Hot primary melts and mantle source for the Paraná-Etendeka flood basalt province : New constraints from Alin-olivine thermometry

Abstract Continental flood basalts (CFB) are amongst the most voluminous volcanic eruptions in Earth's history. They are rapidly emplaced, and in rare cases the thick lava piles are associated with primitive magmas that have high MgO contents. The compositions of these primitive melts are consistent with a deep-sourced, high-temperature mantle plume origin. Whilst the association of CFBs with impacting mantle plumes is widely accepted, the magnitude of the thermal anomaly is not yet resolved. The development of Al-in-olivine thermometry, however, allows the crystallisation temperature of (near-)liquidus olivine to be determined without knowing the composition of the co-existing melt. This provides both a robust minimum estimate of mantle temperature and a value from which potential temperature (TP) can be back-calculated. This technique has previously confirmed that crystallisation temperatures in CFB settings can be a few hundred degrees greater than those estimated for MORB, and the results hint at a diversity in crystallisation temperatures between different CFB settings. In this study, we re-assess the TP of the mantle source of the Parana-Etendeka CFB province by applying the Al-in-olivine thermometer to olivine–spinel pairs from picrites and ferropicrites. We show that the mean crystallisation temperature of olivine with Fo>90 in the picrites is 1458 °C, with a maximum temperature of 1511 °C. Using the mean value, we calculate a preferred TP of 1623 °C, for an assumed lithospheric thickness of 50 km and magma emplacement pressure of 0.5 GPa. This represents a thermal anomaly of around +300 °C relative to ambient mantle, and confirms that the mantle source of the Parana-Etendeka CFB is the second hottest known from Phanerozoic Large Igneous Provinces, after the Caribbean Large Igneous Province. The ferropicrites record a cooler mean olivine crystallisation temperature of 1296 °C. Given that these low-volume melts derive from deeper and earlier melting of mantle pyroxenite, their temperature is not directly comparable to that of the picrites but they appear to require a somewhat cooler mantle source – perhaps found at the front or edges of a rising plume head.


Petrological estimates of mantle temperature
The enormous volumes of magma that formed continental flood basalt (CFB) provinces, and their rapid eruption rates, represent significantly higher magmatic fluxes than the eruptions occurring on Earth's surface today (e.g. Renne et al, 2015). These vast outpourings of CFB lavas are widely considered to be responsible for global environmental change and extinction events throughout the geological record (e.g. Saunders, 2016;Sobolev et al., 2011;Song et al., 2013;Wignall, 2001). The theory that the starting heads of upwelling thermal plumes in the mantle are responsible for CFB provinces is now widely accepted (e.g. Herzberg et al., 2007;Morgan, 1971;Putirka et al., 2007;Richards et al., 1989;White & McKenzie 1989, 1995. The temperatures of the starting plume heads are less well constrained. This uncertainty is important because thermal convection is a key mechanism by which planets lose their internal heat: understanding the occurrence and properties of upwelling plumes, including their thermal history, has implications for interpreting the spatiotemporal patterns of volcanism both on our planet and others. Temperature variations in the convecting mantle are conveniently described by the concept of potential temperature, TP, which is its temperature extrapolated along a solid adiabat to 1 bar (McKenzie and Bickle, 1988). If mid-ocean ridges remote from hotspots sample the ambient mantle (Parsons and Sclater, 1977), then a thermal anomaly in the mantle has an elevated TP relative to that of the MORB-source mantle. The TP of the mantle can be investigated through its melts, with the most primitive magmas retaining properties that are easiest to relate to the mantle source. Techniques involving FeO-MgO olivine-melt equilibrium have been used to obtain crystallisation temperatures of the most primitive melts, which can be considered minimum estimates of TP (Beattie, 1993;Roeder and Emslie, 1970). However, these techniques require a well-constrained estimate of the equilibrium melt composition, and will be inaccurate if olivine related to the modern-day Tristan mantle plume through two complementary seamount ridges across the Atlantic seafloor (e.g. White and McKenzie, 1989).
As with most CFB provinces, the main-sequence of flood lavas are not particularly useful for understanding mantle melting conditions, as they have been extensively altered by fractionation and assimilation processes (e.g. Arndt et al., 1993;Erlank et al., 1984;Peate, 1997).
