Mineralochemical Mechanism for the Formation of Salt Volcanoes: The Case of Mount Dallol (Afar Triangle, Ethiopia)

A genetic model is proposed for the formation and evolution of volcano-like structures from materials other than molten silicate rocks. The model is based on Mount Dallol (Afar Triangle, Ethiopia), currently hosting a conspicuous hydrothermal system with hot, hyper-acidic springs, forming a colorful landscape of unique mineral patterns. We reason that Mount Dallol is the last stage of the formation of a salt volcano driven by the destabilization of a thick sequence of hydrated minerals (the Houston Formation) after the emplacement of an igneous intrusion beneath the thick Danakil evaporitic sequence. Our claim is supported by field studies, calculations of the mineral/water volume balance upon mineral dehydration, and by a geothermal model of the Danakil basin predicting a temperature up to 220 °C at the Houston Formation after the intrusion of a basaltic magma without direct contact with the evaporitic sequence. Although insufficient for salt melting, this heating triggers mineral dehydration and hydrolysis, leading to a total volume increase of at least 25%. The released brine is segregated upward into a pressurized chamber, where the excess volume produced the doming of Mount Dallol. Later, the collapse of the dome formed a caldera and the emission of clastic flows. The resulting structures and materials resemble volcanic lava flows in distribution, structure, and texture but are entirely made of salty materials. This novel mechanism of the generation of pressurized brines and their later eruption extends the relevance of volcanologic studies to lower temperature ranges and unanticipated geologic contexts on Earth and possibly also on other planets.


Methodology for computing volume increases by dehydration
Many previous works (mainly on the engineering of industrial Mg production from bischofite) on the thermal decomposition of hydrated magnesium chloride phases have been published; e.g. see references in Table   1 1 ,where are compiled studies on the dehydration and hydrolysis reactions. The TGA data from that work (see Fig. 7 of ref 1 ) have been used for the following calculations. In the present study, two different samples were used: natural bischofite from Antofagasta Region, Chile, obtained from the brines concentration process and Merck-branded synthetic (> 99% purity) bischofite. The dehydration curves for both samples are shown in Fig. S5. The main differences between the two samples include the weight loss, that is from 5% to 10% less in the natural sample, and the first dehydration step, that goes to the tetrahydtate in the syntetic sample and to the pentahydrate in the natural one. Both differences can be attributed to the presence of impurities present in the natural sample (NaCl, Li 2 SO 4 •H 2 O, KCl and KCl•MgCl 2 •6H 2 O) 1 , leading to a reduced weight loss due to the presence of anhydrous minerals, and to a complex thermal decomposition due to the presence of other hydrated minerals. Since we are interested in the behavior of pure bischofite, while chemical, thermodynamical and structural (density) data are not available for the hypotetical pentahydrate, we used the synthetic sample for presenting the following results.
These TGA data were processed for bischofite, and for other hydrated phases below, by: 1. Digitizing the plot data. Using the software xyscan 2 from the Yale University, a set of representative ( , weight loss) points (24 to 40) on the curve was recorded.
2. Interpolating the data. From these points, a data set was spline-interpolated using the software R 3 into a equispaced data set for T=25,26,27, …499,500.
3. Differentiation. The first derivative of the interpolated dataset was computed using R so that each phase transition shows as a peak spanning the corresponding temperature range and having a maximum at the temperature of the maximum transformation rate. 4. Fitting. These derivatives were fitted (using R) to the sum of Exponential-Gaussian Hybrid distributions to account for peak asymmetry 4 , being the number of decomposition steps of the given phase.

