Abstract

During the development of continental rifts, strain accommodation shifts from border faults to intra-rift faults. This transition represents a critical process in the evolution of rift basins in the East African Rift, resulting in the focusing of strain and, ultimately, continental breakup. An analysis of fault and fluid systems in the younger than 7 Ma Natron and Magadi basins (Kenya-Tanzania border) reveals the transition as a complex interaction between plate flexure, magma emplacement, and magmatic volatile release. Rift basin development was investigated by analyzing fault systems, lava chronology, and geochemistry of spring systems. Results show that extensional strain in the 3 Ma Natron basin is primarily accommodated along the border fault, whereas results from the 7 Ma Magadi basin reveal a transition to intra-rift fault–dominated strain accommodation. The focusing of strain into a system of intra-rift faults in Magadi also occurred without oblique-style rifting, as is observed in Ethiopia, and border fault hanging-wall flexure can account for only a minor portion of faulting along the central rift axis (∼12% or less). Instead, areas of high upper crustal strain coincide with the presence of hydrothermal springs that exhibit carbon isotopes and N2-He-Ar abundances indicating mixing between mantle-derived (magmatic) fluids and air saturated water.

By comparing the distribution of fault-related strain and zones of magmatic fluid release in the 3 Ma Natron and 7 Ma Magadi basins, we present a conceptual model for the evolution of early-stage rifting. In the first 3 m.y., border faults accommodate the majority of regional extension (1.24–1.78 mm yr–1 in Natron at a slip rate ranging 1.93–3.56 mm yr–1), with a significant portion of intra-rift faulting (38%–96%) driven by flexure of the border fault hanging wall. Fluids released from magma bodies ascend along the border fault and then outward into nearby faults forming in the flexing hanging wall. By 7 m.y., there is a reduction in the amount of extension accommodated along the border fault (0.40–0.66 mm yr–1 in Magadi at a slip rate ranging from 0.62 to 1.32 mm yr–1), and regional extension is primarily accommodated in the intra-rift fault population (1.34–1.60 mm yr–1), with an accompanying transition of magmatic volatile release into the rift center. The focusing of magma toward the rift center and concomitant release of magmatic fluids into the flexing hanging wall provides a previously unrecognized mechanism that may help to weaken crust and assist the transition to intra-rift dominated strain accommodation. We conclude that the flow of magmatic fluids within fault systems plays an important role in weakening lithosphere and focusing upper crustal strain in early-stage continental rift basins prior to the establishment of magmatic segments.

INTRODUCTION

The initiation of continental rifting is fundamental to plate tectonic theory, yet the combination and relative importance of processes driving the initial breakup of continental lithosphere remain largely enigmatic. The tectonic forces driving rift initiation (e.g., gravitational potential energy, basal shear from mantle convection; Coblentz and Sandiford, 1994; Bird et al., 2008; Stamps et al., 2015) are theoretically too small to rupture thick lithosphere, particularly in the absence of tractions provided by boundary forces acting on the sides of lithospheric plates (e.g., slab pull; Buck, 2004; Stamps et al., 2014). However, some continental rift zones are also sites of asthenospheric and lithospheric heating (e.g., Kenya Rift; Rooney et al., 2012) that may predate rifting (Courtillot et al., 1999; Campbell, 2005), with subsequent tectonic thinning commonly accompanied by decompression melting and the rise of volatiles and magma through the lithosphere (McKenzie and Bickle, 1988; White and McKenzie, 1989; Geissler et al., 2005; Lee et al., 2016). Therefore, intrusion of magma and volatile release are thought to play a major role in the continental rifting process (Buck, 2004; Reyners et al., 2007; Lindenfeld et al., 2012; Rowland and Simmons, 2012; Ebinger et al., 2013).

The buoyancy of magma enables dike intrusions at continental rift settings to ascend from melt generation zones in the mantle toward the surface (Buck, 2004). Dike opening accommodates extensional strain, while the transfer of heat from intrusions weakens the crust locally (e.g., Buck, 2004; Calais et al., 2008; Bialas et al., 2010; Daniels et al., 2014). The rise of magmatic volatiles from mantle depths may also assist rifting. These fluids are hypothesized to mechanically weaken lithosphere through increases in pore-fluid pressure (Sibson, 2000; Reyners et al., 2007) and hydration mineral reactions (Moore and Rymer, 2007). Earthquake swarms signaling brittle failure in the lower crust and upper mantle have been detected beneath cratonic lithosphere in the East African Rift (EAR), and interpreted as the rise of fluids through the plate (Lindenfeld et al., 2012; Albaric et al., 2014; Lee et al., 2016). Soil CO2 flux and carbon isotope data collected in a companion study (Lee et al., 2016) from the Natron basin, Tanzania, and Magadi basin, Kenya, support these interpretations, revealing that mantle-derived CO2 ascends to the surface along deeply penetrating faults in parts of the EAR.

