Geological data were collected in the field during a two visits to the Mount Dallol area on 21–25 January 2018, and 12–22 January 2019. They include observations of geomorphology, structural features, and evidence of hydrothermal activity, and also acquisition of magnetic data which will be presented elsewhere.

Topography acquired by satellite (such as from the Shuttle Radar Topography Mission and the ASTER Global Digital Elevation Map) are not adapted to the study of areas in which topographic variations are below 1 m metre. For this reason, a UAV-based survey of imagery acquisition was implemented during the 2019 campaign for stereoscopic processing. Flight altitude above ground was 100 m, images were tagged in geographic coordinates with the WGS84 datum. They were then projected to the WGS 84 / UTM zone 37N - EPSG:32637 projected coordinate system; topography was calculated and an orthorectified image generated using the MicMac ¹ photogrammetric suite (Rupnik et al., 2017). The obtained pixel size is 0.5 m. Subtle vertical topographic variations < 1 m can be detected using pixel elevation averaging (such as presented here) and other statistical methods (Choe et al., in preparation).

Most of the images taken by civilian and declassified satellites in the visible and near-infrared ranges until 2020 were examined for this study, as well as a subset of images obtained between 2020 and 2023. Those who were found the most useful for this study are listed in Supplementary Table S1. High-resolution images from KeyHole-4B, OrbView-3, Pleiades-1A/B, QuickBird-2, Spot-6, and WorldView-2 were used to map the details of the YLF and associated features at irregular time steps. The benefits of 10–30 m resolution data from ASTER and Landsat-7 lie in their spectral capabilities, making identification of potential surface composition variations between two consecutive passes, usually not more than 1 or 2 months. They were used to determine the succession of events at the time of YLF formation, in 2004–2005 (Supplementary Table S2).

Moreover, the thermal channels of Terra/ASTER were useful to identify possible surface temperature variations associated with the hydrothermal activity of the YLF, both diurnal, and between different orbits. Sentinel 2A/B data were also analysed, but did not yield information not already available using data from the other multispectral satellites. The YLF is nearly fully covered by four 1x1 km pixels from the Terra/MODIS Land Surface Temperature dataset. In spite of this coarse resolution, surface temperatures are measured daily, during day time and night time, making it possible to monitor whether or not development of the YLF in 2004–2005 was associated with significant thermal anomalies at surface.

Satellite images having visible and near-infrared channels were studied with conventional processing for geologic interpretations. Initial files were selected in a radiometrically corrected and projected format, provided by the source image providers. Panchromatic images were enhanced through histogram stretching. Multispectral images were studied under various colour composite combinations. If the number of bands exceeds 3, various combinations were tested so as to reveal optimal rock composition contrasts. The presented image corresponds to this particular band combination, to which histogram stretching was applied in order to emphasize the contrasts. Infrared bands revealed particularly useful due to spectral variations in rock reflectance in this range, whereas visible wavelengths are devoid of major mineral absorptions. The captions indicate which bands were selected in the corresponding figure. In case both a panchromatic sensor and a multispectral sensor exist, pansharpening of the multispectral channels was done using Gram-Schmidt pansharpening (Laben and Brower, 2000; Maurer, 2013).

A composite (red-green-blue) image showing the difference between two measurement times ( t 0 , “before,” and t 1 , “after” the studied event) was generated by subtracting a given band at t 0 from the same band obtained at t 1 and attributing it to the red channel, and doing the same for two other bands and the two other channels.

Thermal information was directly retrieved from the ASTER L2 Surface Kinetic Temperature distributed by the USGS Land Processes Distributed Active Archive Center. The ASTER five thermal infrared channels were used, after atmospheric correction, to determine surface radiance. Surface radiance data were then used to generate emissivity data, from which temperature is derived.

Image processing was done using ENVI 5.5 and ENVI 5.7 (NV5 Geospatial Solutions, Inc., Broomfield, Colorado, U.S.), and the GIS part using Global Mapper 20 to 24 (Blue Marble Geographics, Hallowell, Maine, U.S.).

