Estimates of the Temperature and Melting Conditions of the Carpathian‐Pannonian Upper Mantle From Volcanism and Seismology

What drives the formation of basaltic melts beneath intraplate volcanoes not associated with extensive thermal anomalies or lithospheric extension? Detailed constraints on the melting conditions and source region are imperative to resolve this question. Here we model the geochemistry of alkali basalts and mantle nodules brought up by young (12–0.1 Ma) intraplate volcanoes distributed across the Carpathian‐Pannonian region and combine the results with geophysical observations. Rare earth element inversion and forward calculation of elemental concentrations show that the basalts require the mantle to have undergone less than 1% melting in the garnet‐spinel transition zone, at depths of about 63–72 km. The calculated melt distributions correspond to a mantle potential temperature of ∼1257°C, equivalent to a real temperature of 1290°C at 65 km beneath the Pannonian Basin. The composition, modal mineralogy, and clinopyroxene geochemistry of some of the entrained mantle nodules closely resemble the basalt source, though the latter equilibrated at greater depths. The gravity anomalies and topography of the Basin reveal no large‐scale features that can account for the post‐extensional volcanism. Instead, the lithospheric thickness and geotherm show that melting occurs because the base of the lithosphere, at ∼50‐km depth, is close to or at the solidus temperature over a large part of the Basin. Hence, only a small amount of upwelling is required to produce minor volumes (up to a few cubic kilometers) of melt. We conclude that the Pannonian volcanism originates from upwelling in the asthenosphere just below thinned lithosphere, which is likely to be driven by thermal buoyancy.


Introduction
The processes that govern the formation of minor volumes of basalts away from active plate boundaries are subject to widespread debate. Although such rocks exhibit geochemical signatures similar to ocean-island basalts (OIBs) and continental flood basalts-both considered the products of upwelling mantle plumes (e.g., Morgan, 1971;Richards et al., 1989;White & McKenzie, 1995)-it is increasingly evident that many instances of small-volume intraplate volcanism may not be driven by deeply sourced excess heat as originally hypothesized (e.g., Foulger & Natland, 2003). Numerous conceptual mechanisms have been proposed to explain such small-volume occurrences. These often involve some form of plate tectonic control, wherein lithospheric extension allows melts to rise to the surface (e.g., Foulger & Natland, 2003), or sublithospheric processes, such as edge-driven convection (King & Anderson, 1995, 1998King & Ritsema, 2000), shear-driven upwelling (Conrad et al., 2010), small-scale convection (Ballmer et al., 2007), or hydrous upwellings (Long et al., 2019;Yang & Faccenda, 2020). The ubiquity and viability of many such driving forces, however, remain contentious.
To determine whether intraplate magma generation is driven by the plate above or by processes below requires an integrated set of observations. Particularly valuable insights are contained in basaltic lavas, whose geochemistry and volume provide a record of the chemical, thermal, and dynamical state of their mantle source regions (e.g., Brown & Lesher, 2016;Ghiorso et al., 2002;McKenzie & O'Nions, 1991;Plank & Forsyth, 2016). Because their compositions reflect the composition and mineralogy of the source, the depth and degree of melting, and the temperature distribution in the mantle, they can be used to probe the physical processes in, and properties of, the melt source. These geochemical observables, used as key input for models of mantle melting, coupled with geophysical signals observable at the Earth's surface, such as gravity anomalies, surface wave velocities, and topography, allow holistic constraints to be placed on the origin of intraplate volcanism (e.g., Ball et al., 2019;Klöcking et al., 2018;McNab et al., 2018).
A region whose intraplate volcanism is not well understood is that of the Pannonian Basin and Carpathians in eastern central Europe. Dispersed throughout this area, small volumes of alkali basalts were erupted over the last 12 million years (Embey-Isztin, Seghedi et al., 2004). The surprising feature of this continental volcanism is that it largely postdates the extension in the Pannonian Basin, which had ceased by the Late Miocene (Horváth et al., 2015). Consequently, a variety of causes has been proposed to explain the generation of basaltic melts. These include decompression melting by asthenospheric upwelling, "finger-like" asthenospheric plumes, lithospheric edge-related mantle flow, plate tectonic processes (Embey-Isztin, Downes, et al., , 2001Harangi et al., 2015;Seghedi et al., 2004Seghedi et al., , 2011Wilson & Downes, 1991), and mantle water content and compression (Kovács et al., 2020). To understand the nature of this post-extensional intraplate volcanism, as well as similar instances worldwide, it is clearly desirable to obtain robust constraints on the generation of melt and its source region.
Here we present the results of melting models based on a judiciously screened geochemical data set of the most primitive erupted products of young intraplate magmatism in the Carpathian-Pannonian region. First, rare earth element (REE) inversion and forward element modeling are employed to determine the nature of the melt source and estimate the melting depth and melt fractions involved. Using the calculated melt distributions and a parameterization of mantle melting, we quantify the mantle potential temperature (T P ) beneath the Pannonian Basin. The calculated composition and mineralogy of the source region is then compared to that of the mantle nodules brought up by the basalts, in order to assess whether these may represent fragments of basalt source rock. Lastly, we examine the geophysical features of the Basin and place integrated constraints on the origin of intraplate volcanism in the region.