The early-stage (the first 10 m.y.) development of continental rift basins is recorded in extensional fault systems observed at the surface. For many continental rifts, active deformation is concentrated in graben or asymmetric half-graben depressions with accompanying synrift sedimentation and volcanic resurfacing (Baker and Mitchell, 1976; Baker, 1986; Rosendahl, 1987; Scholz et al., 1990; Gawthorpe and Leeder, 2000; Ebinger and Scholz, 2012). Sedimentary cover often conceals fault scarps, resulting in uncertainty regarding the geometry and kinematics of fault populations, as well as calculations of bulk strain and time-averaged extension rates. Although these difficulties arise in many parts of the EAR, faults are well exposed in some basins in Ethiopia and the Kenya Rift (Fig. 1B; Le Turdu et al., 1999; Gupta and Scholz, 2000; Manighetti et al., 2001; Vetel et al., 2005; Agostini et al., 2011a). Well-exposed fault systems in the Kenya Rift include the Kino Sogo fault belt of the Turkana basin (Vetel et al., 2005), the Maji-Moto fault system of the Baringo basin (Le Turdu et al., 1999; Le Gall et al., 2000), and faults of the Magadi basin (Baker, 1958, 1963; Baker and Mitchell, 1976; Le Gall et al., 2008). A common feature of these fault systems is that they formed in young volcanic units (ca. 3 Ma or younger) with limited sedimentary cover (Baker and Wohlenberg, 1971; Vetel et al., 2005). Despite covering earlier-formed faults, widespread lavas also provide constraints on the accumulated fault strain since the time of their emplacement (Baker and Mitchell, 1976; Baker et al., 1988; Vetel et al., 2005; Le Gall et al., 2008). Therefore, these locations have served as focus sites for studies addressing the growth and kinematics of fault systems in the Kenya Rift of the EAR (Le Turdu et al., 1999; Morley, 1999; Le Gall et al., 2000, 2008; Vetel et al., 2005).

By comparing fault systems in basins of different ages across the Eastern rift of the EAR (Fig. 1B), we can infer how upper crustal strain is distributed within rifts as they evolve through time (Hayward and Ebinger, 1996; Ebinger, 2005; Corti, 2009; Agostini et al., 2011a). Similarly, geophysical and geochemical studies along the Eastern rift provide insights into how magma and volatile systems assist rift basin development (Birt et al., 1997; Ebinger and Casey, 2001; Pik et al., 2006; Albaric et al., 2014; Halldórsson et al., 2014; Keir et al., 2015). This study uses fault population data and lava chronology in well-exposed fault systems to investigate the evolution of fault-related strain, specifically during early-stage rifting (the first 10 m.y.). To achieve this we target two rift basins, the 3 Ma Natron (Foster et al., 1997) and 7 Ma Magadi basins (Crossley, 1979) (Fig. 1), which are in relatively early developmental stages. These data are examined alongside recently acquired spring geochemistry to investigate the role of magmatic volatile release in rift basin evolution. We show that areas of high fault strain coincide with zones of magmatic volatile release, revealing for the first time a strong link between the evolution of upper crustal strain and release of fluids from deep magma bodies in the EAR.

RIFT BASIN DEVELOPMENT IN THE EASTERN RIFT OF THE EAR

The EAR forms a >3000-km-long zone of individual rift basins, which are typically ∼100 km long and tens of kilometers wide (Fig. 1B). It is the type example of active continental breakup occurring through extensional faulting, magmatism, and ductile thinning of lithosphere. Rifting in the central parts of the EAR occurs at the eastern and western edges of the Tanzanian craton, dividing the system into distinct Eastern and Western rifts (Rosendahl, 1987).

Faulting in the Eastern rift initiated ca. 25 Ma in north Kenya and Ethiopia (Ebinger et al., 2000). Geological and geophysical investigations of rift basins of different ages in the Eastern rift have highlighted major shifts in fault and magma systems during basin development (Hayward and Ebinger, 1996; Ebinger and Casey, 2001; Ebinger, 2005; Corti, 2009; McCartney and Scholz, 2016) (Fig. 2). In early-stage basins (younger than 10 m.y.), strain is focused along rift-bounding faults (termed border faults), which are often hundreds of kilometers long and can accrue >1000 m of throw. Flexure of the downthrown border fault hanging wall produces extensional strains (Vening Meinesz, 1950; Lavier et al., 2000; Cowie et al., 2005), which can manifest as smaller intra-rift faults that are typically tens of kilometers long with tens to hundreds of meters of throw (McClay and Ellis, 1987; Imber et al., 2003; Ebinger, 2005; Corti, 2009). Over the course of rift evolution in the EAR, the locus of strain has migrated away from border faults into these developing intra-rift fault systems (Hayward and Ebinger, 1996; Wolfenden et al., 2004; Corti, 2009). For evolved basins in the Eastern rift (older than 10 m.y.), strain migrated into en echelon magmatic segments in the center of the rift, with dimensions approximately 60 km by 20 km (Ebinger and Casey, 2001; Keranen et al., 2004; Maguire et al., 2006). Deformation in these magmatic segments manifests at the surface as slip along intra-rift faults (Rowland et al., 2007; Corti, 2009). At deeper levels of the crust (∼5–20 km), strain is primarily accommodated through swarms of rift-parallel dikes (Ebinger and Casey, 2001; Rooney et al., 2005; Keir et al., 2006; Wright et al., 2012).