Fortunately, outcrops of more primitive rock types -picrites and ferropicrites -are exposed in the Namibian portion of the Paraná-Etendeka CFB province: these are found predominantly as dykes that are part of the plumbing system of the overlying flood lavas and associated volcanics (Erlank et al., 1984;Marsh et al., 2001;Thompson et al., 2001). These magmas bypassed the fractionation and assimilation processes responsible for the chemical processing of the flood basalts, and retain properties imparted by mantle melting. The ferropicrites represent the earliest magmatic products in this province, as well as in many others Gibson, 2002). The more common picrites are thought to derive from high-fraction melting in the central stem of the Tristan plume, and thus are genetically related to the flood basalts, providing a window into the physical and chemical properties of their mantle source (Jennings et al., 2017;Thompson et al., 2001;Thompson and Gibson, 2000).
Previous estimates of the TP of the Paraná-Etendeka plume source derived from the picrites are contradictory, but indicate a mantle source temperature well above that of ambient mantle (Thompson and Gibson, 2001;Keiding et al., 2011). A significant limitation of the previouslyavailable techniques was that they required the melt composition to be known, and so different methods of calculating the primary melt composition have resulted in different TP estimates. Thompson and Gibson (2000) used the maximum olivine forsterite (Fo) content and whole-rock MgO-Fe/Mg relationships to infer a komatiitic primary melt composition, implying an extremely high corresponding TP of ~1680 °C. By contrast, Keiding et al. (2011) used melt inclusion compositions from high-Fo olivine to define a primary melt that is much lower in MgO and corresponds to a lower TP of ~1520 °C. These two approaches and results are further discussed in section 4.2.1. In this study, we use the Al-in-olivine thermometer, which is independent of melt composition, to investigate crystallisation temperatures of primitive melts and calculate the TP of the mantle source of the Paraná-Etendeka CFB province, in order to understand the magnitude of the mantle thermal anomaly that was responsible for the creation of this voluminous CFB province.

Samples
The samples used in this study are Early Cretaceous (132-133 Ma) picrites and ferropicrites from the Etendeka region of NW Namibia, part of the Paraná-Etendeka CFB province. They are the same set as those used by Jennings et al. (2017) and were previously described by Thompson et al. (2001), Thompson and Gibson (2000) and Gibson et al. (2000); detailed descriptions can be found in those studies. Sample localities are shown in Supplementary Figure S1.
The picrites are from dykes near Horingbaai (known in the literature as "Horingbaai-type" of the high-Ti suite; Marsh et al., 2001). They contain 0.5-2 mm olivine macrocrysts with cores of up to Fo93, and appear to represent near-primary melts rather than having a cumulate origin (Thompson et al., 2001). The Horingbaai picrites have previously been suggested to derive from high-temperature, high-fraction melting of peridotite in the sub-lithospheric mantle, followed by were analysed by QEMSCAN at the University of Cambridge, which performs phase identification on a pixel-by-pixel basis from EDS measurements of composition (see Supplementary Figure S2 for images; see Neave et al., 2017, for technical details The Paraná-Etendeka ferropicrites are olivine-phyric (up to Fo86) and may also contain macrocrysts of sub-calcic augite set in a fine-grained groundmass that is rich in opaque oxides.
The ferropicrites have olivines with low Fo but similarly high Ni contents to those in the picrites.
On the basis of their Fe-rich and Al-poor major element composition, anomalously low HREE abundances, and other features of their elemental and isotopic compositions the ferropicrites have been interpreted as resulting from the melting of mantle pyroxenite at high pressures and temperatures (Gibson, 2002;Jennings et al., 2016;Tuff et al., 2005).

Al-in-olivine thermometry
The Al-in-olivine thermometer exploits the fact that aluminium becomes less incompatible in olivine with increasing temperature (e.g. Wan et al., 2008;De Hoog et al., 2010). This thermometer uses the partition coefficient of Al between spinel and olivine (kd = Al2O3 olivine / Al2O3 spinel , in wt. %) and has been experimentally calibrated for a range of temperatures and a limited range of mafic-ultramafic compositions by Wan et al. (2008) and Coogan et al. (2014).
