, the source organism of OH-GDGTs might also exhibit different thermal responses, namely the OH-GDGTs composition of their membrane lipids, between "shallow-water" and "deep-water" communities. 480 Here, we combined our data with previously published sedimentary OH-GDGTs data in the SCS, with water depths ranging between 3 m and 4405 m (Lü et al., 2015;Yang et al., 2018). We found that [OH-0] is psositively correlated with WD (R 2 = 0.66, p <0.001, Fig. 5j), but [OH-2] correlated negatively with WD (R 2 = 0.53, p <0.001, Fig. 5l). Meanwhile, except one sample (WD = 41 m), two clusters of samples can be separated based on the [OH-0]/[OH-2] ratio, with the ratio value <0.55 for the shallow-water samples (WD <200 m) and >0.55 for the deep-water samples (WD >200 m) (Fig. 7a). This is 485 surprising because the deep-water sediment samples were collected at warmer, lower lattitudes on the slope and in the basin of the SCS , which should induce more abundant  according to eq. (14). We thus speculate that "shallow-water" and "deep-water" communities have different OH-GDGTs compositions, with more  in the deepwater community.
Recently, Yang et al. (2018) found that  is positivly correlated with SSTs at SST <25 °C using a modified Bligh/Dyer 490 extraction method, but this relation is inversed at higher SSTs (>25 °C ). They therefore proposed a different thermal response of archaeal OH-GDGTs at higher temperatures. However, progressive regression analysis of annual SST with , as well as with RI-OH, on our data sequentially removed the outliers that lie outside of the 95 % prediction bands of the respective calibrations and showed that both [OH-2] and RI-OH were positively correlated with SSTs (Fig. 7c, 7d). The annual SSTs of most (n = 11 of 13) data points laying within 95 % prediction were also above 25 °C. We explain the fact 495 that we reach different conclusions than Yang et al. (2018) by the different water depths at which samples considered in the analyses were recovered. Most (n = 17 of 23) samples of Yang et al. (2018) were located in the deep (WD >971 m) basin of the SCS, and their geographical distribution led to an apparent SST increase with WD (Fig. 7b).
The  and RI-OH of seven samples were identified as outliers in our progressive regression analysis (Fig. 7c, 7d).
Three of them correspond to samples taken in the PRE (PRE-A8, PRE-Y6, and PRE-Y11), and one to the samples from the 500 slope (LD-21), similar to where outliers in the iGDGT distribution were recorded. However, the three other outliers (QD00,  cannot be explained at present. Unlike iGDGTs, there are no indices developed to assess the impact from non-thermal factors on OH-GDGT distributions.

Seasonality of RI-OH index
After excluding the seven outliers identified above, we found that temperature estimates using the summer RI-OH-SST 505 calibration (i.e. (15)) correspond well with EASM SSTs on the shelf of the northern SCS (Fig. 2d), with an average residual of 0.0 ± 1.1 °C (Table S5). In comparison, if annual and winter calibrations by Lü et al. (2015) were used, the standard errors of residuals would be 2.3 °C and 3.2 °C, respectively (data not shown here), indicating that the summer calibration provides better estimates. If RI-OH is considered to reflect EASM SSTs rather than annual or EAWM SSTs, which likely indicates that the source organisms proliferate mainly during the EASM season. Such a conclusion is similar to the observation by Lü 510 et al. (2019), who showed that OH-GDGTs in surface water SPM were more abundant in summer than in winter in the ECS.
Comparatively, as discussed above, TEX 86 H is biased to EAWM SSTs in this region. This may indicate that OH-GDGTs and iGDGTs originate from different organisms. However, the source of OH-GDGTs has not been identified yet, and thus more studies on OH-GDGTs in various regions are needed for a better assessment of the proxy.

Implication for paleoclimatic reconstruction 515
After excluding samples with obvious signs of terrigenous supply of the respective lipids, we observed close association between measured annual mean or seasonal SSTs and temperature estimates based on the four proxies discussed here. The relatively poor performance of TEX86 in this setting may result from more complicated processes that needs further investigations. Overall, the good agreements between measured SSTs and temperature estimates suggest that resuspension and lateral transport have only minor impacts on the lipid biomarkers in our study area. 520 The reconstruction of EASM and EAWM, being controlled respectively by processes occurring in the tropical Indian-Pacific oceans and in high-latitude Siberia, is a prerequisite for the understanding of paleoclimate change in East Asia. As the strengths of the EAWM and EASM appear to be anticorrelated at least during the most recent geological history (e.g., Yancheva et al., 2007), the possibility of reconstructing seasonal temperatures will greatly advance the understanding of this system. Our comparison of SST proxies reveals their differential seasonal biases and thus reveals a promising multi-proxy 525 approach to reconstruct EASM and EAWM separately. The coastal and inner shelf of the SCS can provide fine sediment archives accumulated since the early Holocene (Yim et al., 2006;Ge et al. 2014;Gao et al., 2015), which have been retrieved and studied extensively in recent years including the SST reconstructions mostly using alkenones (Kong et al., 2014(Kong et al., , 2017Lee et al., 2019;Zhang et al., 2019 warm-biased (RI-OH) temperatures relative to annual mean SSTs. After excluding from the data set the samples subject to 540 terrestrial input, the temperature estimates based on these proxies could be ascribed to different seasons, which reflects distinctive ecologies of their source organisms as results of seasonal changes in environmental conditions. LDI-SST matched well with annual SSTs, suggesting that marine eustigmatophyte abundance does not vary strongly with nutrient variation in an annual cycle. For the TEX 86 H proxy, a local "shallow-water" calibration based on winter surface water SPM in the SCS appeared to be more appropriate for temperature reconstruction and reduced residuals relative to SSTs in the EAWM season, 545 although an additional cold bias of temperature estimates still exists. In contrast to TEX 86 H indices, RI-OH-based temperature estimates seem to reflect EASM SSTs, hence suggesting a different source organism of OH-GDGTs from that of iGDGTs.
As these proxies appear to reflect preferentially different seasons, their combined use has the potential to allow reconstructing seasonal SSTs controlled separately by the EASM and EAWM, which may improve our understanding of the evolution of the East Asian climate system. 550

Acknowledgements
This work is supported by the National Natural Science Foundation of China (grant No. 41676030)

Author contribution 560
GJ and GM conceived and designed the study. BW and MK collected the samples. BW conducted all the proxy analysis and was aided by JH in the instrument maintenance and data analysis. BW wrote the paper with inputs from GJ, GM, JH, EP, and SW. All the authors reviewed the final manuscript.

Supplement
There is a supplement related to this article. 565 Table 1: Sampling sites, SST and proxy values from the PRE and northern SCS shelf (water depth (WD), sea surface temperature (SST) obtained from WOA18, RI-OH, ring index of OH-GDGTs).