The transition from border fault to intra-rift fault-dominated strain accommodation is a critical process to focus strain as continental rifts evolve toward proto–spreading ridges (Corti, 2009). However, the timing and magmatic-tectonic processes controlling this transition are debated. Evidence of extensive diking below magmatic segments from shear-wave splitting and seismic tomography highlights an active magmatic influence on rift basin development (Keranen et al., 2004; Keir et al., 2005, 2011; Kendall et al., 2005; Bastow et al., 2010). The transition to intra-rift faulting in these segments is also accompanied by extensive volcanism in the rift center, often manifesting as rift-parallel chains of scoria cones and eruptive fissures (Rooney et al., 2011, 2014; Keir et al., 2015; Muirhead et al., 2015; Mazzarini et al., 2016). Recent fault-slip events on intra-rift faults in the Afar depression and the Natron basin coincide with the emplacement of rift-parallel dikes, suggesting that favorable stress conditions for intra-rift faulting are achieved during diking events (Rubin and Pollard, 1988; Wright et al., 2006; Rowland et al., 2007; Baer et al., 2008; Calais et al., 2008; Belachew et al., 2013).

Others suggest that the initiation of intra-rift faulting is a product of oblique continental rifting, and occurs independent of magmatic processes below the rift center (e.g., Corti, 2008). For example, in the Main Ethiopian Rift, the earlier formed (before 2 Ma) large border faults trend oblique to the regional extension direction (as much as 45°; Keir et al., 2015), whereas younger intra-rift faults (ca. 2 Ma to present) trend normal to the regional extension (Bonini et al., 1997; Agostini et al., 2011a). Oblique continental rifting in the Main Ethiopian Rift was previously attributed to changes in the relative motions between the Somalian and Nubian plates, where an initial Miocene–Pliocene phase of approximately northwest-southeast extension was followed by approximately east-west extension during the Quaternary (e.g., Bonini et al.,1997; Boccaletti et al., 1998; Wolfenden et al., 2004). The resulting change in plate kinematics produced sinistral motions on the earlier formed northeast-southwest–trending border faults, and drove the formation of north-south–trending intra-rift faults (Bonini et al., 1997; Corti et al., 2003). Alternatively, oblique rifting in the Main Ethiopian Rift may not require changes in far-field plate motions, but reactivation of preexisting lithospheric weaknesses trending oblique to the regional extension direction (Corti, 2008; Agostini et al., 2009, 2011b). Analog investigations of rift-oblique extension under this model show that the earliest stages of basin development are still characterized by oblique slip on border faults, and then followed by a second stage of dip-slip motion within intra-rift faults, that group into discrete populations arranged en echelon in the rift center (Clifton et al., 2000; Clifton and Schlische, 2001; Corti, 2008; Agostini et al., 2011b). These experimental results are consistent with fault patterns in the Wonji fault belt of the Main Ethiopian Rift (Mohr, 1967; Gibson, 1969; Chorowicz et al., 1994; Bonini et al., 1997; Boccaletti et al., 1998; Agostini et al., 2011a, 2011b; Corti et al., 2013). The experiments suggest that the transition to intra-rift faulting may be driven by an upwelling mantle that (1) imposes upward-bending forces on the lithosphere that restrict slip on border faults, and (2) drives outward flow of an overlying (thinning) lower crust, leading to shear stresses acting on the base of the brittle crust below the rift center (Corti, 2008; Agostini et al., 2009). In the context of these models, the development of intra-rift fault systems may predate magmatism, and uprising of warm mantle and enhanced melt production occurred in areas of focused lithospheric thinning below intra-rift fault systems (Corti, 2008, 2009) to produce the magmatic segments imaged geophysically below the rift (Keranen et al., 2004; Mackenzie et al., 2005; Cornwell et al., 2006; Daly et al., 2008).

The studies described here highlight a fundamental question related to the magmatic-tectonic processes controlling continental rifting. Does rift basin architecture evolve in response to changes in plate kinematics and oblique rifting, or is it controlled by deeper processes, such as decompression melting and enhanced magma production (and related volatile release) below the rift? Early-stage rifts actively transitioning from border fault to intra-rift fault–dominated strain accommodation represent prime targets to investigate processes controlling rift basin evolution. For example, if magmatism drives strain localization in evolving basins, then changes in fault system dynamics should be accompanied by changes in underlying magma systems and/or zones of magmatic volatile release. The 3 Ma Natron and 7 Ma Magadi basins are currently in their earliest stages of rifting, and provide an ideal location to test and advance existing models of continental rift initiation and evolution.

METHODS

Field and remote-sensing analyses were used to investigate faults in the Magadi and Natron basins. Fault traces were mapped from 0.5-m-resolution aerial photographs (provided by the Polar Geospatial Center, University of Minnesota, St. Paul). Individual segments were identified by the presence of visible fault tips. Following the method of Vetel et al. (2005), segments were interpreted to form part of a longer, segmented fault if they (1) exhibited relay zones with small offsets relative to the segment length, (2) had similar dip directions, and (3) displayed regular and systematic changes in throw patterns between segments, indicating that segments are geometrically and kinematically coherent (Peacock and Sanderson, 1991; Walsh and Watterson, 1991; Cartwright et al., 1995; Bohnenstiehl and Kleinrock, 1999; Walsh et al., 2003). Fault throws were measured from 30 m SRTM (Shuttle Radar Topography Mission) data, which have an estimated vertical error of 7.6 m (95% confidence) in the study area (Rodríguez et al., 2005). These remote-sensing analyses were verified with three field investigations in the Magadi and Natron basins from 2012 to 2014 (Fig. 3). Throws, and occasionally lengths, were measured on border fault and intra-rift fault segments in the field using a Trimble global positioning system (GPS) (GeoExplorer 6000 Series; GeoXT Handheld). Collected GPS data had a mean vertical error of 3.0 m (95% confidence) (Fig. 4). Faults were also examined for evidence of recent rupturing, such as fresh scarp surfaces or open fissures (Figs. 3A–3D). Throws were converted to displacements and heaves by assuming an optimal 60° dip for normal faults (Fig. 5). These data relate to fault displacements in only the most recently emplaced volcanics (e.g., Magadi Trachyte, Gelai lavas, Singaraini basalt; Baker and Mitchell, 1976; Mana et al., 2015), and do not reflect possible displacements that predate the emplacement of these units. Each basin has been affected by widespread volcanic resurfacing, where voluminous outpourings of lava erased the surface expression of earlier faulting (e.g., the 295 km3 Magadi Trachyte; Baker and Mitchell, 1976; Guth, 2015). The strain analyses performed in this study, and described in the following, relate to horizontal extensional strains since the emplacement of the lava units in which faults are contained. Maximum displacements (Dmax) refer only to displacements accrued on surficial fault scarps, and do not reflect possible fault displacements that occurred prior to emplacement of the faulted lavas. Maximum ages represent the maximum age of only the surficial fault scarp, and not necessarily the fault plane at depth.