Carpathian-Pannonian Geodynamics
The Pannonian Basin is a major extensional basin located in eastern central Europe and constitutes part of the Alpine-Mediterranean orogenic system (Horváth et al., 2006;Royden, 1988). It is surrounded by the arcuate Carpathian orogen to the north, east, and south and the Eastern Alps and Dinarides to the west (Figures 1 and 2). The Basin is a result of E-W-directed extension during the Miocene, which started at ca. 20 Ma and continued until 8-9 Ma, with peak rates around 11-15 Ma (Balázs et al., 2016(Balázs et al., , 2017Horváth et al., 2006). This extension was largely coeval with, and of similar magnitude to, the shortening recorded in the outer Carpathians (Horváth et al., 2015). A 2-5 km thick sequence of Miocene sediments was deposited in the Basin during this period, predominantly onto the deformed Cretaceous-Paleogene rock units of the AlCaPa and Tisza-Dacia microplates (Balázs et al., 2017;Horváth et al., 2006;Schmid et al., 2020). The regional stretching was accompanied by severe attenuation of the lithosphere, and as a result, the Basin is presently underlain by a thin continental lithosphere (∼50 km; Figure 1b), of which the upper 21-35 km is crust (Horváth et al., 2006;Priestley et al., 2018).
(inset Figure 1a), perhaps as a response to the counterclockwise rotation and NNE-directed indentation of the Adriatic microplate (Balázs et al., 2016;Dombrádi et al., 2010;Horváth et al., 2015). Active seismicity presently occurs beneath Vrancea in the southern East Carpathians, where it is associated with a near-vertical slab still attached to the surface (e.g., Martin & Wenzel, 2006). The other lithospheric slabs that were previously subducted along the Carpathian arc have already been detached: their remnants are imaged at depths of 410-660 km by seismic tomography (Wortel & Spakman, 2000).
Perşani, and at the local eruptive centers of Burgenland, Little Hungarian Plain, Banat ("Lucareţ"), and Central Slovakia ("Štiavnica") (e.g., Embey-Isztin, Harangi et al., 2015;Seghedi et al., 2004). Boreholes in the center of the Basin, near Kecel (Figure 1a), have further revealed various alkali basalts of Late Miocene age (8-11 Ma) buried beneath sediments (Seghedi et al., 2004, and references therein), but no detailed petrological data are yet available from these basalts. Indeed, seismic sections, boreholes, and gravity and magnetic anomaly data concerned with the interior of the Basin (e.g., Zelenka et al., 2004) suggest that similar rocks have been buried by sedimentation in those parts that experienced rapid subsidence. Hence, the present-day surface distribution of the basalts is probably biased toward the margins of the Basin and toward relatively young occurrences.
sourced from the asthenosphere and that some were modified by interaction with the lithospheric mantle during ascent (Embey-Isztin, Harangi et al., 2015;Seghedi et al., 2004). Trace element concentrations in melt inclusions in olivines from Perşani show averages that coincide with whole-rock data (Bracco Gartner et al., 2020), demonstrating that (near-)primary melt characteristics were largely preserved in the bulk lavas.
Intraplate volcanism was accompanied by sporadic eruptions of low-volume, potassium-rich products during the Middle Miocene (13-16 Ma) and Quaternary (0.9-2.2 Ma) (Harangi, Wilson, et al., 1995;Seghedi et al., 2004Seghedi et al., , 2011. These are shown as orange triangles in Figure 1a. Those of Middle Miocene age occur close to where the alkali basalts of Styria and Balaton would later erupt. The Quaternary occurrences are found more dispersed throughout the Basin, near Bár, Gătaia, Uroi, and in Southern Harghita. These potassic melts seem to record a signature of lithospheric domains previously metasomatized by subduction-related agents (e.g., Harangi, Wilson, et al., 1995;Seghedi et al., 2011) and are likely to be of shallower origin than the alkali basalts.

Sample Selection
An important prerequisite for the present study is the selection of rock samples that most closely reflect the composition of their respective primary magmas. Hence, judicious screening is required to mitigate the effects of processes such as fractional crystallization and lithospheric/crustal contamination, which confuse the relationship between erupted products and their primary melts. To identify and account for these effects, detailed major and trace element analyses are required. Hence, the synthesized data set includes only those samples for which both are analyzed and are sufficiently precise.
The assembled data set comprises major and trace element and, where available, radiogenic isotope analyses for the most primitive alkali basalts from all exposed volcanic districts (supporting information; Figure 3). The least-evolved and least-fractionated rocks were selected by adopting cut-off values of SiO 2 ≤ 50 wt% ( Figure 3a) and MgO ≥ 8 wt% (Figure 3b), which ensure that olivine was the dominant fractionating  Nions, 1991Nions, , 1995. Average composition of ocean-island basalt (OIB; grey dashed line) is from Sun and McDonough (1989). Data sources as in Figure 3. phase. A cautious exception was made for Nógrád (MgO ≥ 7.3 wt%, whose trace element concentrations closely resemble the less-fractionated samples), in order to yield a greater data set. Lack of correlation between light-over-middle REEs such as La/Sm, which are known to increase during extensive clinopyroxene crystallization, and indicators of clinopyroxene crystallization (e.g., Sc/Y and Ca/Al), demonstrate that this process did not strongly affect the screened samples. Trace element ratios were employed to identify OIB-like samples and discard those which show evidence of secondary modification or interaction with lithospheric and/or metasomatized domains. The selected samples plot on the mantle array in Nb/Y-Th/Y space (Pearce & Peate, 1995) (Figure 3c) and within the field of global OIBs in Rb/Nb-Ba/Nb space (Willbold & Stracke, 2006) (Figure 3d).
The basalts are grouped into four suites corresponding to the locations in Figure 1: NW (Styria, Burgenland, Little Hungarian Plain, and Balaton), N (Central Slovakia and Nógrád), S (Banat), and SE (Perşani). Their primitive mantle-normalized incompatible element patterns and calculated averages (±1 standard deviation [SD]) are shown in Figure 4.