(K) = 10000 0.575(0.162) + 0.884(0.043)Cr# − 0.897(0.025) ln d [1] where Cr# = Cr spinel ( Cr spinel + Al spinel ), in atomic %, and the values in parentheses are the published standard error of the parameter When olivine and spinel simultaneously saturate and grow in the same melt, we assume they are in equilibrium with one-another. Since spinel and olivine tend to crystallise at, or close to, the liquidus in primitive melts the crystallisation temperature that they record approximates the liquidus temperature, which itself represents the minimum temperature of mantle melting (because melts could only have cooled since they were generated). This thermometry technique is considered robust to post-crystallisation diffusive resetting because the diffusion rate of Al in olivine is extremely slow at low silica activity (Spandler and O'Neill, 2010;Zhukova et al., 2017), and is ideally used when spinel crystals are trapped and isolated in a closed system within the olivine structure.

Analytical
In order to use the Al-in-olivine thermometer as calibrated by Wan et al. (2008) Table S1; counting times of other elements and for spinel measurements given in Supplementary Table S2). These uncertainties are propagated through temperature calculations.
The analytical points in olivine were measured approximately 20-30 μm from the inclusion edge. Given that Al is a major element in spinel and a trace element in olivine, there is a risk that the apparent Al2O3 concentration of the olivine (Al2O3 olivine ) could be increased by secondary fluoresced X-rays from within the spinel inclusion (e.g. Borisova et al., 2018). We checked this with Monte Carlo simulations using PENEPMA (v. 2014;Llovet and Salvat, 2017), converting k-ratios to concentrations using CalcZAF (derived from CITZAF; Armstrong, 1995). In the simulation, a pure MgAl2O4 spinel half-sphere was set within pure Fo90 olivine. At various distances from the interface, the sample was bombarded with a simulated 15 keV electron beam for several hours to reach an arbitrary low uncertainty. An annular detector spanning a 35-45° take-off angle was used to represent the 40° take-off angle of a microprobe. We found that the fluoresced apparent Al2O3 signal from MgAl2O4 spinel dropped below 0.005 wt.% (became negligible) at around 4 μm from the edge of a 20 μm inclusion (Figure 2), although we caution that a much higher level of secondary fluorescence of Al is expected in the case of an olivine inclusion within a larger spinel crystal.
Spinel Fe 3+ /ΣFe and Fe 3+ per formula unit (p.f.u.) were initially calculated stoichiometrically from the EPMA data according to the method of Droop (1987) on the basis of four oxygens.
Nine spinel standards, that had previously been characterised by Mössbauer and EPMA (Davis et al., 2017;Ionov and Wood, 1992;Wood and Virgo, 1989) and XANES, were analysed in the same analytical session: their Fe 3+ /ΣFe was, on average, overestimated by a factor of 1.7, so the Fe 3+ /ΣFe of the unknown spinels was multiplied by a correction factor of 0.6. The final Fe 3+ values should therefore be considered approximate values only. However, they indicate that the spinel samples fall within the Fe 3+ /ΣFe calibration range of the Al-in-olivine thermometer (Coogan et al., 2014;Wan et al., 2008).
In the study of Coogan et al. (2014), the thermometer reproduced the experimental temperatures of 42 of the 45 calibration experiments to within 20 °C, indicating an inherent 1σ uncertainty of the thermometer of 11 °C (note that this is lower than the uncertainty derived by propagating the errors of fit of the individual coefficients, which relates to correlation between those coefficients' uncertainties). In all cases, the analytical uncertainty in Al2O3 olivine is greater than that of the thermometer, so we propagate this analytical uncertainty to define temperature uncertainties.

Initial data filtering
The Al-in-olivine technique is appropriate for fairly primitive Cr-spinel and olivine pairs, and is calibrated using a limited range of spinel compositions. Spinel has solid solutions with a range of end-members: a high proportion of the spinel (MgAl2O4) and magnesiochromite (MgCr2O4) end-members are indicative of crystallisation from a more primitive melt, whereas increasing concentrations of Ti and Fe 3+ in the B site, as well increasing Fe/Mg in the tetrahedral A site, are expected with melt evolution. Although most inclusions in the Paraná-Etendeka olivines were found to be rich in spinel/chromite-series components, several were rich in Ti and Fe 3+ , and a few inclusions in low-forsterite olivine were in fact ilmenite or magnetite. The data were therefore initially filtered, such that inclusion measurements where Al+Cr < 1.4 p.f.u. were discarded. The complete compositional data (excluding these low Al+Cr analyses) is provided in Supplementary Table S1.