Strain Calculations

Fault-scaling relations in the Magadi and Natron basins were investigated to calculate horizontal normal strain. First, to estimate individual fault displacements within the large fault population (>1000 faults), fault displacement-length ratios were constrained in the Magadi-Natron region. The conventional relationship between maximum displacement, Dmax, and fault length, L, for continental faults is defined as: 
graphic
where g is the scaling factor or displacement-length coefficient, and n is a power-law exponent that defines the rate of displacement accumulation with increasing fault length (Scholz and Cowie, 1990; Cowie et al., 1993; Dawers et al., 1993; Cartwright et al., 1995; Schultz et al., 2006). Although the value of n for individual fault data sets has been shown to be as high as 2 in some cases (Walsh and Watterson, 1988; Marrett and Allmendinger, 1991; Gillespie et al., 1992), linear elastic fracture mechanics models predict n = 1 (Cowie and Scholz, 1992; Bonnet et al., 2001), which is consistent with the full range of global displacement-length data sets (Cowie and Scholz, 1992; Gudmundsson, 1992; Dawers et al., 1993; Cowie et al., 1993, 1994; Carbotte and MacDonald, 1994; Dawers and Anders, 1995; Cartwright et al., 1995; Schlische et al., 1996; Bohnenstiehl and Kleinrock, 1999, 2000; Schultz et al., 2006, 2010). Therefore, studies quantifying extensional strains in fault populations often assume a linear relationship between fault displacement and length (e.g., Cowie et al., 1993; Carbotte and MacDonald, 1994; Schultz et al., 2010; Nahm and Schultz, 2011). The displacement-length coefficient, however, varies between fault populations depending on the tectonic setting and mechanical properties of the host rock (Cowie and Scholz, 1992; Bohnenstiehl and Kleinrock, 2000), and therefore must be determined by analyzing displacement-length ratios for each system of interest. On Earth, values of g typically range from 10–1 to 10–3 (Schultz et al., 2010). Constraining this relationship within a given fault population (see following), and assuming a characteristic fault dip at depth, allows an estimate of Dmax on any fault within a population from only measurements of L (Cowie et al., 1993, 1994; Carbotte and MacDonald, 1994; Nahm and Schultz, 2011).

Although Dmax can be inferred from fault length data, estimates of crustal strain are based on average, rather than maximum, displacements along faults. The displacement profile along the trace of an individual fault tends to be bell shaped or elliptical (Dawers et al., 1993; Cartwright et al., 1995; Gupta and Scholz, 2000; Kattenhorn and Pollard, 2001), where the displacement remains relatively constant over at least the central third of the fault before reducing to zero at the fault tips (Walsh and Watterson, 1988; Cowie et al., 1993; Willemse et al., 1996; Peacock, 2002). Therefore, to estimate the average displacement along a given fault, the maximum displacement must be corrected by a value, k, depending on the shape of the displacement distribution along the fault trace. Fault formation under the assumption of linear elastic fracture mechanics behavior (Pollard and Fletcher, 2005) is expected to produce an ideal elliptical displacement profile that would result in a k of 0.7854 (Schultz et al., 2010).