Melting Model and Application
Relative abundances of incompatible elements allow quantitative constraints to be placed on the conditions of primary melt generation (McKenzie & O'Nions, 1991. Owing to their inter-elemental differences in compatibility, REEs are particularly sensitive to the cumulative amount of melting and the relative proportion of melting in the garnet and spinel stability fields. Assuming that the melt source is homogeneous and the resulting basalt is a homogeneous mixture of the fractional melts, REEs can be used to determine the depth of the melting interval and the melt fraction involved. The best-fitting solution can be sought iteratively by employing computational inversion, such as performed by the INVMEL (inversion melting) algorithm. This scheme inverts observed REE concentrations for the best-fitting melt distribution as a monotonic function of depth for a specified source composition (see McKenzie & O'Nions, 1991, for the original version and theoretical basis). For a given inversion, the melting depth interval, the top of the melting column, and initial distribution of melt fraction with depth are varied iteratively, and fractional melts are summed over incremental depths. The minimum root mean square misfit (M) between the observed (c o i ) and calculated concentration ratios (c c i ) of the REEs (elements i) is sought using Powell's conjugate direction method. M is defined as where N is the number of REEs and σ i is the SD. Once the optimal solution is obtained, the forward model calculates concentrations of other elements by integrating instantaneous fractional melts over incremental depths. Comparison of the calculated and observed concentrations of elements other than the REEs provides an independent test of the melting model obtained from the inversion.
INVMEL exploits a consistent set of mineral-melt REE partition coefficients determined by lattice strain parameterization (Blundy & Wood, 1994;Brice, 1975;Wood & Blundy, 1997). Details of these calculations for the REEs, as well as the partition coefficients for all elements used in the modeling, are provided in Appendix A and the supporting information, respectively. The effects of temperature and phase composition on these values are small relative to the other uncertainties inherent to the modeling (McKenzie & Blundy, 1999).
The composition of the melt source region was gauged by the observed Nd isotopic ratios (ϵ Nd values) of the screened samples (following White & McKenzie, 1995). The average value of ϵ Nd for all the basalts discussed here is 3.3. The composition of the source model used for modeling was therefore calculated by combining the primitive source of O'Nions (1991, 1995), with ϵ Nd ¼ 0, and their MORB source, with ϵ Nd ¼ 10, in appropriate fractions to generate a source with ϵ Nd ¼ 3:3. The composition of this source, here termed Pannonian Upper MAntle (PUMA), is tabulated in the supporting information. The modal mineralogy corresponding to a source with this bulk composition is Ol 0.58 Opx 0.27 Cpx 0.12 Sp 0.03 (in the spinel peridotite field) and Ol 0.62 Opx 0.14 Cpx 0.11 Gt 0.13 (in the garnet peridotite field).
The calculated depth of melting is principally governed by the depth of the phase change from spinel to garnet peridotite. Experimental studies have shown that this transition is sensitive to temperature (Klemme & O'Neill, 2000). Based on their laboratory experiments, Klemme and O'Neill (2000) proposed that this transition occurred between 60 and 80 km for a T P of 1315°C. Thermodynamic modeling by Jennings and Holland (2015), however, suggests that the pressure of the garnet-spinel transition was previously overestimated due to the simplicity of the phase assemblage considered (i.e., Mg-Al-Si without Ca). These authors allowed for the activity of Ca in garnet and show that the transition intersects the solidus at 2.14-2.17 GPa. We used 63-72 km (2.1-2.4 GPa) for the transition from spinel to garnet peridotite.
The calculated MgO concentrations in the basalts were all greater than 8 wt% and were used to estimate the fraction of olivine that was removed from the magmas by crystallization. This was done by comparing the observed MgO and FeO concentrations with those in a primary melt in the mantle, prior to fractionation, using the expressions of Irvine (1979) and O'Nions (1991, 1992). The fraction of olivine removed was between 7 and 14 wt% and has no appreciable effect on the final melt fractions, which are small ( Figure 5).
The melt distributions from the inversions were used to estimate the mantle potential temperature using the parameterization of hydrous mantle melting of mantle peridotite by Katz et al. (2003), which calculates the melt fraction from pressure, temperature, water content, and the modal proportion of clinopyroxene.
INVMEL now includes this parameterization. The potential temperature was estimated by calculating the melting profile for isentropic melting, using the parameterization of Katz et al. (2003) and a water content of 238 ppm (see section 4.2), that minimizes its root mean square difference from that obtained from the inversion. Using the same parameterization and a water content of 100 ppm, the MORB source composition of O'Nions (1991, 1995) generates 7-km thick crust with the average composition of MORB (Gale et al., 2013) if the average potential temperature of the mantle is 1315°C (see McKenzie, 2020, for a comprehensive inversion of Gale et al.'s data).
The sensitivity of the present modeling approach to variations in input parameters was investigated by Brodie et al. (1994) and Klöcking (2018). They considered variations in the source composition and thickness and depth of the melting column and find typical uncertainties of <2% in cumulative melt fraction and 2-5 km in the depth of the top of the melting column, which amount to ±10-30°C in the estimated potential temperature. Placing the garnet-spinel transition zone at greater depth, for example, due to a greater amount of Fe 3+ or Cr in the source (Jennings & Holland, 2015), would act to increase the inferred potential temperature (Klöcking, 2018).