Crystal chemistry and thermometer applicability
Measured olivine crystals with appropriate spinel inclusions had compositional ranges of  De Hoog et al., 2010). The thermometer indicates that olivines crystallised at higher temperatures will contain more Al2O3, so the generally positive correlation between Fo and Al2O3 olivine shown in Figure 3 is expected. For the subset of olivines in which P2O5 was measured, the average concentration was below the detection limit, so corrections are not required for P-Al substitution (cf. Coogan et al., 2014). The activity of SiO2 in the picrite and ferropicrite melts was likely low and similar to the calibration experiments, so little added uncertainty is expected from an additional silica + vacancy substitution mechanism (Coogan et al., 2014). In addition, Al2O3 olivine may be affected by high TiO2 activity because the incorporation of Ti into olivine octahedral sites could be charge-coupled with substitution of Si for Al in olivine tetrahedral sites. This would lead to anomalously high temperature estimates for olivine with high Ti concentrations, although alternative solution mechanisms for Ti in olivine also exist (Hermann et al., 2005). However, this is not considered to be a problem here as TiO2 olivine concentrations were close to, or below, detection limits of 0.01 to 0.03 wt.% (depending on setup).
Aspects of the spinel inclusion compositions are shown in Figure 4a  and Ti 4+ in spinel need to be considered, because the presence of magnetite and ulvöspinel components of spinel will affect the activity coefficients of aluminium and chromium endmembers. The range of spinel Fe 3+ (following the initial filtering described in section 3.1) is well within the limit of the thermometer calibration range, so does not pose a problem. However, the measured spinels reach higher Ti contents than the calibration set. This is particularly true of the ferropicrite spinels, because these magmas are relatively high in Ti (as shown by whole-rock measurements and melt inclusions; Gibson et al., 2000;Jennings et al., 2017). The extent to which high Ti contents will affect the thermometry results is uncertain, but we err on the side of caution and restrict our dataset to spinel-olivine pairs where the maximum spinel Ti content is not far from the maximum in the calibration dataset. An upper limit of 0.05 Ti p.f.u. was chosen for spinels in the picrite samples and corresponds to ~2 wt.% TiO2. A higher limit of 0.08 p.f.u.
(~3 wt. % TiO2) is applied to the spinels in the ferropicrites. The corresponding temperatures are shown in Figure 4c and d, where the few high-temperature outliers in the picrite data are seen for spinels with high Ti contents. Data points above our stated limits were removed from the dataset for the subsequent results and discussion.

Crystallisation temperature
Histograms and Probability Distribution Functions (PDF) of calculated crystallisation temperatures are shown in Figure 5. The PDF curves incorporate the uncertainties in temperature estimates, which tend to be larger at lower temperatures because of the lower Al2O3 olivine concentrations, resulting in a low-temperature tail. The mean equilibration temperature of all picrite spinel-olivine pairs is 1372±13 °C (1σ = 84°C; Figure 5a). However, the highest Fo olivines must have crystallised first and would record a temperature closer to the liquidus. The mean temperature for spinel-olivine pairs where olivine is Fo>90 is therefore higher, at 1458 ± 11 °C (1σ = 36°C; Figure 5b). We consider this small group of ten measurements to approximate the liquidus temperature. The highest temperature estimate was obtained from an olivine (Fo92.9) in sample 97SB33, with an equilibration temperature of 1511 °C (1σ = 24 °C). This is in stark contrast to the range of equilibration temperatures obtained from olivine-inclusion MgO partitioning by Keiding et al. (2011Keiding et al. ( , 2013 of 1279 -1361 °C from samples of similar Horingbaai picrites with equally high Fo olivine. The lower temperatures obtained by Keiding et al. (2011Keiding et al. ( , 2013 may reflect the higher mobility of Fe-Mg compared to Al during heating and rehomogenisation of the melt inclusions (see section 4.2.1).