The brittle horizontal normal strain (en) represented by a normal fault population (number of faults = N) may be estimated by summing the extensional strain accommodated along all faults, and is given by a modified version of the Kostrov summation (Kostrov, 1974; Scholz and Cowie, 1990; Nahm and Schultz, 2011): 
graphic
where d is the fault dip (assumed to be optimally 60° for normal faults), V is the volume of the faulted region, L is the fault length, H is the fault depth extent, and D is the average displacement (k × Dmax). Studies investigating fault population strain on Earth’s surface (Gupta and Scholz, 2000) and seafloor (Cowie et al., 1993), and other planetary bodies (e.g., Mars; Nahm and Schultz, 2011), make the simplifying assumption that H = L. The assumptions made by these studies therefore require a downward-propagating fault model, which is consistent with observations from many studies at magmatic rift settings (e.g., Ethiopia and Iceland; Gudmundsson and Bäckström, 1991; Forslund and Gudmundsson, 1992; Gudmundsson, 1992; Acocella et al., 2003, 2008; Trippanera et al., 2015). For example, field observations in Tertiary fault systems in Iceland show that the greatest frequency of faults occurs in the upper few kilometers of the crust, suggesting that these faults nucleated at relatively shallow depths (∼1 km) and propagated downward (Forslund and Gudmundsson, 1992; Gudmundsson, 1992). Downward fault propagation in magmatic rifts may also be promoted where fault activation is assisted in part by crustal flexural (e.g., bending of the border fault hanging wall or the flexural margins of subsiding crust; Vening Meinesz, 1950; Lavier et al., 2000; Acocella et al., 2008; Corti et al., 2015), as flexure produces the highest extensional stresses at the near surface (Turcotte and Schubert, 1982).
Once L, and hence H, is equal to the depth of the brittle crust, faults can only grow by increasing L while H remains constant (Cowie et al., 1993). As a result, horizontal extensional strains may be estimated from two separate equations for small (L < brittle crust thickness) and large (L > brittle crust thickness) faults from the method of Cowie et al. (1993). For small faults, the brittle horizontal normal strain can be calculated using: 
graphic
whereas the brittle horizontal normal strain for large faults may be calculated from the equation: 
graphic
where A is the area of the faulted region.

The equations and scaling relations described here were used to estimate horizontal normal strain within fault populations of the Magadi and Natron basins. Brittle crust thickness estimates are from Albaric et al. (2009) for central Magadi and the north Natron–south Magadi border (11 and 19 km, respectively).

Measuring Fault Lengths, Throws, and Slip Rates from Remote Sensing and Field Data

Broad-scale analyses of fault populations from remote-sensing data are prone to both natural and detection biases (Soliva and Schultz, 2008; Schultz et al., 2010). Natural biases are a consequence of processes such as erosion and basin infilling, whereas detection biases relate to the resolution of the data acquisition method (e.g., aerial or satellite photos, SRTM data). These factors lead to truncation biases, which underestimate the size and number of smaller faults in a given population (Schultz et al., 2010), and cannot resolve the contributions of small tension fractures. The resolution of aerial photos used to measure fault lengths in this study is 0.5 m/pixel; given the estimated throw-to-length ratio in this study of 0.005 (Fig. 6E) and an average ∼20° slope measured for degraded scarps throughout the field (Fig. 4), these images should resolve the hanging-wall and footwall cutoffs of faults with lengths >36 m. However, faults with small throws (e.g., <5 m) are also prone to natural bias because they are easily eroded and buried by sediments. For example, faults with lengths <250 m could not be identified using aerial photography in the Magadi and Natron basins, and only 7 faults were identified with lengths between 250 and 500 m (i.e., faults with estimated throws <∼2.5 m). As a result, it is conservatively assumed in this study that faults with throws <5 m are frequently below the detection limit of the aerial photos due to natural biases, resulting in a fault length truncation bias of 1000 m (Fig. 6C). This truncation bias is consistent with that applied by Soliva and Schultz (2008) for faults in the Afar region of the EAR.

Fault throws could not be measured from aerial photographs; therefore, throw data were remotely acquired from the 30 m SRTM data set. The accuracy of the throw data depends on the estimated vertical precision of this data set for the African continent (7.6 m to 95% confidence; Rodríguez et al., 2005). Based on the precision of the SRTM, it is inferred that throws <10 m cannot be confidently measured (i.e., across faults <∼2000 m in length), however, throws of these smaller can be confidently estimated by assuming a linear relationship between fault length and throw (Schlische et al., 1996). Maximum throws were estimated using SRTM data for 40 intra-rift faults in the Natron and Magadi basins by systematically measuring throw along the lengths of faults (e.g., Figs. 6A, 6B). Comparison of fault throws measured in the SRTM with those measured using a Trimble GPS in the field shows good agreement (Fig. 4). To avoid biases related to erosion and burial by sediments, throws were remotely measured only on well-exposed faults with relatively steep scarps (typically 15°–30° dips on the SRTM) that were not surrounded or covered by extensive rift sediments. However, the precision of throw profile analyses will also be affected by sediments that collect against the fault scarp on the downthrown hanging wall (Cartwright et al., 1995). Although the sedimentary cover is expected to be only a few meters thick in arid, extensional basins (e.g., Volcanic Tablelands and Canyonlands, USA; Dawers and Anders, 1995; Moore and Schultz, 1999), the measured throws likely represent minimum values. To avoid detection biases related to the resolution of the SRTM, faults with throws <10 m were not included in comparative throw-length analyses (e.g., Figs. 6A, 6B, 6E). However, the throws of smaller faults were inferred assuming a linear relationship between fault length and throw (e.g., Cowie et al., 1993; Schultz et al., 2010; Nahm and Schultz, 2011), and then used in the horizontal extensional strain calculations. Throw values were also converted to displacement and heave by assuming an optimal 60° dip for normal faults (Fig. 5). Together, fault lengths and throws along 40 intra-rift faults give a best-fit displacement-length coefficient (i.e., g in Equation 1) of 6.3 ± 0.4 × 10–3 (Fig. 6E), which is in good agreement with faults dissecting Quaternary lavas in rift zones elsewhere, such as Iceland (Gudmundsson, 1992; Angelier et al., 1997), the Kino Sogo fault belt of the Turkana basin, Kenya (Vetel et al., 2005), and the East Pacific Rise (Cowie et al., 1993, 1994; Carbotte and MacDonald, 1994).