Alkali Basalts
The left column in Figure 5 shows the results of the REE inversions, grouped into four suites corresponding to the locations in Figure 1. It shows the average observed concentration (black circles ±1 SD) and the best-fitting concentration ratios calculated by the inversion (solid lines). The calculated concentrations for the NW and N groups show that the light (L-) REE concentrations are slightly underpredicted and those of the most incompatible elements (Cs, Rb, K, and Ba; third column) slightly overpredicted. These differences may result from using the primitive mantle as one of the end-members to construct a source with ϵ Nd ¼ 3:3. The values of ϵ Nd from basalts from oceanic islands (Stracke et al., 2003) cover a larger range than do those from the Pannonian Basin. Basalts from oceanic islands must result from melting the mantle and not from assimilation of continental crust. Those with negative ϵ Nd values obviously cannot be produced from a source that is a mixture of the primitive mantle (ϵ Nd ¼ 0) and the MORB source (ϵ Nd ¼ 10). The compositions of such enriched regions in the mantle would require a multistage melting model to be more accurately reproduced. The application of such a model for the Pannonian basalts is considered below.
The REE inversion provides a good match to the compositions of the S and SE groups (Banat and Perşani), although the observed slope of the heavy (H-) REEs in the Perşani basalts is flatter than that calculated by single-stage melting. This behavior may result from the source having been depleted and then enriched by a metasomatic melt generated by a small amount of melting in the garnet stability field (see McKenzie & O'Nions, 1995;White & McKenzie, 1995). Here, such multistage melting comprises initial (5%) depletion of the mantle in the garnet-spinel transition zone, followed by source enrichment by a small-fraction (2%) metasomatic melt generated by minor (0.4%) melting of the MORB source at depths where garnet is stable. The dashed line in Figure 5m shows the concentration ratios resulting from such a model, which can accurately reproduce the observed pattern for Perşani. Mantle nodules brought up in Perşani show that lithospheric domains have undergone just such an evolution (Downes et al., 1992;Faccini et al., 2020;Vaselli et al., 1995).
Theoretically, such a multistage model can also improve the fits to the REEs of the NW and N groups. The dashed lines in Figures 5a and 5e show the results of such a model for these groups, whose sources were enriched by adding a small-fraction (3% and 1.5%, respectively) metasomatic melt generated by minor (0.4%) melting of the MORB source at depths where garnet is stable. It is obvious that the more complicated the model, the better the fit that can be obtained. However, none of these groups show the characteristically 10.1029/2020GC009334

Geochemistry, Geophysics, Geosystems
flat HREE pattern as observed in Perşani, meaning that there is presently no strong argument-other than their improving the REE fit-to impose a more complex source model than that which involves single-stage melting. Importantly, both models yield nearly identical estimates of melt fractions, as well as melting depth, meaning that for the present purpose, rather than ad hoc adjustments to the source model, the simplest model is preferred. Indeed, the calculated concentrations of elements other than the REEs and Cs-Ba agree well with those observed ( Figure 5, second and third column), meaning the simple mixing scenario is effective in capturing the key features.
The principal result of the modeling shown in Figure 5 is that the compositions of all the basalts from the Carpathian-Pannonian region require the mantle to have undergone less than 1% melting in the garnet-spinel transition zone. This value is slightly lower than, but probably within error of, previous estimates, that is, 2-3%  or 1-4% (Seghedi et al., 2004). For Perşani, the calculated melt fraction is only slightly smaller than the 1-2% estimated from trace elements in olivine-hosted melt inclusions from Racoş, a lava flow in Perşani (Bracco Gartner et al., 2020).