The crystallisation temperatures of the olivines in the ferropicrite are generally lower than those of the picrites, which is somewhat expected given their lower Fo content (Roeder and Emslie, 1970). The mean equilibration temperature of ferropicrite spinel-olivine pairs is 1296 ± 7 °C (1σ = 29 °C; Figure 5c). Given the narrow spread in the data, we use this value to approximate the liquidus temperature of the ferropicrites. Primary ferropicrite liquids are Fe-rich , so the lower Fo content of ferropicrite olivine does not imply significant fractionation.
We examine the temperature variations with olivine Fo content in Figure (supplementary table S3). This trend, also shown in Figure 6, indicates that temperature should rapidly decrease at the onset of crystallisation, with a moderate decrease in Fo. After plagioclase and clinopyroxene saturate, temperature decreases more gently with Fo.
The high-temperature olivines in picrite sample 97SB41 crystallised at higher temperatures than can be related to the Fo>90 cluster by simple fractional crystallisation. One possibility is that the spread in temperatures at a given Fo content reflects Fe-Mg re-equilibration: whereas Al diffuses slowly in olivine, Fe-Mg exchange is relatively fast (e.g. Chakraborty, 1997 (Jennings et al., 2017).
In comparison to the temperatures recorded by olivine-spinel pairs from different samples of Paraná-Etendeka picrites, those for the ferropicrite are all within uncertainty of one-another.
The highest temperature is 1346 °C (1σ = 54 °C), which surprisingly comes from a rather low Fo  Table 1). This provides strong evidence for a pronounced thermal anomaly in the mantle source of the Paraná-Etendeka CFB province. More generally, Figure 7b shows PDFs of Al-in-olivine temperatures for samples from different tectonic setting. The bandwidth is related to typical analytical uncertainties. It is clear that there is a large offset in the highest probability (the mode) temperature of approximately 100 °C between MORB, Iceland and CFBs. Interestingly, the CFB distribution appears to be bimodal; this may be due to sampling bias but it is apparent that samples from CFB provinces have a probability tail that extends to much higher temperatures than MORB and Iceland. The maximum olivine temperatures recorded are the closest to the liquidus of the primary melt, so we can consider the high temperature limits of these distributions (or the 95 th percentile temperature to reduce bias from low sampling rates; Table 1). The 95 th percentile temperatures for different CFB provinces range from 1379 to 1568 °C, which represents an offset from MORB of 125-314 °C, implying a much hotter mantle source.
4.2 Potential temperature in the proto-Tristan mantle plume from the Etendeka picrites The potential temperature of the mantle must be higher than the crystallisation temperature In order to calculate the TP of the mantle source of the picrites in the Paraná-Etendeka from the crystallisation temperature, we must reconstruct the thermal pathway from melt generation through to crustal emplacement. The method used here is similar, but not identical, to that used by Matthews et al. (2016), who determined a TP of 1480 °C for the mantle source of recent Icelandic melts by calculating the thermal pathway of the melts from source to emplacement. We calculate this with the assumption that the mantle source of the Etendeka picrites is a KLB1-like peridotite; this is supported by major element, trace element and isotope geochemistry (Jennings et al., 2016;Thompson et al., 2001).
Assuming that the spinel-olivine equilibrium temperature is at or below the liquidus, we can use the liquidus temperature (taken to be the average of the Fo>90 olivine temperatures, 1458 °C), to constrain a minimum crustal emplacement temperature, and so a minimum mantle TP for the Paraná-Etendeka picrites. We then produce a forward model of decompression, melting, and the cessation of melting at the base of the lithosphere, and allow the TP to vary until the predicted emplacement temperature matches this Al-in-olivine temperature.