Volcanic rocks at the tops of scarps were sampled for 40Ar-39Ar age analyses (Table 1; see Supplemental Materials, Section S51), providing constraints on the ages of major lava units and maximum ages of the measured faults. By comparing faulted lava ages with fault throw values, and converting throw to both displacement and heave, these data can be used to estimate maximum slip rates and extension rates, respectively. As shown in Figure 5, slip rates refer to changes in displacement over time, whereas extension rates relate to changes in heave over time.

RESULTS: FAULT CHARACTERISTICS, LAVA CHRONOLOGY, AND SPRING GEOCHEMISTRY OF THE NATRON AND MAGADI BASINS

General Observations

The 3 Ma Natron basin (Foster et al., 1997) is bounded by an ∼600-m-high border fault escarpment (Natron border fault) along its western margin. Rift-normal topographic profiles north of Gelai volcano in the Natron basin (Fig. 1D) show the lowest elevations (595 m) in the lake-filled basin adjacent to the currently active Natron border fault, where a number of hydrothermal springs expel water into the lake at the base of the border fault scarp. In addition to the main escarpment (bold black line in Fig. 1A), the Natron basin comprises a number of smaller, subsidiary segments with throws of as much as 200 m. Incised gullies and ravines within the footwall of the border fault also expose subsidiary faults with measured throws of as much as 91 m (e.g., sample 10, Table 1). Locally, the border fault exhibits recent fissures displacing Holocene alluvium (Sherrod et al., 2013) (Fig. 3B) and at least one relatively young fault segment (Fig. 3A) offsets tuff (maximum age of 200 ± 5 ka; Table 1) and debris avalanche deposits sourced from Oldoinyo Lengai (Kervyn et al., 2008; Sherrod et al., 2013). This fault exhibits as much as 28.1 ± 1.9 m of throw, implying recent and probably ongoing border fault activity (0.16 mm yr–1 since the late Pleistocene on this subsidiary segment). Tuff deposits with 40Ar/39Ar ages ranging from 91 ± 4 to 230 ± 6 ka are observed draping some subsidiary border fault segments with no accompanying fault displacements; this suggests that not all the observed (subsidiary) fault segments along the border fault have produced surface ruptures since ca. 230 ka.

For this study the intra-rift faults of the Natron basin dissecting Gelai volcano lavas were measured and sampled in the field in the southernmost ∼40 km of the basin. Logistical restrictions prevented access to intra-rift faults in Magadi Trachyte lavas between Lenderut volcano and the northern slopes of Gelai (Fig. 7A; see also Supplemental Fig. S12). Faults analyzed in the field on and around Gelai volcano exhibit typical throws of between 50 and 100 m. Age dating of faulted lavas around Gelai reveals that these faults have been active since ca. 1.2 Ma (Table 1), and ongoing intra-rift fault activity in the Natron basin is suggested by persistent shallow (<10 km) tectonic earthquakes below Gelai, with magnitudes ranging from 1.0 to 4.5 (Lee et al., 2016), as well as a 2007 dike-induced surface rupture (Baer et al., 2008; Calais et al., 2008). This surface rupture produced a small dilational fault scarp (as much as 65 cm throw; Wauthier, 2011; Fig. 3C), which could still be observed ∼7 yr later as a series of aligned pit chains (to 5 m long and 2 m wide; Fig. 3D). Throws documented by Wauthier (2011) were rapidly removed by erosion of the original small scarp in loose surface materials.

The 7 Ma Magadi basin (Crossley, 1979) comprises an ∼1600-m-high border fault escarpment (the Nguruman fault). During field work, there was no identified evidence of recent rupturing, such as recently opened fissures or fault displacements on younger than 250 ka volcanics or sediments, in contrast to the Natron border fault (Figs. 3A, 3B). Based on field observations, discussions with the local residents, and existing hydrothermal spring maps (e.g., Baker, 1958; Eugster, 1970), no springs are present along the Nguruman border fault escarpment of the Magadi basin. Late Pleistocene lake deposits (e.g., Oloronga Beds; Baker, 1958; Behr and Rohricht, 2000) within ∼20 km of the Nguruman escarpment suggest that a paleolake was once situated along the western side of the Magadi basin, forming a northern extension of present-day Lake Natron (Crossley, 1979; see Supplemental Fig. S1 [footnote 2]).

Rift-normal elevation profiles show that the deepest part of the Magadi basin occurs away from the border fault in the rift center (603 m elevation compared to 680 m at the base of the Nguruman escarpment), where faults pervasively dissect 1.37–0.95 Ma trachyte lavas (Fig. 1C). Recent east-southeast–west-northwest extension is indicated by rift-parallel, dilational joint sets in late Pleistocene and Holocene lake sediments in the intra-rift graben and around Lake Magadi (Atmaoui and Hollnack, 2003; Muirhead, 2016). This assertion is further supported by a surface rupturing event during a period of enhanced seismicity in 1998. Focal mechanisms from this event reflect normal faulting in response to a west-northwest–east-southeast–directed extensional stress field, which at the surface produced a small, north-northeast–south-southwest–trending dilational fault scarp (∼10 cm throw; Ibs-von Seht et al., 2001; Atmaoui and Hollnack, 2003) (Fig. 7). Within the intra-rift fault system, numerous springs feed water into perennial Lake Magadi. Much like Lake Natron, this shallow lake (only a few meters deep) has little surface drainage input and a high evaporation rate, yet survives due to the influx of CO2-rich, saline fluids from below (Eugster, 1970; Roberts et al., 1993; Lee et al., 2016). The combination of CO2-rich spring waters, alkali-rich volcanics, saline fluids, and arid conditions create unique environmental conditions necessary to produce economically viable trona and nahcolite deposits in these lakes (Eugster, 1970; Jones et al., 1977; Monnin and Schott, 1984; Jagniecki et al., 2015; see also Supplemental Fig. S1 [footnote 2]).

