Performance of long-chain mid-chain diol based temperature and productivity proxies at test: a five-years sediment trap record from the Mauritanian upwelling

Long-chain mid-chain diol (shortly diol) based proxies obtain increasing interest to reconstruct past upper ocean temperature and productivity. Here we evaluate performance of the sea surface temperature proxies; long chain diol index (LDI), the diol saturation index (DSI) and the diol chain-length index (DCI), productivity/upwelling intensity proxies: the two diol indices DIR (Rampen et al., 2008) and DIW (Willmott et al., 2010) and the combined diol index (CDI), as well as the nutrient diol index (NDI) as proxy for phosphate and nitrate levels. This evaluation is based on comparison of the diols in 20 sediment trap samples from the upwelling region off NW Africa collected at 1.28 km water depth with daily satellite derived sea surface temperatures (SSTSAT), subsurface temperatures, productivity, the plankton composition from the trap location, monthly PO4 and NO3 concentrations, wind speed and wind direction from the nearby Nouadhibou airport. The diol based SST reconstructions are also compared the long chain alkenones based U37 proxy reconstructions (SSTUK). At the trap site, most diol proxies lag wind speed (phase φ = 30 days) and can be related to upwelling. Correlation with the 25 abundance of upwelling species and wind speed is best for the DCI, DSI and NDI whereas the DI and CDI perform comparatively poorly. The nutrient proxy NDI shows no significant correlation to monthly PO4 and NO3 concentrations in the upper waters and a negative correlation with wind-induced upwelling (r=0.28, φ = 32 days) as well as the abundance of upwelling species (r=0.38; Table 4). It is suggested that this proxy reflects upwelling intensity rather than upper ocean nutrient concentrations. 30 At the trap site, SSTSAT lags wind speed forced upwelling by about 4 months (φ = 129 d). The LDI based SST (SSTLDI) correlate poorly (r = 0.17) to SSTSAT which we attribute to variability in 1,13 diol abundance unrelated to SST such as https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c © Author(s) 2021. CC BY 4.0 License.

sea surface temperatures (SST SAT ), subsurface temperatures, productivity, the plankton composition from the trap location, monthly PO 4 3and NO 3 concentrations, wind speed and wind direction from the nearby Nouadhibou airport. The diol based SST reconstructions are also compared the long chain alkenones based U K' 37 proxy reconstructions (SST UK ). At the trap site, most diol proxies lag wind speed (phase φ = 30 days) and can be related to upwelling. Correlation with the 25 abundance of upwelling species and wind speed is best for the DCI, DSI and NDI whereas the DI and CDI perform comparatively poorly.
The nutrient proxy NDI shows no significant correlation to monthly PO 4 3and NO 3 concentrations in the upper waters and a negative correlation with wind-induced upwelling (r 2 =0.28, φ = 32 days) as well as the abundance of upwelling species (r 2 =0.38; Table 4). It is suggested that this proxy reflects upwelling intensity rather than upper ocean nutrient concentrations. 30 productivity. The SST UK correlates best with SST SAT (r 2 = 0.60). Also amplitude and absolute values agree very well and the flux corrected SST UK time series average equals the SST SAT annual average.

Introduction (as Heading 1) 35
Upper ocean temperature and productivity reconstructions are important for assessing past climate and environment. For organisms, optimal functioning of their membranes and transport in the cell is crucial and since temperature has a large influence on the solubility and viscosity of lipids, organisms adapt their lipid composition to temperature. Common responses to a decrease in temperature are (1) reducing average chain-length of the lipid molecules, (2) increasing the number of double bonds or (3) the degree of cyclisation (e.g. Suutari et al., 1994;Elling et al., 2015;Sollich et al., 2017). 40 This response of organisms to adapt their metabolism and metabolite composition to prevailing ocean conditions we can use to reconstruct past ocean temperature and productivity through searching for relevant metabolites in the fossil record and taking them as environmental proxies. This is not an unequivocal enterprise since the metabolite composition of organisms is often influenced by a combination of environmental variables and, therefore, proxies based on these metabolites bear the risk of reflecting this. Furthermore, the same metabolites may be produced by different organisms, each having its own complex 45 response to environment. To obtain robust and reliable temperature, nutrient and productivity reconstructions from fossil metabolites it is essential to test the performance of these proxies in present day conditions. In this study, we do so by investigating the long-chain mid-chain diol composition in sediment trap samples in relation to environment during a multiyear trap deployment. We performed this study in the upwelling area off Cape Blanc, one of the most productive regions in the world (Chavez and Messié, 2009). From this region detailed (daily to monthly) records of water temperature, 50 productivity, nutrient concentrations and upwelling dynamics in the upper ocean as well as atmospheric parameters like wind direction and wind intensity are available.
Other diol-based proxies have been defined but these either strongly correlate with the proxies mentioned above or are not relevant in the Mauritanian setting such as a diol proxy for terrestrial/fresh water input (Versteegh et al., 1997;Lattaud et al., 60 2017a,b). We also evaluate the diol temperature proxies in relation to the lipid based temperature proxies: the long-chain alkenone-based U K' 37 (Prahl and Wakeham 1987;Brassell et al., 1986a). fraction was extracted with hexane, and then fractionated into three polarity fractions using Bond-Elut silica gel cartridges.