The thermal model and parameters used here are the peridotite model of Jennings et al. (2016), which is based on the equations and method of Katz et al. (2003) but uses an updated parameterisation of KLB-1 peridotite melting based on Powell (2011) andHolland (2015). This assumes that upwelling peridotite follows an adiabat (that defines TP) and then intersects the solidus. Following this, the enthalpy of fusion further reduces the temperature during batch decompression melting: the melt and residue are assumed to maintain thermal equilibrium. Whilst this assumption represents a simplification, the timescales of thermal equilibration are much shorter than those of mass transfer required for chemical equilibrium, and the temperature and chemical properties of fractional melts can become decoupled, as discussed by Matthews et al. (2016). The appropriateness of the equilibrium assumption was tested by Jennings et al. (2016), who found limited divergence between incremental batch melting and equilibrium melting thermal pathways.
Melting ceases once upwelling material reaches the base of the lithosphere: at this point the liquid is assumed to segregate from the solid and follows a basaltic liquid adiabat to the pressure of the magma chamber in which it is emplaced. An example of a calculated thermal path is shown in Figure 8.
The results for the best fit TP required to reproduce a given liquidus temperature (in this case 1458 °C) are dependent upon lithospheric thickness (because the lithospheric thickness dictates the shallow limit to melting and thus melt fraction F) and crystallisation pressure (Pcryst). The picrites were generated during a period of rifting and lithospheric thinning associated with the opening of the Atlantic Ocean. The lithospheric thickness at the time of picrite generation has been estimated from REE inversion techniques to be around 50 km depth (Thompson et al., 2001;Thompson and Gibson, 2000). To understand how different lithospheric thicknesses might affect the calculated TP required to match the picrite crystallisation temperatures, we performed the calculations over a complete range of lithospheric thickness, corresponding to the pressure just above the peridotite solidus (125 km or 112 km to match the melt temperature emplaced at 10 or 5 kbar, respectively) through to the emplacement pressure (30 or 15 km; Figure 9). We find that the calculated TP is fairly insensitive to lithospheric thickness (and therefore also F), varying by <100 °C over the entire lithospheric thickness range.
The effect of the assumed crystallisation pressure on the calculated TP is moderate, where TP calculated at Pcryst = 1.0 GPa tends to be around 50 °C cooler than that calculated at Pcryst = 0.5 GPa. 0.5 GPa is consistent with the co-saturation of olivine and plagioclase feldspar in these samples (Thompson et al., 2001) and with an average pressure of 5.3 kbar from clinopyroxene thermobarometry of five Horingbaai picrite samples (Keiding et al., 2013;barometer uncertainty ± 1.3 kbar, Putirka, 2008). The TP of the ferropicrite mantle source is more difficult to calculate due to the uncertainty of the inferred pyroxenite composition, mineralogy, thermodynamic properties and melt productivity (Gibson, 2002;Jennings et al., 2016;Tuff et al., 2005). In addition, an unknown water or other volatile content may somewhat depress the pyroxenite solidus.
Silica-undersaturated mantle pyroxenite is thought to form by either solid-state hybridisation of subducted crust (i.e. eclogite) with peridotite, or through the reaction of eclogite-derived partial melts with peridotite (e.g. Yaxley and Green, 1998;Herzberg et al., 2011); previous studies on pyroxenite melting have tended to consider starting materials with a chemistry that reflect these processes. Jennings et al. (2016) used thermodynamic modelling to investigate the composition of the partial melts of the anhydrous 50:50 peridotite-basalt hybrid composition (KG1-like, after of Kogiso et al., 1998), and found that low-fraction melts produced at elevated TP were broadly similar to the Etendeka ferropicrites. Given that we have little additional information available about the exact composition and origin of pyroxenite in the Etendeka mantle source, we have chosen to use that same composition to represent pyroxenite melting in the present study. We simplify the modelling by assuming a single hybrid lithology, and disregard the effect of heat transfer between heterogeneous lithologies (cf. Phipps Morgan, 2001).
The uncertainty in pyroxenite composition and melting behaviour add to that introduced by the unknown effect of the high Ti content of the ferropicritic spinels on the thermometer accuracy. For these reasons, a ferropicrite mantle source estimation should be considered an approximate value only.
We use the properties and melting characteristics parameterised by Jennings et al. (2016) of a KG1-like composition to approximately model the thermal pathway of pyroxenite partial melts, employing the same method as described for peridotite melting. For the mean crystallisation temperature of 1296 °C, we find TP solutions of 1360-1500 °C for the permissible ranges of lithospheric thickness at our preferred crystallisation pressure of 1.0 GPa .