Geochronology of Volcanic Units and Fault Slip Rates

The timing of fault activity was investigated using new 40Ar/39Ar dating of volcanic units (Table 1; Supplemental Table S13) and by reviewing geological maps and previously published K-Ar and 40Ar/39Ar ages (see Supplemental Fig. S1 [footnote 2]). Volcanic rocks dissected by 17 faults were sampled for 40Ar/39Ar age analyses in the Magadi and Natron basin, including 13 intra-rift faults (6 from Magadi and 7 from Natron) and 4 fault segments that collectively form part of the Natron border fault escarpment. Sampled rocks include basalt, trachyte, and nephelinite lavas, as well as volcanic tuff. All samples were collected along the footwall cutoffs at the tops of fault scarps and away from fault tips to avoid the reduced fault throw in those regions. Determined ages range from 4.05 to 0.10 Ma.

These data provide new constraints on the timing of volcanic events that were previously dated using K-Ar techniques (Isaacs and Curtis, 1974; Baker and Mitchell, 1976; Crossley, 1979; Baker et al., 1988; Foster et al., 1997). In the Natron basin, nephelinite (sample 3, Fig. 7A; Table 1) sampled on the Natron border fault and the eastern flank of Mosonik volcano (Supplemental Fig. S1 [see footnote 2]) gives an age of 4.06 Ma, suggesting this volcano began erupting ∼0.5 m.y. earlier than previously estimated (Dawson, 1992; Foster et al., 1997). Basalts from Oldonyo Sambu volcano (Supplemental Fig. S1 [see footnote 2]), sampled at the top of the footwall of the Sangan fault segment of the Natron border fault (sample 7), give an age of 1.97 Ma. Although this date is consistent with previous K-Ar studies, which range from 3.5 to 1.8 Ma (Isaacs and Curtis, 1974), the date of 1.8 Ma was originally considered anomalous, with the authors favoring an age of 3.5–3.1 Ma for the volcano. This new 40Ar/39Ar age, however, suggests that Oldonyo Sambu was active until ca. 1.97 Ma. Sekenge crater rim pyroclastic deposits were also sampled (Neukirchen et al., 2010; Sherrod et al., 2013; sample 2, Fig. 7A). These deposits are displaced by a subsidiary fault segment in the footwall of the Natron border fault escarpment. They contain rare basaltic breadcrust bombs (as much as 30 cm diameter), which give a maximum age of 1.25 Ma. Tuff in close proximity to Oldoinyo Lengai (samples 9, 11, and 12) range in age from 0.23 to 0.10 Ma. Some of these deposits are offset by small segments of the Natron border fault in the hanging wall of the main escarpment. Although these deposits cannot be correlated to Oldoinyo Lengai with complete certainty, the dates are within the range of published ages for the volcano (minimum age of 350 ka to present; Sherrod et al., 2013). Faulted lavas in the intra-rift portion of the Natron valley were derived from Gelai volcano. The new ages presented here demonstrate a longer duration of activity for Gelai (minimum of 230 k.y.), with ages in the range of 1.19–0.96 Ma (samples 1, 4, 5, 6, and 8), compared to 1.19–1.11 Ma reported in Mana et al. (2015).

In the Magadi basin, age data suggest that the Singaraini basalt was emplaced by 1.96 Ma (sample 15; Fig. 7A; Table 1), rather than the 2.31 Ma originally proposed by Baker and Mitchell (1976) based on K-Ar dating. This study also provides the first published 40Ar/39Ar ages for Magadi Trachyte lavas (samples 13, 14, 16, 17), a recent and widespread lava unit observed along the center of the Natron and Magadi basins. The new ages of 1.37–0.95 Ma are within previously documented K-Ar ages ranging from 1.4 to 0.8 Ma (Baker and Mitchell, 1976; Baker et al., 1988). However, only the youngest lavas were sampled, exposed at the tops of fault scarps; therefore, these data only reflect the ages of lavas emplaced during the later stages of the Magadi Trachyte volcanic event. A new date of 2.05 Ma for Limiru trachyte (sample 18) is consistent with the 2.01–1.88 Ma age of Baker and Mitchell (1976).

Age dating of faulted lavas at the tops of fault scarps provides the maximum ages of sampled faults. When combined with measurements of fault throw, the dates also constrain time-averaged minimum slip rates (Fig. 5). Minimum slip rates (Table 1) for intra-rift faults in the Natron basin range from 0.06 to 0.09 mm yr–1. Subsidiary border fault segments in Natron exhibit higher minimum slip rates, ranging from 0.09 to 0.16 mm yr–1. Time-averaged minimum slip rates estimated for sampled intra-rift faults of the Magadi basin range from 0.01 to 0.13 mm yr–1, but are generally higher than slip rates in equivalent-age lavas in the Natron basin. However, these data (Table 1) do not include the largest fault throws (as much as 362 m) measured from SRTM data in Magadi Trachyte lavas, which would suggest that intra-rift faults have maximum slip rates ranging 0.30–0.45 mm yr–1 in the 0.95–1.37 Ma Magadi Trachyte.