Fluids were collected at two hydrothermal pools and at Yellow Lake. The samples were collected by completely immersing autoclaved sterile glass tubes. Water temperature, pH and conductivity were measured. After sampling, the glass tubes were hermetically closed with a thick rubber sealing septum. To prevent unwanted leaking or external contamination further, the flasks were firmly sealed by forcibly crimping of an aluminium closure ring on top of the septum. After transport towards the laboratory, the samples were treated and chemically analysed as in Moors et al. (2023).

The uncertainty on the initiation time of the YLF, constrained to the period between October 13 and 14 November 2004 encompasses the whole period of Black Mountain dyke injection inferred from the seismic record, i.e., October 22–November 4 (Nobile et al., 2012). Independently, InSAR modelling constrains dyke intrusion to two events, the main one between October 22 and October 26, and a minor event between October 27 and 1 December 2004. Figure 11 summarizes the time constraints of tectonic and hydrothermal activity at the YLF, compared to the timing of seismic ruptures recorded at nearby seismic stations and dyke emplacement inferred from InSAR modelling.

The model inferred from the InSAR ground deformation calculations for the period 22/09/2004–01/12/2004 includes the Black Mountain dyke, as well as a blind normal fault located west of the dyke (Nobile et al., 2012). The fault is parallel to the dyke and to the Danakil rift margin, where seismic profiles indeed show rift-parallel border faults (Figure 3), one of which might have been reactivated by dyke emplacement (Bastow et al., 2018). The model residual does not show any influence of YLF development on ground deformation during this period. However, when the narrower, October 27—December 1st time span is considered, during which 13% of the dyke was injected (Nobile et al., 2012), the InSAR model residual appears consistent with ground displacement along the YLF (Supplementary Figure S6). Combining information from satellite image monitoring and InSAR modelling, it appears likely that the YLF formed between October 27 and 14 November 2004, consistent with the timing of second magma injection in the Black Mountain dyke, strongly suggesting a genetic link.

The red salt crust exposures have been observed for several decades. Bubbling at Yellow Lake has also been reported since Holwerda and Hutchinson (1968), testifying to permanently pressured groundwater underneath. White Lake, the largest water body in the YLF area, was vigorously bubbling during the 2019 field campaign (Figure 5A), and KeyHole imagery indicates that it is a long-living lake as well, developed from the merging of two smaller lakes visible in 1965 (Figure 6). Evidence of groundwater release, and therefore pressuring underneath, exists at these sites for as long as data are available. It requires a degassing source at depth much before the 2004 Black Mountain dyke intruded the crust. This source may be a magmatic intrusion underneath, old enough for the temperature of the hydrothermal fluids to have cooled to the ambient temperature at surface.

We follow the interpretation of a less than 6,000 years-old sill intruding the salt formations, as advocated by López-García et al. (2020), for the following reasons. First, Holwerda and Hutchinson (1968) mentioned the existence of rare altered basalt fragments in salt dykes on Mount Dallol. Dolerite enclaves were indeed fund in salt dykes on the western slope of Mount Dallol during our 2019 campaign (Supplementary Figure S7), strengthening the hypothesis of a dyke or sill underneath. Secondly, a N-S trending seismic profile acquired for Yara Dallol BV north of Mount Dallol shows that the evaporite beds are increasingly uplifted toward Mount Dallol (Supplementary Figure S8; Environmental Resources Management Southern Africa (Pty) Ltd., 2014), which might be an effect of layer upwarping at the contact of the intruding sill (Galland and Scheibert, 2013).