5.
Integration. The resulting peaks were integrated separately to obtain smooth, noise-free, zerobackground, cumulative weight loss for each dehydration steps. These functions, after normalization to remove experimental/theoretical differences (always smaller than 10%), are the final product of experimental data processing. The distribution of all mineral phases, the yield of the decomposition reaction, and the volume balances were computed using those functions. Weight loss curves in units of % mass, grams and moles were computed at this stage.
6. Separation of parallel reactions. For the case of the MgCl 2 *H 2 O decomposition, as discussed above, the parallel equations 4 and 14 were taken into account. The weight loss function obtained in the previous step was decomposed in two for the respective reactions using the -dependent ratio as P HCl 14 /P 2 14 explained before. Figure S5 shows the original TGA data for bischofite as reported in ref 1 . Using these data and following the process described above and illustrated in Fig. S6, we obtained the amount of each of the mineral phases and liquids illustrated in Figs S7 and S8.    To convert these mole fractions into volume fractions, the density of the different phases must be considered.
The density data used in this work were computed from crystallographic data, (multiplicity times the molecular weight divided by the unit cell volume) obtained from the American Mineralogist Crystal Structure Database 5 .
Data for the oxychloride are not available from this database, and were obtained from the PAULING FILE Multinaries Edition one 6 . Data for the hexa-hydrate are from ref 7 , and for the tetra-di-and mono-hydrates from ref 8 . Table S1 summarizes the data used. The density of water as a function of pressure and temperature was computed using the IAPWS Formulation 1995 9 implemented in the R package IAPWS95 10 . The P/T dependency of the liquid/gas boundary, the triple and critical points, defining the state of water in the brine chamber, and the response of these fluids to heating and decompression, were obtained from the same source.
Volume fractions were calculated using data in Table S1. Figure S9 summarizes the evolution of the thermal decomposition of bischofite in terms of volume changes. The initial point is 1m 3 of pure bischofite. Successive dehydration of this volume of hexa-hydrate (black thin solid-line curve) produced a decreasing volume of the tetra-di-and mono-hydrate (thin solid-line red, green, blue and cyan curves, respectively) and the corresponding volume of water (dashed lines of the same color).
The volume of HCl is not plotted because it is produced as gas that get dissolved in the water, increasing nonlinearly the water volume. The water volume increased even after the decomposition due to the reduction of water density with increasing temperature. The solid lines show total volumes (for solids, fluids and total).
We have modeled the two-step dehydration of carnallite, according reactions to 17 and 18 and using the same methodology described for bischofite. The only difference is that the peak corresponding to the first dehydration has a wide foot, which does not fit well with a EGH peak. Matching it to the sum of two such curves improves a lot the coordination of the overall decomposition.
In the case of carnallite, volumes computed from this model were based on the crystallographic data shown in Table S2. Data for the hexa-hydrate were obtained from the American Mineralogist Crystal Structure Database (database code "amcsd 0001005") 11 . The data for the anhydrous chloride were obtained from the JCPDS no.
01-074-2233 reccord 12 . No unit cell information is included in this record, just an unindexed list of powder diffraction peaks 13 . From this list we have computed a unit cell and indexing list using the dicvol 14 software.
This unit cell is the one listed in Table S2 and used in the calculations.
Using these data, the volume balance upon thermal decomposition of carnallite has been computed using the same methodology as in bischofite. The result of this model is shown in Fig. S10.  The initial point is 1 3 of pure carnallite. Successive dehydration of this volume of hexa-hydrate (thin black solid-line curve) produced a decreasing volume of the di-hydrate (thin green solid-line curve) and anhydrous salt (blue thin solid line), and the corresponding volume of water (dashed lines of the same color). The water volume increased even after the decomposition due to the reduction of water density with increasing temperature. The thick solid lines show total volumes (for solids, fluids and total).

The impact of the hydrolysis reactions
As discussed above, up to 200 ºC the total volume increases as water accumulates, and causes a volume increase of almost 25%. For higher temperatures, the dominant reaction is the hydrolysis of the mono-hydrate, that releases HCl, but not water. So, basically, the system behaves as a source of water up to 200 °C and a source of HCl for T>200 ºC. The effect of HCl release on total volume is not included in Fig. S9, because density data for HCl saturated solutions at these P/T conditions temperatures is not available. Figure S11 compares the density of water at 21.2 MPa (the computed hydrostatic pressure) and the corresponding temperature and density of a solution at atmospheric pressure and at the same temperature. Data for these solutions at atmospheric pressure are tabulated for concentration values from 0 to saturation in ref 15 . These data were fitted as linear functions of the concentration and temperature and extrapolated to the 25<T<500ºC range. Figure S12 shows the concentration of a hypothetical solution produced by dissolving the temperaturedependent HCl yield by hydrolysis reactions into the water produced by the previous dehydration reactions.
The final concentration is slightly higher than the solubility at room temperature and pressure (around 38%), so probably some HCl is present as gas, but solubility at the simulated conditions is not known, so the presence of this gas is hypothetical. In any case, solutions with concentration close to 40% have negative pH values close to -1 or -2 at room temperature.

Thermal evolution of the Dallol area
The heat transfer equation describing the transient temperature field in the absence of a heat source in a lagrangian framework, where , , , and represent temperature, time, density, specific heat and thermal conductivity respectively, was numerically integrated using a finite difference algorithm with explicit Euler timestepping: (2) were is an empiric parameter. Values for , , and are listed for the involved lithologies in Table S3 0 (adapted from ref 16 ). Figure S13 shows the resulting thermal diffusivities ( ) used in the numerical calculation for each lithology and temperature.    Table S3). Note that this value is not homogeneous within each lithology because it 0 is temperature-dependent. These variations are particularly important within salt and sandstone. During the calculation, each 100 iterations, the Temperature matrix (Fig. S16) is saved to disk and a plot like those in Fig. 9 is generated for this time-step. These figures were created programmatically using R 19 , using the ggplot2 graphical package 20 , under the control of the calculation script that generates the corresponding R script and launches the application.
A video showing all the generated frames in sequence was encoded from the generated plots using the ffmpeg software 21 , and can be downloaded as part of the supplementary material.