Lipid data aquisition (as Heading 2)
For diol analyses the polar fractions were silylated with 100 µl N,O-bis (trimethylsilyl) trifluoroacetamide (BSTFA) for 1 hour at 70°C. Except CBeu5 (see below), the concentrations of the different diol isomers were determined using a time of flight mass spectrometer (TOF-MS) LECO Pegasus III (LECO Corp., St. Joseph, MI) interfaced to an Agilent 6890 gas chromatograph (GC). The GC was equipped with a 15m x 0.18mm i.d. Rtx-1MS (Restek Corp., USA) column (film 100 thickness: 0.10µm) with an integrated 5m guard column. A temperature-programmable cooled injection system (CIS4, Gerstel) in combination with an automated liner exchange facility (ALEX, Gerstel) and a multi-purpose autosampler (MPS 2, Gerstel) were used for sample injection. The CIS was operated in solvent vent mode (2.5 µL injection volume) and initially held at 40 °C for 0.05 min, then heated at 12 °C s -1 to 80 °C (held 0.1 min) with the split valve open, a vent flow of 100 mL min -1 and a vent time of 10 sec. After the solvent vent time, the split valve was closed, the CIS heated at 12 °C s -1 to 105 340 °C and held for 2 min for sample transfer to the GC-column. The GC oven was initially held at 60 °C for 1 min, then heated at 50 °C min -1 to 250 °C and at 30 °C min -1 to 310 °C (held 2.5 min), resulting in an analysis time of 9.3 min per sample. Helium was used as carrier gas in constant flow mode at a rate of 1.5 mL min -1 .
The ion source was operated at a temperature of 220 °C with the GC connected to the source by means of a heated transfer line set to 280 °C. The pressure in the flight tube of the MS was 2 x 10 -7 Torr. Full range mass spectra from m/z 50-600 110 obtained under EI conditions at 70 eV were recorded at a data rate of 40 spectra s -1 . Processing of the TOF-MS data was accomplished with the software (ChromaTOF™) provided by the manufacturer and included automated baseline and peak finding, spectral deconvolution of overlapping chromatographic peaks, peak area calculations, and extraction of compound specific m/z (mass-to-charge ratio) ion chromatograms.
Relative amounts of alkyl diol isomers were estimated from peak areas of specific ions resulting from α-cleavage of 115 trimethylsilyl-ethers at various mid-chain positions (de Leeuw et al., 1981) As an example, for C 28 diols peaks at m/z 299 (1,14-diol) and m/z 313 (1,13-diol), and for C 30 diols peaks at m/z 313 (1,15 diol) and m/z 327 (1,14-diols) were integrated (Rampen et al., 2008;Versteegh et al., 1997). For the diol quantification, selected reference samples containing alkyl diols in high relative amounts and without other interfering compounds (as indicated by GC/TOF-MS analyses) were analyzed by GC-FID on an Agilent 6890 GC equipped with a 60m x 0.32mm i.d. DB1-MS (Agilent J&W) column (film thickness: 120 0.25µm). Samples were injected via an on-column injector. The GC oven was initially held at 60 °C for 3 min, and then heated at 20 °C min -1 to 150 °C and at 6 °C min -1 to 320 °C (held 30 min). Helium was used as carrier gas in constant flow mode at a rate of 1 mL min -1 . Alkyl diols were identified by retention times, elution order and relative abundances in comparison to the corresponding GC/TOF-MS analysis. GC-FID areas of the respective diols were used for quantification by comparison to the peak area of an external n-C 28 1-alkanol standard. If the diol peaks consisted of a coeluting isomeric 125 https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c Author(s) 2021. CC BY 4.0 License. mixture, isomer proportions were quantified by their relative ratios as obtained by analysis of isomer-specific m/z ions from GC/TOF-MS. Quantified amounts of diols were than used to obtain mass-specific response factors, that allowed subsequent direct quantification of diols by GC/TOF-MS for the rest of the samples.
The diols of CBeu 5 were analyzed by GC-MS using an Agilent 6850 GC coupled to an Agilent 5975C MSD equipped with a fused silica capillary column (Restek Rxi-1ms; length 30 m; diameter 250 µm; film thickness 0.25 µm). The temperature 130 program for the oven was as follows: held at 60 °C for 3 min, increased to 150 °C at 20 °C min -1 , increased to 320 °C at 4 °C min -1 , held at 320 °C for 15 min. Helium was used as carrier gas and the flow was held constant at 1.2mL min -1 . The MS source was held at 230 °C and the quadrupole at 150 °C. The electron impact ionization energy of the source was 70 eV.
Relative amounts of alkyl diol isomers were estimated from peak areas of specific ions as described above. Absolute amounts of diols were obtained by comparison to the peak area of the known amount of the internal standard androstanol 135 (m/z 333).

Calculation of diol indices
Diol indices were calculated on the basis of relative abundances of the diol isomers.

Temperature proxies (as Heading 3)
Long Chain Diol Index (Rampen et al., 2012 According to Rampen et al. (2014a) this index has a high correlation to temperature in Proboscia cultures (Rampen et al., 2009) following the relation: Diol Chain length Index (Rampen et al., 2009(Rampen et al., , 2014a 160 The DSI and DCI showed high correlations to culture temperature but these correlations were absent in a global survey relating DSI and DCI derived from core-tops to SST (Rampen et al., 2014a).