Sill development from a dyke, as suggested by Warren (2015) at Mount Dallol, is mechanically well understood (e.g., Gressier et al., 2010). In particular, the effect of the specific rheology of salts especially hydrated, on this evolution, has been investigated (e.g., Schofield et al., 2014). The Houston Formation offers a range of hydrated mineral-rich layers (mainly, from bottom to top, kainite, carnallite, bischofite, and carnallite; Rauche van der Klauw, 2015), the mechanical strength of which is small (Urai et al., 1986; 2017; Rowan et al., 2019) and easily further decreased by decomposition and melting at low temperature. The melting temperature of carnallite and bischofite, 110°C–120°C (Warren, 2016), may allow them to melt ahead of a propagating intrusion even before their contact, depending on the combination of reaction kinetics, ambient thermal diffusivity, and intrusion propagation speed. The resulting brines would recrystallize further away around the intrusion as new, more competent, anhydrous or less hydrated minerals (Schofield et al., 2014). The strata of the Houston formation containing hydrous minerals is capped with layers exclusively composed of anhydrous minerals of higher mechanical strength (halite, anhydrite, and sylvite; Rauche and van der Klauw, 2015). These layers would overlie the intruded sill.

If Mount Dallol indeed uplifted in response to sill intrusion, the dimensions of the sill (hereafter called the Dallol sill) should be large enough to uplift Dallol by tens of metres, from which a complex intrusion geometry would be expected; in particular, the Dallol sill may have extended to adjacent areas as thinner, secondary magma sheets connected to the main intrusion. It would have propagated far enough toward Yellow Lake to continue nowadays to pressure groundwater in this area, fed by the fluid-saturated sediments (Talbot, 2008) between the sill and the surface.

From drilling core data, Rauche and van der Klauw (2015) extrapolated the depth of the top of the Marker Beds to 350–400 m south of Yellow Lake (Supplementary Figure S9). The Houston Formation is shallower at Yellow Lake and the southern YLF than in most other place of lake Asale south of Dallol in areas where drilling cores are available. This may be the reason why hydrothermal activity visible at the surface south of Dallol is concentrated in the YLF area instead of anywhere else.

The following two-stage evolution is an attempt to reconcile the long-lasting hydrothermal activity in the YLF area and the recent (2004) opening of the fissure (Figure 14).

1) The Dallol sill, then not yet cooled (López-García et al., 2020), propagated in the Potash Beds of the Houston Formation (Warren, 2015). This stratigraphic level is particularly shallow, and the branch of the propagated sill would probably be closer to the surface in the YLF area than in many other areas south of Mount Dallol (Figure 14A), triggering more visible hydrothermal activity at surface in the fluid-saturated evaporite basin. This situation would have started earlier than the satellite image data record. Hydrothermal activity would also explain the deggasing of White Lake, and be involved in the formation of hydrothermal pools. López-García et al. (2020) already suggested that Yellow Lake hydrothermal activity is an effect of Dallol sill degassing.

In the second stage, the YLF initiates at a site where the salt pile was close to failure (Figure 14C, top). Continuous groundwater pressuring in the YLF area (Figure 14C, bottom), or groundwater pressuring in response to sealing by some salt levels, brought the stress deviator close to rupture, and promoted tensile fracturing (Tingay et al., 2003; Mourgues et al., 2011). The distribution of the dyke, the normal fault and YLF is actually close to models of host rock fractured around intruding dykes in extensional environment or with free lateral boundary conditions. For instance, Rubin and Pollard (1988) and Abdelmalak et al. (2012) found that emplacement of vertical dykes at depth (i.e., dykes whose upper tip does not rotate) promotes normal faulting on the dyke upper lateral side, and tension fracturing at a higher level and to the surface. Physical models that include mechanical layering and expandable lateral boundary conditions, both particularly adapted to the lake Asale conditions, yielded similar conclusions (Mastin and Pollard, 1988). Following these results, formation of the YLF may be viewed as an anticipated consequence of the intrusion of the Black Mountain dyke.

It is also possible that a proto-YLF already existed in the past, although not observed on satellite imagery before 2004. For instance, Figures 6A, B shows linear patterns in 1965 and 2003 which be interpreted as surface flows, but their linear pattern as well as the inability of the seasonal floods to remove them from the surface between 1965 and 2003 favours a tectonic instead of a surficial origin.