Mantle Temperature
The presence of young basalts in small volumes that are widely distributed across the Pannonian Basin has important implications for its geothermal structure. The melt distributions calculated by the inversions can be used to estimate the potential temperature of the upper mantle from the parameterization of Katz et al. (2003). The parameterization requires an estimate of the water content of the source, here taken to be 238 ppm, calculated from estimates of the H 2 O content of the primitive mantle (330 ± 55 ppm) and the MORB source (100 ± 50 ppm) (Gibson & Richards, 2018, and references therein), and the average observed value of ϵ Nd (3.3) for the basalts. This estimate is corroborated by observational constraints on the water content of clinopyroxene megacrysts in the Balaton area (Kovács et al., 2020), which translates to 240 ± 90 ppm H 2 O in the source, using their expressions and our average melt fraction ( Figure 5).
The parameterization of Katz et al. (2003) accounts for the presence of H 2 O, but not CO 2 , in the source. Like H 2 O, the addition of CO 2 acts to lower the solidus and reduce the inferred potential temperature (e.g., Dasgupta et al., 2007Dasgupta et al., , 2013. There is at present no direct measure of the CO 2 content of the Pannonian, or indeed most other, primary magmas, because most of the CO 2 is lost due to degassing during magma ascent. While there is evidence for CO 2 -rich phases in the Pannonian lithosphere-particularly in the Balaton area -their presence is clearly linked to (earlier) subduction-related metasomatic processes, not the host basalts

10.1029/2020GC009334
Geochemistry, Geophysics, Geosystems (e.g., Bali, Zajacz, et al., 2008;Bali, Zanetti, et al., 2008;Demény et al., 2004Demény et al., , 2010Szabó et al., 1996). There is no evidence in the compositions of the selected basalts for a strongly carbonated source: they neither show particularly elevated CaO contents nor conspicuous trace element signatures (e.g., Sr, Ba, Zr, Hf; Veksler et al., 1998). In addition, strong effects of CO 2 on the melt composition, and thus inferred P-T conditions, are only expected at pressures >2 GPa, because at lower pressures, there is a major decrease in the capacity of melt to contain CO 2 and depress the solidus (Falloon & Green, 1989). The effect of CO 2 , of lowering the solidus temperature, is also about three times smaller than that of H 2 O (Dasgupta et al., 2007). The effect of CO 2 is therefore likely to be minor, but nonetheless, we note that the parameterization yields an upper bound on the potential temperature. Figure 6b compares the melt distribution from the REE inversion ( Figure 5d) with that calculated from the parameterization. It shows that the two can be closely matched. The corresponding, best-fitting potential temperature to the melt distributions, of both the NW and S groups, is 1257°C. Adding 33°C, which corresponds to an isentropic gradient of 0.5°C/km, this T P corresponds to a real temperature of 1290°C at a depth of 65 km beneath the Pannonian Basin. This value exceeds temperatures estimated from the spinel peridotite nodules (mostly 900-1150°C; e.g., Dobosi et al., 2010;Embey-Isztin et al., 2001;Faccini et al., 2020;Liptai et al., 2017;Vaselli et al., 1995), which is compatible with their being derived from shallower regions in the upper mantle.
The principal question then is whether the thermal structure of the Carpathian-Pannonian upper mantle can account for the generation of melt. To map the thermal structure and lithospheric thickness in the region, we use the surface wave tomographic model of Priestley et al. (2018), which is based on the conversion of a shear wave speed (V Sv ) and anisotropy model for the upper mantle (CAM2016Vsv; an update of PM_v2_2012; Priestley & McKenzie, 2013) to a temperature model as described in Priestley and McKenzie (2013) and Priestley et al. (2018). Importantly, the parameterization T(V s , z) makes no assumptions about the rheological properties and grain size and their relationship to seismic velocity. Fitting T(z) profiles with geotherms allows determination of the depth of the transition from the conducting to advecting geotherm, that is, where heat transport changes from conduction in the shallower mantle to advection in the deeper mantle. This depth corresponds to the base of the lithosphere. Both the lithospheric thickness and temperature data are available from IRIS (2018) (files CAM2016Litho.tgzand CAM2016T-200.tgz, respectively). Figure 6a shows one of the geotherms, in the center of the Pannonian Basin (47°N, 19°E), that was used to determine the lithospheric thickness shown in Figures 1b (map) and 9 (profile). The regional geotherm shows two important features. First, it shows that the lithosphere in the center of the Basin is thin-only about 50 km. This thickness is compatible with previous estimates (e.g., Horváth, 1993;Horváth et al., 2006). Second, it shows that the temperature at the base of the lithosphere is very close to that of the dry solidus. Clearly, small volumes of melt can easily be generated near the base of the lithosphere, which indeed is just where the REE inversion shows the melting columns are located. The presence of any water in the mantle-and it is likely that there is some beneath the Basin (e.g., Aradi et al., 2017;Falus et al., 2008;Kovács et al., 2020;Lange et al., 2019;Patkó et al., 2019)-would facilitate melting at the conditions of the dry solidus.

Mantle Nodules
Like many alkali basalts, those from the Pannonian Basin often contain mantle nodules (e.g., Szabó et al., 2004, and references in the caption to Figure 8). Such nodules have various possible origins. They may come from the mantle with the same composition as that of the source of the basalts. Alternatively, they may come from the older, shallower part of the mantle which was stretched when the Basin formed. Whatever their origin, their composition may have been changed by infiltration of melts (such as their host basalts). A clear example of such infiltration is found in the Nógrád area, where both spinel lherzolite nodules (Liptai et al., 2017) and wehrlite nodules (of metasomatic origin; Patkó et al., 2020) were brought up by the basalts. It is of interest to compare the observed composition and mineralogy of these nodulesparticularly those unaffected by extraneous metasomatic modifications-with that of the expected source of the basalts. This comparison serves to answer the question of whether some of the nodules could be fragments of basalt source rock. By reverse reasoning, finding a suitable match in the nodules also provides further validation of the calculated source used for the inversions.