Upwelling / upper ocean productivity proxies
Diol Index (Rampen et al., 2008) DI R = (1,14C 28 + 1,14C 30 ) / (1,14C 28 + 1,14C 30 + 1,15C 30 ) (10) 165 This has been suggested as a proxy for southwest monsoon upwelling in the Arabian Sea (Rampen et al., 2008) and is also applicable to the Namibian upwelling (see Versteegh et al., 2000) Diol Index (Willmott et al., 2010) As may be expected, the ratio of the slopes of these transfer functions is 1:16, the Redfield Ratio (Redfield 1963). 180 Generally, fluxes recorded in sediment traps have a logarithmic nature. As a result, linear correlations between diols are subject to bias by single high values. To overcome this bias, we based our correlations on the log-transformed diol concentrations (Sup. Fig. S1). Diol concentrations of CBeu1-4 are on average more than one order of magnitude (20.71 times) lower than for CBeu5 (Sup. Fig. S2). Although this does not affect the diol proxy ratio's, it does affect the direct 185 comparison of diol fluxes. Concentrations of CBeu1-4 have been calculated without standard but by using the peak areas, response factors, injection volume and sediment mass. In contrast, diol concentrations of CBeu5 have been calculated using peak areas, response factors and an internal standard and these fluxes are in good agreement with those reported from other sediment traps from high productivity regions (Rampen et al., 2008, DeBar et al., 2019.

Alkenone based U K' 37 SST proxy
Lipid SST proxy data on long chain alkenone based U K' 37 have been determined for trap CBeu5 (material collected between 195 28 March 2007 and17 March 2008). Data have been combined with those of Mollenhauer et al. (2015) for CBeu1-4.
The alkenones in fraction 2 were analysed using an Agilent 5890 gas chromatograph equipped with a DB5-MS capillary column and a flame ionization detector. Alkenone identification is by relative retention times and comparison with a laboratory-internal standard sediment. The U K' 37 index was calculated using the peak areas of the C 37:2 and C 37:3 alkenones (Prahl and Wakeham, 1987) Analytical precision based on repeated analyses of the standard sediment is ± 0.01 units of the U K' 37 index. The U K' 37 values were converted to temperatures using the calibration for suspended particulate matter (Conte et al., 2006) A systematic and statistically significant difference between the average SST UK of CBeu1-4 (Mollenhauer et al., 2015) and CBeu5 is observed (t-test for the same mean p = 0.0014; F-test for same variance p= 0.024; Monte Carlo permutation, mean 210 p= 0.0014, variance p = 0.0016 with 10 4 permutations; Epps-Singleton test for the same distribution of two univariate samples p = 1.5 10 -6 ). Correcting of this analytical discrepancy has been performed on the basis that alkenone composition reflects reliably the ambient temperature during production (Conte et al., 2006). Furthermore, we allowed for a small degree of selective degradation during transport through the water column, resulting in slightly higher SST UK (global calibration of U K' 37 from suspended particulate matter (SPM) to ambient SST (Conte et al., 2006). By subtracting 0.094 U K' 37 units, the 215 CBeu5 SST UK average equals that of CBeu1-4. Application of the global calibration reveals a deviation of 0.45°C from SST SAT for the entire series CBeu1-5 (SST SAT = 0.723 SST UK + 5.56; φ = 35 days; r 2 =0.60, average 21.76°C) and since the slope of the global calibration differs from our dataset, the reconstructed SST UK show an offset of almost +0.8°C from SST SAT at the lower temperatures (18°C) and an offset of -1.9°C at 26°C. Logically, the alternative, a local calibration removes this offset. For details about this correction and the local cubic calibration we refer to supplemental material (Sup. 220

Diatom data
Diatom counting has been performed as described in Romero et al., (2020). Data for Proboscia alata are presented here for the first time.

Insolation (as Heading 3)
Solar insolation at 20°N has been obtained from https://www.pveducation.org/pvcdrom/properties-of-sunlight/calculation-ofsolar-insolation. This record provided insolation for each 5 th day. The four most important frequency components were extracted using the sum of sinusoids model of PAST4 of the insolation explaining 99.994 % of its variance have been used to generate a 5-years insolation record with daily resolution for comparison with the daily SST SAT record (Fig. 2). 230 The insolation is not a simple sinus wave, the lowest half of the insolation amplitude (7.777-10.744 W m -2 d -1 ) includes 41.2% of the year (152 d), the highest half accounts for 58.8% (213 d). The 10% highest insolation (>10.447 W m -2 d -1 ) occurs for 28% (103 d) of the year, the 10% lowest insolation values (< 08.074 W m -2 d -1 ) occur at 17% (62d) of the year ( Fig. 2a).

Daily sea surface temperatures 235
Satellite-derived SST SAT values for the CBeu have been obtained from the ERDDAP daily optimum interpolation (OI), AVHRR dataset (Dataset ID: ncdcOisst21Agg_LonPM180) for 20.875N, 18.625W, the grid point closest to the location of CBeu and representing the square enclosed by 21°N, 18.75°W and 20.75°N, 18.5°W ( Fig. 1, green square). For each cup, the daily SST SAT values have been averaged (binned) for the respective deployment time (each bin being the interval between the starting and ending dates for each cup deployment). Since it takes time for material produced in the water column to 240 arrive in the cup, the SST SAT have also been calculated for bins shifted by one day (φ = 1) to 140 days (φ = 140) back in time. In this study we assume that φ is constant over time. Since the purpose of the SST proxies is to reconstruct SST, we took the φ that gave the best correlation between the binned SST SAT record and the proxy-based reconstructed SST as collected by the sediment trap.