In the study area, the Dallol sill intrusion propagating in one of the weakest evaporite levels generates, on the one hand, long-lasting, diffuse hydrothermal activity in the evaporites, and on the other hand, mechanical changes in these evaporites that help focus further deformation. Dyke injection then provides the deviatoric stress increase that triggers directional, fissure-driven, hydrothermal activity. Does such a magmatic-tectonic-evaporitic-hydrothermal system exist in other extensional environments?

The scale of the processes studied here goes actually beyond the resolution that most data from past or recent offshore cruises can reveal. Not surprisingly, therefore, the proposed mechanism has not been described in the literature yet. It may, however, be operative at slow and ultraslow spreading centres.

One such extensional system may be the slow-spreading Red Sea spreading centre. The Danakil depression is an incipient spreading ridge connected to the Red Sea. Using InSAR data, Moore et al. (2019) interpret stacked large melt lenses in the upper crust feeding shallow magma chambers in the Erta ‘Ale volcano, which sits in the Danakil depression 25 km south of the YLF. North of the Danakil depression, the Red Sea seafloor consists of a volcanic, oceanic crust (Augustin et al., 2021; Delaunay et al., 2023) widely covered by evaporites (Mitchell et al., 2010; Augustin et al., 2014; Feldens and Mitchell, 2015; Delaunay et al., 2023). Moreover, seismic profiles across the central Red Sea by Sang et al. (2023) show disturbances in the evaporite sequence which may be interpreted as interstratified sills. The conditions for the interplay between evaporites, dyke and sill intrusions, and hydrothermal circulation as discussed in the present work may therefore plausibly extend from the Erta Ale to the whole Red Sea.

Another possibly relevant slow-spreading ridge may be the ridge entering the Gulf of California, separating Baja California from North America, where the seafloor of some segments is also partly covered by evaporites, deposited on top of the volcanic basement (Miller and Lizzaralde, 2013; Helenes et al., 2020). For instance, at the axis of the Pescadero Basin, Ramírez-Zerpa et al. (2022) showed from seismic reflection that hydrothermal vents developed on sediments as a result of the intrusion of sills underneath along the rift axis. Vent activity is explained by advective flow in the possibly saturated sediments above the sills. This interpretation is close to the interpretation of hydrothermal activity at White Lake and Yellow Lake in the YLF area (Figure 14A).

Fracture-controlled hydrothermal activity in response to dyke emplacement as in Figure 14B may still be considered as a viable mechanism for linear hydrothermal vents even in the absence of salts, because the Mastin and Pollard (1988) dyke emplacement model which it is inspired of does not specifically require evaporites. For instance, McDermott et al. (2022) noted that in addition to dyke-controlled hydrothermal vents at the East Pacific Rise axis, off-axis fault-controlled hydrothermal vents also formed at a distance of ∼750 m to the ridge axis. This distance is smaller than, but of the same order of, the 2 km measured in the Dallol area (Figure 14B). Although the East Pacific Rise is a fast-spreading ridge, involving significantly different relationships between tectonics and magmatism from those prevailing in the Dallol area, mechanical relationships between dyking and fracturing might also follow the Mastin and Pollard (1988) model, providing a potential explanation for the 750 m offset.

Generally, sill and hydrothermal activity offset with respect to the active spreading axis have been reported from ridges at all spreading rates. On the Mid-Atlantic Ridge, the Trans-Atlantic Geotraverse (TAG) hydrothermal field was shown from bathymetry, seismic refraction data, and earthquake distribution, to be offset from the axial magma chamber in relation to a detachment fault system (deMartin et al., 2007; Szitkar et al., 2019), as also observed on numerous oceanic core complexes of the slow and ultraslow poorly magmatic spreading centres (e.g., Dyment et al., 2018; Zhou et al., 2022). At the intermediate Juan de Fuca ridge, where the crust builds by dyking and sea-floor eruptions, seismic reflection profiles reveal the presence of sills offset from the axial magma centre. Hydrothermal vents are also offset from the spreading axis (Canales et al., 2009). Higher-resolution bathymetry and geophysics in such areas may reveal in the future whether relationships between dyke and sill emplacement on the one hand, and hydrothermal activity on the other hand, such as proposed in this work is frequent at oceanic spreading centres.