10.1029/2020GC009334
Geochemistry, Geophysics, Geosystems Figure 7 shows a comparison between the composition of the source used in the inversions and the average observed REE and (other) incompatible element composition of the peridotite nodules from the NW, N, and SE groups (data in supporting information). Because anomalous nodule compositions (e.g., small or otherwise unrepresentative samples) can exert disproportionate control over the average values, we identified statistical outliers in the nodule groups to obtain a more representative average composition. Outliers were identified by calculating the lower and upper quartiles Q 1 and Q 3 ; values outside the range [Q 1 − 1.5(Q 3 − Q 1 ), Q 3 + 1.5(Q 3 − Q 1 )] were rejected. The resulting averages are similar to the medians. Since the amount of melt that a mantle of variable composition will contribute depends on its fertility, the average composition c i was calculated by weighting the observed concentrations c n i by the concentrations themselves: where N is the number of samples. The composition of the nodules is clearly variable; the error bars on the concentrations in Figure 7 correspond to 1 SD. Nonetheless, the figure shows that the average observed REE and incompatible element composition of the Pannonian peridotite nodules is similar to that of the source used in the inversions. This consistency suggests that these nodule suites may in fact be fragments of source rock, which were entrained by the magmas they generated. The REE inversions, however, require the melt to be generated within the garnet-spinel transition zone, and none of the nodules contain garnet. Therefore, as might be expected, they must come from depths that are shallower than the source regions of the basalts.
More insight is provided by comparing the mineralogy and clinopyroxene composition of the nodules and those of the calculated source. The calculated composition of clinopyroxene in the source depends on the mineral mode, which, because the source region spans the garnet-spinel transition zone, in turn depends on depth. For direct comparison with the nodules, it is therefore most informative to consider the source characteristics at depths where spinel, and not garnet, is stable. Figure 8a compares the mineral mode of the source, calculated for a spinel peridotite whose bulk composition is the same as that of the source used in the inversions (Ol 0.58 Opx 0.27 Cpx 0.12 Sp 0.03 ), to the average of that observed in the nodules. In addition, Figure 8b compares the average composition of "LREE-depleted" clinopyroxenes in nodules from the NW Embey-Isztin et al., 2014), N (Liptai et al., 2017), and SE (Faccini et al., 2020)-which are thought to represent the unmetasomatized, or "precursor," mantle beneath their respective regions-  Nions (1991Nions ( , 1995. Published nodule whole-rock trace element data (Embey-Isztin et al., 2014;Liptai et al., 2017;Ntaflos et al., 2017;Vaselli et al., 1995) are tabulated in the supporting information.
with that calculated for clinopyroxene in the source, also at depths where spinel is stable. Both the expected mineral mode and the clinopyroxene composition of the basalt source closely resemble that of the nodule suites and are consistent with their representing the basalt source rock.
In summary, there is strong consistency between the calculated basalt source and the observational data (bulk-rock trace elements, mineralogy, and clinopyroxene REEs) from various suites of mantle "xenoliths" that were brought up by these magmas throughout the region. It is probably better to refer to all such inclusions as nodules, rather than xenoliths, because, like those discussed here, they may be pieces of the source rock, not foreign inclusions. The principal difference between the basalt source and the mantle nodules concerns the depth at which they equilibrated: the garnet-free nodules must originate from depths shallower than the source regions of the basalts.

Melt Generation and Intraplate Volcanism
The REE inversions show that fractional melting of a homogeneous source can reproduce the elemental compositions of the erupted basalts in the Carpathian-Pannonian region. The composition of this source, here termed PUMA (supporting information), is similar to that of some of the mantle material brought up as nodules throughout the region. These results do not imply that the melt source is homogeneous; it rather suggests that any small-scale lithological or compositional heterogeneities present in the source region were homogenized as the melts rose to the surface. It is probable that the source regions contain small-scale inhomogeneities like those beneath other basaltic volcanoes.
The inversions show that the young intraplate basalts in the Carpathian-Pannonian region are sourced from the garnet-spinel transition zone, at depths of about 63-72 km. Figure 9 shows a section across the Pannonian Basin along the profile trace shown in Figure 1b and illustrates the Basin's geophysical architecture derived from the surface wave tomography of Priestley and McKenzie (2013) and Priestley et al. (2018) (see Figure 6a). The base of the lithosphere is at a temperature of about 1300°C and melt is generated within the garnet-spinel transition zone, where the geotherm reaches the solidus. Figure 9 shows that all the  (Aradi et al., 2017;Downes et al., 1992;Embey-Isztin et al., 1989, 1990Faccini et al., 2020;Liptai et al., 2017;Ntaflos et al., 2017;Szabó & Taylor, 1994;Szabó et al., 1995;Vaselli et al., 1995Vaselli et al., , 1996 are tabulated in the supporting information. Note that statistical outliers were identified according to the criteria described in the main text; samples were excluded if outlying modes were identified. (b) REE distributions in clinopyroxene in the basalt source, calculated for a spinel peridotite whose bulk composition is the same as that of the source used in the inversions, versus those observed in "LREE-depleted" clinopyroxene in the nodules (data in supporting information; Dobosi et al., 2010;Embey-Isztin et al., 2014;Faccini et al., 2020;Liptai et al., 2017). Average observed concentrations are weighted by the concentrations themselves (see main text). Error bars are given as ±1 SD. Concentrations are normalized to the MORB source of McKenzie and O'Nions (1991).