Monthly subsurface water temperatures, salinity and nutrient data 245
Subsurface temperatures and salinities were obtained from the World Ocean Atlas 2018 Zweng et al., 2019) using the statistical mean temperature on a 1° square for all decades centred at 18.5° W, 20.5° N. (Fig. 3

Wind direction and wind speed 250
Wind direction and strength are based on 3 hourly observations from Nouadhibou airport (20° 56′ N, 17° 2′ W; Institut Mauretanien de Recherches Océanographiques et des Pêches, Nouadhibou, Mauritania). Since these data are vector-based (direction and speed) they can't be simply averaged (The average of 1 h 10 m/s at 350° and 1h 10 m/s at 10° is 1h 9.85 m/s at 0° and not 1 h 10 m/s 180°). Therefore, the data have been decomposed into their North-South (V n ) and East-West components (V e ), averaged and converted back to directions and speeds (see Grange 2014 and 255 http://www.webmet.com/met_monitoring/622.html\). Since there is a high variability between the days, 11 days moving averages have been used to reduce this high frequency contribution (Fig. 2). Furthermore, the deviation from north is presented (-90° is west, + 90° east) rather than the full 360° scale).

Dust events
The dust data (

Statistics
Statistical analyses have been performed with the software package PAST4.0.4 (Hammer 2001) and with R packages 'grDevices', 'stats', 'EnvStats', 'methods' and 'car'. Phase shifts reported between proxy records and SST SAT represent the 265 phase for which the correlation is highest. Phases are represented by φ, wavelengths by λ.

Daily sea surface temperatures
Daily SST SAT are above average (21.3°C) from July to December (5 months) and low and relatively constant from December 270 to July (7 months Fig. 2b). The maximum temperature is reached at Sept 30 th (day 273, just over 6 months after the minimum).
The records of binned SST SAT averages (representing the averaged SST for the deployment period of each cup) varies depending on the assumed phase shift (φ) between the genesis of a signal at the sea surface and its arrival in the cup of the sediment trap. Minima lay between 17.9-18.4 °C, maxima between 26.2-27.2 °C and annual amplitudes between 7.8-8.9°C. 275 https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c Author(s) 2021. CC BY 4.0 License.

Monthly subsurface water temperatures, salinity and nutrients
The temperature-salinity diagram for the upper 600 m (Fig. 3) shows for November-December a predominantly South Atlantic Central Water (SACW) signature with relatively low salinities for a given temperature (or high temperatures for a given salinity). From January, the contribution of North Atlantic Central Waters (NACW) increases. From March to June the upper 80 m show admixture of cool, low salinity waters (but absent in April) so that despite increasing insolation the SST 280 stay below 19°C for this period.

Wind direction and wind speed
From 2003 to 2008 the 11 days averaged wind blows 87% of the time from NW to NE (36°W -21°E) and 52% of the time from NNW to NNE (17°W-7°E). Most easterly directions occur in a relatively short period from November to February, 290 when wind speed is at its lowest on average (Fig. 2d). This phase with more westerly winds was poorly developed in 2005/06. This is followed by a period of high wind speeds from March to July with winds more or less constant from the North (or NE in 2006 and2007;Fig. 2e). A third period, from June to November, has the same overall wind as the preceding period but intercalates more eastward outbreaks and wind speed tends to be lower. Over the investigated period there is an eastward trend (D = -0.0124 d -7.49, whereby D is the direction in degrees from N and d = Julian day since 01-01-2003; 295 r 2 =0.14; p=9.45 10 -9 n = 1989). This implies that at the beginning of 2003 the average wind direction is 7.5° W and by the end of the study period (after 2000 d) this is 17.4°E. This effect is predominantly due to the years 2007 and 2008 for which from the beginning of May to the end of October the wind has a more easterly direction than during this period in the preceding years (Fig. 2). Frequency analysis shows apart from this long-term trend an annual cycle (363 d) and a 6 months cycle (184.6 d) with its most eastern direction at December 10 th . 300 The 11 days averaged wind speed varies between 1 and 9 m s -1 and shows no trend (slope 1.4 10 -4 ). Frequency analysis shows an annual cycle (371 d) and a 6 months cycle (181 d). The phase of the cycle is 45 d (from Jan. 1 st ) with the maximum wind speed ¼ π later, (45 + 93 = 138 d) which is May 18 th .

Dust events
The dust record shows a strong seasonal cycle (φ = 55.33 days, λ = 363.2 d) and the abundance of dust events lags wind 305 speed by 10 days.

Sediment trap mass fluxes and upwelling species abundance
The relative abundance of upwelling species (%Upw) show a clear seasonal cycle and lag wind speed by 21 days (r 2 =0.29, p=3.5 10 -9 ) and conversely dust event frequency by 11 days. The %Upw. species correlates well with the SST SAT (negative r 2 = 0.53 φ = 76 p<2.5 10 -20 ; positive r 2 =0.40, φ = -119 p < 2 10 14 ). Total mass flux maxima occur in concentrated intervals 310 of which the timing and amplitude may differ between the years. Total mass fluxes, and fluxes of carbonate, C org , biogenic large maximum is missing (Fig. 2). As a result of this irregular timing of total mass flux maxima, the most important frequency component is not the annual cycle but a cycle with a length of 257 d (r 2 =0.17, p = 4.04 10 -6 ).