This increases to 1360-1550 °C if crystallisation occurred at 0.5 GPa. While these values are only approximate, and are limited by the stated caveats, they do appear to indicate that the ferropicrites were derived from mantle that was somewhat cooler than the picrite source, but warmer than ambient mantle. One plausible explanation is that the pyroxenite source of the ferropicrites was entrained within the upwelling head of the proto-Tristan plume. Alternatively, pyroxenite may have been present in the cooler peripheries of the plume head, but because pyroxenite is expected to be denser than peridotite in the mantle, this is more difficult to reconcile with the requirement for buoyancy.

Comparison with previous TP estimates of the proto-Tristan plume
The MgO content of any primary mantle melt should be directly related to mantle TP and the extent of melting (F), and simple parameterisations of this relationship have been proposed (e.g. Herzberg et al., 2007). It is therefore unsurprising that debate has focused on the method of deducing the primary melt MgO. Previous estimates of magmatic temperatures and TP of primitive magmas from the Paraná-Etendeka CFB province have been based on the Fo content (so the Fe/Mg ratio) of olivine, which preserves a more reliable record of the primary melt than whole-rock composition. Thompson and Gibson (2000) examined some of the same samples as our study: their highest temperatures are also obtained from sample 97SB33, which contains the most forsteritic olivines in the collection (up to Fo93.3). In order to calculate TP, Thompson and Gibson (2000) first The melt FeO and MgO can rapidly undergo diffusive re-equilibration with the host olivine during heating in a furnace, which will: (i) alter their concentrations in a way that cannot be easily corrected for; and (ii) result in MgO concentrations that tend towards equilibrium according to the furnace temperature (Jennings et al., 2017). The MgO contents measured by Keiding et al. (2011) correspond to Fo-MgO equilibrium temperatures of 1246 to 1395 °C (rehomogenised at 1350 °C).
The TP range suggested by our Al-in-olivine thermometry study of 1600-1640 °C is very high even by the standards of all other Phanerozoic LIPs (except for CLIP; Trela et al., 2017) and lies between the estimates of Thompson and Gibson (2000) and Keiding et al. (2011). We stress that the different thermometry methodologies used by Thompson and Gibson (2000), Keiding et al. (2011), and in our present investigation, all point to significantly elevated TP in the Parana-Etendeka CFB mantle source: they only disagree on the exact size of that temperature offset from ambient. Our high temperature and associated high F are supported by generally depleted incompatible trace element compositions, both in whole-rock samples and olivine-hosted melt inclusions (Jennings et al., 2017;Keiding et al., 2011;Thompson et al., 2001). This high TP of the Tristan plume starting head contrasts with the lower estimated TP of around 1360 °C for the modern-day Tristan hotspot (Weit et al., 2017). These estimates were also made using olivinemelt Mg exchange thermometry, using both bulk rock and olivine-hosted melt inclusion compositions. If the lower apparent TP is not related to the different thermometry material used, it indicates a significant cooling of the plume through time.We suggest that using the Al-inolivine thermometer to estimate of crystallisation temperatures is more robust than Fe-Mg exchange-based methods, although the weakness still resides in the assumptions required to convert crystallisation temperature to TP, making the crystallisation temperature more certain than its source temperature. This estimate of TP is somewhat lower than that derived from modelling of Al-in-olivine thermometry results, which may in part be explained by different uncertainties associated with the two methods, including differences in the underlying thermodynamic models. However, we emphasize that two completely independent methods of estimating TP (olivine-spinel Alexchange vs. applying empirical parameterisations to melt major element compositions) both point towards the same conclusion: that the Paraná-Etendeka CFB mantle source was significantly hotter, by hundreds of degrees, than ambient mantle.