10.1029/2020GC009334
Geochemistry, Geophysics, Geosystems magmas beneath the Basin are probably sourced from the asthenosphere, close to the base of the thermal boundary layer.
The thin lithosphere is a result of extension, which occurred when the subduction zone beneath the Carpathian arc rolled back (Balázs et al., 2016;Horváth et al., 2006;Royden, 1988;Ustaszewski et al., 2008). The same process is now taking place in the Aegean further south (e.g., Jolivet & Brun, 2010). Subduction stopped in the northeast when it reached the thick lithosphere of the Russian Platform (see Figure 1b). Because the extension rate was slow, mantle upwelling was not sufficiently fast to generate large volumes of melt. Any small volumes that erupted were buried beneath sediments in those parts of the Basin that were rapidly subsiding (e.g., Zelenka et al., 2004). Heat loss to the surface allowed the upwelling mantle to cool below its solidus and to become part of the mechanical boundary layer. Some of this material has then been brought to the surface as spinel peridotite nodules in the magmas. Yet when volcanism appears to flare up, in the Late Miocene, the extension in the Pannonian Basin had effectively ceased. What causes these melts to form and rise up from beneath the plate's interior?
The gravity anomalies and topography of the Basin (Figure 2) show no systematic pattern of swells and/or troughs at wavelengths between 100 km and that of the Basin, ∼500 km. These features signify that there is no large, hot structure present immediately beneath the plate's interior. Just as the small melt fractions imply, and other studies have construed (e.g., Embey-Isztin, Huismans et al., 2001), the nature of the volcanism is clearly not compatible with an extensive thermal plume. Instead, the lithospheric thickness ( Figure 1b) and geotherm (Figure 6a) show that melting occurs because the base of the lithosphere is close to or at the solidus temperature over a large part of the Basin. This observation is corroborated by the calculated conditions of mantle-melt equilibration based on the major element content of the basalts , which similarly suggest that the sublithospheric upper mantle is at a temperature close to the peridotite solidus.
Only the Perşani volcanics appear to be underlain by slightly thicker lithosphere than those in the Pannonian Basin. However, unlike the vertical resolution (30-40 km), the horizontal resolution of the surface wave tomography is 200-300 km (Priestley & McKenzie, 2013) and is therefore not sufficient to resolve the vertical slab beneath Vrancea (as imaged by e.g., Martin & Wenzel, 2006). Because intraplate volcanism in Perşani developed only about 100 km northwest of this region, contemporaneously with the last episodes of calc-alkaline volcanism in the southern East Carpathians (Seghedi et al., 2016), it should probably be   (Seghedi et al., 2016) show that extensional tectonics were instrumental in the development of volcanism. We concur with Seghedi et al. (2011) that steepening and rollback of the Vrancea slab is most likely to have driven the lithospheric stretching that gave rise to small-degree melting beneath Perşani.
By contrast, intraplate volcanism in the Pannonian Basin during the Quaternary is clearly not the result of ongoing extension. Though the amount of melting is small, the time difference between the extension and volcanism cannot be due to the slow movement of melt. There is strong evidence, both theoretical and observational, that small melt fractions (≤0.5%, down to 0.2%) separate rapidly (on a time scale of ∼1 ka) from their residues, with no appreciable volume remaining in the source region (e.g., Eksinchol et al., 2019;Jull & McKenzie, 1996;Kokfelt et al., 2003;Liang & Peng, 2010;McKenzie, 1985;Slater et al., 2001). Furthermore, unless, for example, water or carbon dioxide is introduced into the mantle, the solidus temperature is a monotonically increasing function of depth. Therefore, it appears that there must be some upwelling to account for the volcanism. Since the lithosphere is cooling and its thickness is increasing after the stretching event, this upwelling and melting is more likely to occur in the asthenosphere than in the lithosphere. This is consistent with our results (Figure 9), which show that the melts seem to originate from the asthenosphere immediately below the lithosphere. Such upwelling can produce melts because the lithosphere is thin and the upwelling material is hotter than that which surrounds it. However, because the volume of melt is so small (i.e., <1% melting, amounting to just a few cubic kilometers of lava throughout the Basin), the upwelling clearly moves mantle material upwards just enough for it to cross its solidus.
The geometry of such upwelling is not well constrained by the surface observations ( Figure 2). The viscosity of the asthenosphere near the top of the mantle is probably about 10 19 Pa·s. Such a low viscosity means that the gravity anomalies and dynamic topography resulting from the hot upwelling will be small. It is then difficult to distinguish the surface expression of such upwelling from gravity anomalies and topography generated by the variations in crustal thickness that resulted from the earlier tectonic deformation. The volume of the volcanism is small, and probably overlies centers of upwelling, but it is not yet possible to confirm the existence or the geometry of the upwelling using other types of observations. Because the topography and gravity anomalies are dominated by the recent tectonics, the presence of young basalts widely distributed across the Basin appears at present to be the only clearly discernible signal of vertical convective motion beneath the area.
It is important to emphasize that the combined geophysical and geochemical constraints described here do not necessitate, and are not compatible with, a major thermal plume as a causal mechanism for melt generation. Instead, the results are more compatible with previous studies that surmised a link between lithospheric thinning and subsequent decompression-induced mantle melting (e.g., Embey-Isztin, Seghedi et al., 2004;Wilson & Downes, 1991). Clearly, only a small amount of upwelling is needed to form minor volumes (up to a few cubic kilometers) of basaltic melt immediately below the lithosphere when its base is close to the solidus temperature. Though the thinned lithosphere of the Carpathian-Pannonian region is the result of stretching, ongoing extension is not a prerequisite for the formation of intraplate volcanism. What appears to be required is a relatively thin lithosphere, below which thermal buoyancy is likely to drive upwelling and hence melt production.