Diol ratios and indices
The productivity related NDI, DI W , DI R , DSI and CDI all largely depend on the same diol isomers and except for some DSI relations covary to a considerable extent (r 2 > 0.46; Table 4, Fig. 4). Due to their annual cyclicity, their statistical relations to environment encompass usually two maxima of correlation of opposite sign and about 181 days apart (½ cycle). For some indices the causal relation to environment isn't clear a priori. In case of a negative correlation with environment the 335 correlation is positive if shifted by ½ cycle but this is also true for the unshifted reciprocal of the proxy. Below we often https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c Author(s) 2021. CC BY 4.0 License. provide both and negative correlations and their phases with environmental variables without further interpretation which will be added in the discussion.
The DSI correlates strongly with the NDI and DCI (Table 4). The positive correlation with SST SAT (r 2 = 0.25, φ = 77 p=5.7 10 -5 ) is much lower than the negative correlation (r 2 = 0.42, φ = -124 p=3.4 10 -8 ). For a considerable number of 365 samples no unsaturated diols could be detected, these samples were omitted from the calculations.

Upwelling and upper ocean productivity proxies DIR, DIW and CDI
The DI R , as a proxy for the Proboscia contribution shows correlates significantly with the ln P. alata flux (r 2 = 0.29, p = 1.6 10 -4 ) but not with the total mass flux or its logarithm (Fig. 7).
The DI W as a proxy for upwelling, correlates significantly with the ln P. alata flux (r 2 = 0.12; p = 0.024) but not with the total 370 mass flux (r 2 = 0.011, p = 0.25). The correlation of the DI w with the LDI is negative low (r 2 =0.04) and barely significant (p=0.09), for the DI r and LDI the correlation is also negative but much better (r 2 =0.24 p=6.3 10 -7 ). Both the DI R and DI W show a negative correlation with wind strength which is best at φ = 79 (Table 4).
The Combined Diol Index (Rampen et al., 2014a) CDI, is almost identical to the DI R (r 2 =0.995) and therefore has not been evaluated separately. 375

Upper ocean nitrate and phosphate concentration proxy NDI
The NDI, shares about half its variance with the CDI, DI W and DI R , even higher with the DCI (74%) and DSI (67%). The NDI shows no correlation with P. alata flux (r 2 =0.005, p=0.6) nor with the total mass flux (r 2 =0.0014, p=0.7) but correlates well (r 2 = 0.38 p=4.1 10 -12 ) with the relative abundance of upwelling indicating species (%Upw) in the samples.

Alkenone based U K' 37 and SST UK 380
The SST UK values are 18.0 -26.1°C and best fit to SST SAT if SST UK at φ = 35 days (Fig. 5). For this phase lag the binned SST SAT are 18.5-26.1°C and the SST UK record thus has about the same amplitude extending 0.5°C below the minimum and equalling the maximum SST SAT . The measured U K' 37 explains 59% of the SST SAT variance and the SST UK 60%. The integrated production temperature (flux corrected) = 21.14°C, nearly the SST SAT for φ = 35 (21.37°C).
The regressions of U K' 37 and SST UK to binned SST SAT do not differ much in the explained variance and phase (φ 35 d, r 2 385 =0.59 vs 0.60). The flux corrected SST UK for CBeu1-5 prior to correction is 21.54, after correction it 21.14°C which is only 0.23°C lower than the SST SAT for the same period (21.37 φ=35) and well within the standard error of the method (1.2°C) (Conte et al., 2006).

Water temperatures during the sampling period 390
During the entire record, SST SAT tend to remain relatively constant for periods of one to three weeks with rapid shifts of 1-2°C between them (Fig. 2). During summer these shifts tend to be larger and they may be up to 4.5 °C within a week at the transitions between summer and winter (in 2004). The length of these 1-3 weeks periods of stable temperatures is similar to the duration of cup deployment (8-24 days) so that these periods are partly reflected by, and consistent features of the binned https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c Author(s) 2021. CC BY 4.0 License.
SST SAT record, irrespective of the phase difference between the SST SAT binned averages and the corresponding cup 395 deployment times. This is much more apparent after November 2006 when the cup deployment is systematically less than 10 days and close to the duration of the shorter-term events of the system. This demonstrates that to capture the systems dynamics, these shorter deployment times are the preferred mode of operation, if not a requirement.
The SST SAT does not simply follow the insolation curve, but a substantial rise in temperature only starts from the beginning of June, only a few weeks before the insolation maximum. We attribute the delay to maximum intensity of the trade winds 400 during late winter and spring, intensifying upwelling and pumping cool, low salinity and nutrient-rich SACW waters to the surface along the coast and subsequently spread these westward over the trap site. Wind speeds are strongly reduced and upwelling is at a minimum for July to October, enabling the surface waters to strongly increase in temperature. The temperature gradient is always small for the upper 20 m, the permanently mixed layer.
The most prominent feature of the monthly water temperature profiles is the strong annual cycle at the sea surface which 405 becomes smaller with increasing depth. In the upper 100 m the relatively short distances between the temperature profiles for different depths (the strength of the temperature gradient) are clearly smaller from January to June compared to the rest of the year. We attribute this to mixing and upwelling. This seems to be most intense in May and June. In May this is indicated by elevated temperatures up to 400 m depth and in June the upwelling and mixing result in SST that are even lower than the preceding months, despite stronger solar insolation. We also observe that the highest temperatures at depth (80-200m) lead 410 the SST by about three months (Fig. 3) which in case of temperature proxies generated at these depths could lead to proxyderived temperature records leading SST.