Conclusions
Al-in-olivine thermometry provides an alternative method to traditional Mg-Fe-based techniques to assess crystallisation temperatures in primitive melts, with the advantages that: (i) no assumptions need to be made about the equilibrium melt composition; and (ii) the thermometry is more robust to diffusive resetting in slow cooling scenarios (Coogan et al., 2014;Wan et al., 2008). The EPMA technique is ideal for making the required spinel and olivine measurements because of both its high spatial resolution and ability to measure trace element concentrations down to 10s -100s ppm for some elements. We have demonstrated that secondary fluorescence from Al in small spinel inclusions (< 20 μm) will have a negligible effect on typical Al concentrations measured in olivine at 15 kV, provided that measurements are made at a minimum distance of least ~4 μm from the interface. However, contamination of the olivine measurement by fluorescence will increase for larger grains or higher accelerating voltages, increasing the corresponding minimum distance from the spinel-olivine interface required for an accurate measurement.
Crystallisation temperatures based on the Al-in-olivine thermometer for picrites from the Paraná-Etendeka CFB province reach up to 1511 °C (1σ = 24 °C; Fo92.9) with a mean of 1458±11 °C for 11 measurements obtained for olivine with Fo>90. By contrast, the precursor ferropicrites have a maximum olivine-spinel co-crystallisation temperature of 1346 °C (1σ = 53 °C) and an average temperature of 1296±7 °C. The lower temperature estimates for the most primitive ferropicrites indicate that their pyroxenitic mantle source was somewhat cooler than the picrite source, and could be explained by the entrainment of pyroxenite by the upwelling head of the proto-Tristan plume or the presence of pyroxenite in the cooler peripheries of the plume.
Given that olivine and spinel would co-saturate at, or close to, the liquidus, the highest crystallisation temperatures are interpreted to represent the minimum temperature of the liquidus, which in turn represents a minimum temperature of the melt during emplacement in a crustal magma reservoir. We assign a minimum liquidus temperature for the picrites of 1458 °C, which is around 100-150 °C hotter than the TP of ambient mantle. We converted the crystallisation temperature of magmas from the Paraná-Etendeka CFB to a mantle TP by forward-modelling the PT path of decompression melting by assuming a peridotite source composition and using the equilibrium melting method of Katz et al. (2003) and the melting parameterisation of Jennings et al. (2016). The TP required to match the liquidus temperature slightly depends on the assumed thickness of the lithosphere as well as the depth of emplacement: for the Paraná-Etendeka picrites the best estimate of these parameters based on previous studies (around 50 km lithospheric thickness and 0.5 GPa emplacement pressure) results in a TP estimate of 1623 °C, corresponding to a high melt fraction of 0.33. This represents a significant thermal anomaly relative to ambient mantle of around +300 °C, and is extremely hot, even by CFB standards: the only higher crystallisation temperatures and interpreted mantle TP in Phanerozoic primitive magmas have been found in the CLIP province (Trela et al., 2017).  Thompson and Gibson, 2000), and higher than methods based on melt inclusions (TP = 1520 °C; Keiding et al., 2011). Such high crystallisation temperatures and mantle potential temperature estimates, together with 3 He/ 4 He ratios up to 10 times RA (Stroncik et al., 2017) in the Paraná-Etendeka picrites are consistent with conceptual models that invoke a mantle plume starting head in their generation, as well as in the generation of other comparable CFB provinces.
Measurements of a more recently-erupted sample from the Tristan mantle plume indicate a cooler TP of around 1440 °C (Herzberg and Asimow, 2008), which is broadly consistent with

Acknowledgements
We would like to thank Iris Buisman for technical support during the EPMA measurements, as well as for producing the QEMSCAN images. We are grateful to Godfrey Fitton, Claude Herzberg and Lawrence Coogan for their careful and insightful reviews that led to improvements of this manuscript, and to Catherine Chauvel for editorial handling. We also thank Simon      Closer than 2 μm, beam convolution leads to a much higher (primary) signal that is not shown.
The detection limit range of this study is shaded in orange, indicating that for a 10 -20 μm diameter inclusion, fluorescence is below detection when olivine is measured at least 4 μm from the interface; greater distances will be needed for larger inclusions.      indicates the melt path: initially at depth, a solid adiabat is followed, but on reaching the solidus, the enthalpy of fusion decreases the temperature more rapidly with pressure. A kink is seen where clinopyroxene is exhausted (cpx-out) and melt productivity decreases. Melting ceases at the LAB, and a liquid adiabat is followed from here to the crystallisation pressure. The calculation is run as a forward model to find a TP to match the Fo>90 olivine temperature.