Conclusions
To place constraints on the generation of melt and its source region beneath young (12-0.1 Ma) Carpathian-Pannonian intraplate volcanoes, we modeled the geochemistry of the erupted basalts and mantle nodules and integrated the results with evidence from surface wave tomography, gravity anomalies, and topography.
1. The REE inversion and forward calculation of elemental concentrations are found to provide close fits to the observed compositions of the basalts. The inversion shows that their composition requires the mantle to have undergone less than 1% melting in the garnet-spinel transition zone, corresponding to a depth range of about 63 to 72 km, that is, in the shallow upper mantle just below the base of the lithosphere.

10.1029/2020GC009334
Geochemistry, Geophysics, Geosystems 2. Using the melt distributions from the inversions and a parameterization of mantle melting, we obtain a mantle potential temperature of ∼1257°C. It corresponds to a real temperature of 1290°C at a depth of 65 km beneath the Pannonian Basin and is compatible with the observed volume of volcanic rock and the small melt fractions involved. 3. The composition, mineral modes, and clinopyroxene geochemistry of some of the mantle nodules brought up by the basalts throughout the region closely resemble those calculated for the basalt source. The principal difference is that the basalt source equilibrated at greater depths. 4. The lithospheric thickness and geotherm show that melting occurs because the base of the lithosphere, at ∼50-km depth, is close to or at the solidus temperature over a large part of the Basin. The volcanism appears to originate from upwelling in the asthenosphere just below the thinned lithosphere, which moves mantle material upwards just enough for it to cross its solidus and produce small volumes (up to a few cubic kilometers) of melt. 5. Because the topography and gravity anomalies of the Basin are dominated by earlier tectonic deformation, the post-extensional volcanism seems at present to be the only discernible signal that resolves the details of vertical convective motion beneath the region.

Appendix A: Mineral-Melt Partition Coefficients
Table S1 lists all the mineral-melt partition coefficients for the REEs (used in the inversion) and elements other than the REEs (used in the forward modeling). Partition coefficients are taken from Beattie (1993), Johnson (1998), Mallmann and O'Neill (2009), and McKenzie and O'Nions (1991, except for the REEs in clinopyroxene and garnet. The lattice-strain model of Blundy and Wood (1994) and Wood and Blundy (1997) based on Brice (1975) was used to obtain a consistent set of partition coefficients for REEs in clinopyroxene and garnet and obtain values for those REEs for which no (reliable) measurements are available. Their expression for the partition coefficient for element i (D i ) with ionic radius r i between a particular mineral and melt is where c s and c l are the elemental concentrations in the solid and liquid, respectively; N A is Avogadro's number; R is the gas constant; and T k the absolute temperature. The radius of the crystallographic site (r o ), the Young's modulus (E) of that site, and the strain-free partition coefficient (D 0 ) are constants that depend on the mineral and the valency of the substituting element. These constants were calculated for clinopyroxene and garnet by fitting Brice's expression to the D values of McKenzie and O'Nions (1991) Figure A1. Partition coefficients D for trivalent rare earth elements in (a) clinopyroxene (cpx) and (b) pyrope garnet as a function of ionic and crystal radius (in pm), respectively. Constants D 0 , r 0 , and E for clinopyroxene and garnet were determined by fitting the expression of Blundy and Wood (1994) and Brice (1975)  Geochemistry, Geophysics, Geosystems using T = 1573 K ( Figure A1). The resultant values are 103.0 and 93 pm (r 0 ), 279.5 and 450 GPa (E), and 0.330 and 10 (D 0 ), respectively. The D values for trivalent REEs and Y were calculated assuming substitution in the clinopyroxene M2-site with eightfold coordination and garnet Y-site with sixfold coordination, using the effective ionic and crystal radii of Shannon (1976), respectively.