The annual cycle, wind, dust SST and upwelling
All environmental parameters investigated show a dominant annual cycle modulated by a semi-annual component. This is also true for most of the proxy records. This implies that a significant correlation of a proxy parameter to an environmental 415 variable, with a given phase shift has a high chance to also provide significant correlations with other environmental parameters, albeit with different phase relations. This is no problem for proxy records that are well investigated and relatively well understood, such as the U K' 37 for which we know a causal relation with temperature. However, for several of the diol-based records only statistical correlations are known and causal mechanistic understanding of the relations to environment are largely absent. Due to the cyclic nature of data, a proxy record has two correlation optima per cycle. One 420 optimum is a positive correlation, the next optimum is an anti-correlation, located ½ cycle distance from the first optimum, if the cycle is perfectly sinusoid. If we don't know the proxy-environment relation, this second, antic-correlation can also be seen as a positive correlation with the inverse (B/A in stead of A/B) of the proxy and since the relation is statistical rather than causal, there is no a priori argument against considering the reciprocal of the proxy as the better one relating to environmental change. Additional arguments, e.g., from the data structure and/or functioning of the system are needed to 425 decide if the proxy or its reciprocal is the better one. In our system, off Cap Blanc, the modulation of the semi-annual https://doi.org/10.5194/bg-2021-309 Preprint. Discussion started: 18 November 2021 c Author(s) 2021. CC BY 4.0 License. frequency component causes asymmetry in the cycles of the environmental variables (Fig. 2). As a result, phases of best correlation and anti-correlation are not ½ cycle apart. Comparison of wind speeds and SST SAT in this region shows that both have a short period of high values, followed by a relatively long period of low values. If proxy records follow environmental change closely, they also should display this asymmetry in the annual cycle. High wind speeds in late winter and spring drive 430 upwelling, preventing SST SAT to increase with solar insolation so that only when wind ceases by the beginning of June, SST SAT can rapidly rise. This results in maximum wind speed leading maximum SST by 122 days (based on 11 day means, r 2 =0.38) and minimum SST leading maximum wind speed by 91 days (r 2 =0.37). The asymmetry in the annual cycle thus leads to an alternation of 122+91=213 day difference between correlation optima, followed by a 365-213=152 day difference between the next correlation optima. We expect to see this asymmetry also in the proxy records. The frequency of dust 435 events lags wind speed by 10 days so that most events occur during the period of most active upwelling. Due to this small phase difference in relation to the generally two to three times longer sample frequency of the sediment traps it is impossible to separate the effects of upwelling and dust input on the export flux, species and lipid composition. Since wind speed is driving both upwelling and dust event frequency, only relations to wind speed are further considered.

Diol based temperature proxies LDI, DSI and DCI 440
The LDI uses the underlying assumption that the percentages of 1,13 diols (relative to the 1,15C 30 diol) decrease with temperature. For the SST LDI outliers (values below 18°C) we couldn't find any analytical explanation. These outliers arise from excess production of both 1,13 diols. The occurrence of these anomalies suggests that this excess production is unrelated to temperature. We observe that all three events with excess 1,13 diols (and very low LDI values) occur during Total Flux maxima (compare Figs. 5,9) and as such we speculatively suggest that factors leading to elevated export 445 productivity play a role. However, also with the outliers omitted, we observe a poor relationship between LDI and SST SAT and a phase lag of 41 days (r 2 =0.17) but still a significant seasonal cycle (λ=347 d, r 2 =0.36, p=1.2 10 -10 ). This contrasts with trap studies from the equatorial Atlantic, Mozambique Channel and to a lesser extent the Cariaco Basin where a general absence of a seasonal cycle in the LDI has been observed (de Bar et al., 2019). The SST LDI integrated production temperature (23.6°C) is well above the annual SST SAT (21.3°C). This may be explained by the general observation that 1,13 and 1,15 diol 450 production is weighted towards the non-upwelling season, which has above average temperatures (as is also indicated by the poor correlation of 1,15 and 1,13 diols with 1,14 diols Fig. 7).
The LDI shows strong covariance with DSI and DCI. Since these proxies are based on completely different diols (LDI 1,15 and 1,13 diols, DSI and DCI 1.14 diols) this covariance has thus no mathematical basis but rather should be sought in common sources or responses to environmental change. applicable to this region. The DSI, lags SST SAT (r 2 = 0.25, φ = 76, n=58) which is a much longer time span than observed for the LDI (φ = 41). This difference may indicate that sinking speed of the 1,14 diols (comprising the DSI) differs from that of 460 the 1,13 and 1,15 diols (comprising the LDI). It may also be that changes in the 1,14 diol composition represent an environmental response different from the response changing the 1,13 and 1,15 diols composition. One explanation may be that the 1,14 diols are produced by different organisms than the 1,13 and 1,15 diols, which allows for both different sinking speeds and different environmental responses. In the marine environment the 1,13 and 1,15 diols are considered to be predominantly derived from marine eustigmatophytes (e.g. Gelin et al., 1997;Volkman et al., 1992Volkman et al., , 1999Versteegh et al., 465 2000;Rampen et al., 2014b). The 1,14C 32 diol is a major diol in the diatom Apedinella radians (Rampen et al., 2011) but since we do not observe this diol in our cup samples, we infer that this species doesn't significantly contribute diols in our case. In our samples we do observe a relative good correlation between fluxes of the diatom Proboscia alata with the 1,14 diols (31% explained variance, Fig. 8). This correlation is considerably better than the correlation between P. alata fluxes and 1,13+1,15 diols (4%) or between total diatoms and 1,14 diols (17%). This suggests that Proboscia significantly 470 contributes 1,14 diols whereas its 1,13+1,15 diol contribution is insignificant. Culture experiments demonstrate this ability for P. alata to produce 1,14 diols (Sinninghe Damsté et al., 2003). Our observations also agree well with diol data from sediment traps from the Arabian Sea where abundance of 1,14 diols covaries with upwelling and appears unrelated to the abundance of 1,15 diols (Rampen et al., 2007). Also a recent sediment trap study from the East Sea (Gal et al., 2021) supports the hypothesis of 1,14 diols being produced by a different plankton population than the 1,13 and 1,15 diols. In Assuming the upwelling related Proboscia diatoms being a major 1,14 diol source in our research area, it makes sense to investigate if and how the DSI relates to wind induced upwelling and %Upw. During the investigated time interval, SST SAT 480 lags wind speed by about one season (φ =122 r 2 =0.342). With diatom sinking rates in this region of 100 -250 m d -1 (Fischer and Karakaş 2009) and 1280 m trap depth, particles are expected to arrive at the trap within 5-12 days after starting to sink.
We observe that the abundance of upwelling species (%Upw) lags wind speed by 21 days (r 2 =0.26, p=6.8 10 -9 ) agreeing reasonably well with the estimated sinking speed (Romero et al., 2020). We therefore expect the DSI to behave similar to the %Upw lagging the wind speed by 21 days and leading SST SAT by 101 days. We observe the DSI lags the SST SAT by 76 days 485 and is not leading it as was assumed based on the arguments above. However, the inverse DSI (higher values more diol unsaturation) provide a completely different picture lagging wind speed with the same phase (φ = 21, r 2 =0.24 p=9.7 10 -5 ) and correlates well with %Upw (r 2 =0.39). It also correlates surprisingly well with SST SAT leading it by 124 days (φ=-124, r 2 =0. 42). We therefore, propose that the inverse DSI reflects upwelling related changes with a higher contribution of unsaturated diols upon stronger upwelling. Interestingly, also in the East Sea sediment trap (Gal et al., 2021)  export production (Kim et al., 2017). Gal et al. (2021) ascribe the spring 1,14C 28:1 diol maximum to P. alata whereas they relate the autumn maximum including also the 1,14C 30 diols to P. indica, a species absent from Cap Blanc. The DSI thus seems to be heavily modulated by processes related to high productivity obscuring a possible relation to water temperature, 495 if present.
The DCI leads the SST SAT (φ=-77, r 2 =0.29). Since the reconstructed SST can't precede the actual SST, it is unlikely that the DCI reflects SST. Moreover, available transfer functions (see Methods 3.4) lead to unrealistic temperatures for CBeu. The inverse of the DCI lags SST (best correlation r 2 =0.35 at φ=117) but simultaneously its average chain length decreases with temperature, which is opposite to the common physiological response of organisms to temperature (higher 500 lipid melting point/viscosity at higher temperatures). Consequently, a causal relation through physiological adaptation of the diol-producing organisms to SST must be considered unlikely.
The DCI follows wind speed by 27-31 days (r 2 =0.28, φ=27, p= 7.5 10 -9 ; φ=31, p=6.4 10 -9 ) and upwelling species (%Upw) by 6-9 days (r 2 =0.42). This suggests that like the DSI, also the DCI reflects changes in the upwelling regime. Interestingly, both these supposedly upwelling related DSI and the DCI correlate better to SST SAT than the temperature proxy LDI which we 505 attribute to the (annual) cyclic nature of the system rather than to a causal relation with temperature.

NDI
The NDI, DSI and DCI correlate well (r 2 >0.68). This is expected since although the NDI consists of all diols encountered, it combines the DSI and inverse DCI. The 1,14 diols in these two latter proxies obviously dominate the NDI. The NDI has 510 been proposed to reflect NO 3 and PO 4 3concentrations in the surface waters (Gal et al., 2018(Gal et al., , 2019(Gal et al., , 2021. If in our setting, we assume that upwelling and the relative abundance of upwelling species (%Upw) are associated with nutrient-rich conditions at the CBeu site, a significant positive correlation between NDI and %Upw is expected. However, our results show a negative correlation (Table 4). The interpretation of the NDI as reflecting NO 3 and PO 4 3concentrations in the surface waters is further complicated by a clear annual cycle in our dataset whereas monthly PO 4 3and NO 3 concentrations 515 show no cyclicity. Furthermore, mostly concentrations of these nutrients show no correlation with the NDI (p <0.04).
A closer look at the NDI reveals that, the slopes of the respective transfer functions relate to each other according to the Redfield Ratio N:P = 16:1. This may be expected since the original calibration between NDI and nutrients (Gal et al., 2018) is primarily based on sediment-derived NDI values and photic zone annual nutrient levels (n=216), the latter following the Redfield ratio. SPM summer samples with associated summer nutrient concentrations are also included in this Gal et al. 520 (2018)

DI W , DI R and CDI
The DI w , DI R and CDI have the same 1,14 diols in the numerator and only differ in the combination of 1,13 and 1,15 diols in the denominator. The CDI only adds the 1,15C 30 diol to the DI R . The DI w has been directed as a proxy for the contribution of Proboscia diols relative to other diols. In our dataset, its correlation with the Proboscia fluxes is low but still significant 530 (r 2 =0.12 p=0.024). The proxies for upwelling strength DI R and CDI correlate even better with Proboscia (Table 4).
The DI R has been suggested to be proxy reflecting total export productivity. We observe high values (>0.9) of this proxy which would indicate a permanent high production in the region. Both the DI W and DI R correlate negatively with wind speed (at φ = 79) and SST SAT (at φ = -69). Closer observation shows that generally, all diol fluxes more or less covary except in 2007 where pulses in 1,13 and 1,15 diols precede those of the 1,14 diols ( Fig. 9) and the DI R sinks below 0.8. Comparison to 535 the total mass flux (correlation insignificant) and to the %Upw reveals that this change in diol composition is not accompanied by a consistent change in export flux so that an explanation for the relative increase in 1,15 diols seems to require a more subtile knowledge of the relation between diol abundance and environment on the CBeu trap data alone. ). It appears that for the 1,13+1,15 diols our average of 3.9 µg m -2 d -1 for CBeu5 is only 1.5 times the average for the nearest trap M1U (2.6 µg m -2 d -1 ), and about 3 times the flux of the oligotrophic central Atlantic M2U (1.2 µg m -2 d -1 ) but only half that of the westernmost M4U (7 µg m -2 d -1 ). For the 1,14 diols the average of 1.7 µg m -2 d -1 for CBeu5 is least three times higher than for the other traps (0.5 for M1U, 0.01 for M2U and 0.3 for M4U). Since CBeu5 is the only site under permanent 545 upwelling influence, we infer that the 1,14 diols are particularly abundant in the (coastal) upwelling whereas the 1,13 and 1,15 diols also increase with increasing productivity but seem to be less bound to upwelling.

U K' 37 and SST UK
The 35 days phase lag of SST UK relative to SST SAT suggests that it takes about one month between fixation of the alkenone composition in the cell and the collection of the alkenones in the cup of the sediment trap which implies an average sinking 550 rate of 38 m d -1 .
This differs strongly between the years. From winter to summer 2007 both SST records nicely overlap, suggesting a much higher sinking rate, also in the following autumn and winter 2003/2004 and 2005/2006  This we may expect since our data cover only a fraction (18-27°C) of the global temperature range and for this fraction the cubic transformation behaves almost linear. Since the explained variance is much lower (60%) than for the global calibration (97%) we infer that other factors than temperature influence the alkenone composition. If the reconstructed SST UK would perfectly project to the SST SAT the slope of the regression would be 1 and the intercept 0. However, the slope of the SST UK regression to binned SST SAT is 0.72 and the intercept 5.6, which implies that the global SPM calibration overestimates local 560 SST SAT below 20°C and underestimates local values above this temperature. If we assume that the global calibration is correct and the alkenones reflect the water temperature the organisms were subject to, the conclusion would be that above 20°C the organisms live in water that is slightly cooler than that at the sea surface (SST SAT ) and slightly warmer below this temperature. This could be explained by a higher influence of solar irradiation at the sea surface than at the subsurface. This is only feasible if the alkenone-producing organisms live partly or entirely below the surface mixed layer. Alternatively, we 565 may adjust the SPM calibration of Conte et al. (2006) 570 with φ = 33 and r 2 =0.54. Just like the global calibration of Conte et al. (2006) this regional calibration performs much better than any of the diol proxies discussed above.

Conclusions
The variation in alkenone, and diol, fluxes and relative abundances as observed in sediment trap cups off Mauritania from 2003 to 2008 have been compared with environmental conditions and plankton composition for the same region and time 575 period. From this comparison, it appears that: 1. Peak total mass fluxes of material to the sediment trap do not show a statistically significant annual cycle but may occur throughout the year. Nevertheless, total mass flux maxima are most abundant during spring. We explain this rather unpredictable occurrence of these flux maxima by attributing them to result from the passage of upwelling filaments over the sediment trap which occurs most often during spring, but is not limited to this. 580 2. Off Cap Blanc, upwelling variability is the major environmental variable. It shows a strong annual cycle in response to the strength of the trade winds. Sea surface temperature also shows a strong annual cycle remaining low in winter and during vernal upwelling and following insolation when upwelling is reduced during summer. It lags upwelling by 130 days. As a result of the predominant annual cycle in both temperature and upwelling, correlations between parameters and/or proxy records should be interpreted with care and phase relations should be considered to identify the most likely forcing 585 mechanism. 5. The diol-derived nutrient index NDI, the DCI as well as the percentage of upwelling species correlate better with SST than the LDI. However, they lead the SST by several months and their variability is most likely a response to upwelling associated processes rather than temperature. A rather intriguing result is the anti-correlation between the diol-derived nutrient proxy NDI and upwelling intensity.

Code availability 595
Not applicable

Author contribution
GJMV interpreted the data and wrote most of the manuscript. KAFZ discussed and edited the paper prior to submission. JH 600 processed performed the lipid analyses of CBeu5 and contributed to the Material and Methods section. OER contributed the diatom data and together with GM critically reviewed earlier versions of the manuscript. GF coordinated the sediment trap project.

Competing interests
The authors declare that they have no conflict of interest. 605

Disclaimer
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.