Fault Statistics of Intra-Rift Fault Populations

We identified 1474 intra-rift faults in the Natron and Magadi basins. The mean trend of intra-rift faults (008°) roughly parallels the border fault system (006° mean trend) (Figs. 7B, 7C). Intra-rift fault lengths range over 3 orders of magnitude, from 0.26 to 60.7 km (Figs. 6C, 6D). Fault throws were measured along 40 well-exposed intra-rift faults. Throw profiles are typically elliptical or bell-shaped and consistent with unrestricted fault growth, where fault tips grow unimpeded during each slip event (Cowie and Scholz, 1992; Cartwright et al., 1995; Nicol et al., 1996) (Fig. 6A). Throw profiles along some faults in the Natron and Magadi basins also exhibit varying degrees of asymmetry (Kattenhorn et al., 2013) (Fig. 6B). Asymmetric throw profiles are indicative of restricted fault growth, where propagation at one end of the fault is stymied by a mechanical barrier (Nicol et al., 1996; Manighetti et al., 2001). Maximum throws are as high as 362 m within the Magadi Trachyte in the Magadi basin (Fig. 6E). The displacement-length coefficient of 6.3 ± 0.4 × 10–3 is low compared to normal fault populations in softer (typically sedimentary) rocks (e.g., 10–1 to 10–2; Schultz et al., 2006), but is comparable to faults dissecting Quaternary lavas in rift zones elsewhere (Cowie et al., 1993, 1994; Carbotte and MacDonald, 1994; Angelier et al., 1997; Vetel et al., 2005).

Fault-Related Extensional Strain

Faults were subdivided into populations based on the ages of the volcanic units they dissect (Table 2; Supplemental Fig. S1 [see footnote 2]). Ages were determined from the new and existing 40Ar/39Ar analyses. In instances where no 40Ar/39Ar dates were available, existing K-Ar ages were applied. Total fault-related strain was then calculated within each lava unit.

Extension Rates in Lava Groups

We identify 14 major volcanic units in this study in order to subdivide the fault population. These units give maximum ages for the faults dissecting them, given that the time interval between lava eruption and subsequent faulting cannot be determined. The ages of faults can then be translated into a time-averaged extension rate since the eruption of each respective unit based on the calculated horizontal extensional strain within the fault population (Fig. 5; Table 2).

Extensional strain was calculated in each of the fault populations using the method described in this study. The time-averaged extension rate was then estimated by dividing total extension by the age of the lava unit (Table 2). The mean time-averaged extension rates are as high as 0.93 ± 0.28 mm yr–1 (unit 3; Table 2) for intra-rift fault populations in the 7 Ma Magadi basin. Estimated extension rates are lower in intra-rift fault populations in the 3 Ma Natron basin (as much as 0.47 ± 0.14 mm yr–1), consistent with the generally lower slip rates along intra-rift faults in the Natron basin compared to the Magadi basin. These extension rates are within the modeled rates (i.e., less than) of 1.8 and 2.0 mm yr–1 for the Natron and Magadi basins, respectively (Saria et al., 2014; see Supplemental Materials, Section S3 [footnote 1]). The highest extension rates occur in the center of each intra-rift fault population, decreasing to 0.20–0.30 mm yr–1 in the eastern ∼20 km of the intra-rift fault population in the Singaraini basalt and Limiru trachyte lavas. In each basin, extension rates are lowest around the extinct Olorgesailie, Eyaseti, Lenderut, and Shompole volcanoes (<0.03 mm yr–1). Away from these volcanoes, extension rates are lowest (< 0.1 mm yr–1) in the western 10–20 km of the intra-rift fault populations. For example, late Pleistocene lake sediments in southwest Magadi basin are relatively unfaulted, suggesting little to no recent fault activity. In the Natron basin, however, Holocene sediments from ephemeral Lake Natron possibly mask the surface expression of recent faults and thus prohibit a detailed analysis of strain on the western side of the Natron basin (discussed further in the following section). Recently observed seismicity implies active normal faulting below the lake (Lee et al., 2016). However, displacements on these faults are too low to produce scarps high enough to rise above the shallow Lake Natron waters (a few meters deep).

Extension Rates across the Magadi and Natron Basins

To understand how strain is distributed within intra-rift faults across-axis fault-related extensional strain was analyzed along two rift-normal transects through the Magadi and Natron basins (Fig. 8). Rift-normal extensional strain was calculated in 15 rectangular areas, 4 km wide by 15 km long, with long axes trending parallel to rift faults. Results of these analyses, presented in Figures 8B and 8D, show how intra-rift fault-strain is distributed across each basin. Similar to results shown in Table 2, analyzed faults in the Magadi basin accommodate more extensional strain (as much as 6.8%) than those in Natron (as much as 4.0%). Within the center of the Magadi basin, total extension exceeds 3% in 10 of the 15 analyzed areas (Fig. 8B). In contrast, only 2 of the 15 analyzed areas in Natron exhibit extension in excess of 3% (Fig. 8D).

Time-averaged extension rates were estimated by dividing the total extension by the age of the lavas displaced by faults (Fig. 8). By summing values across each transect from west to east in Figures 8C and 8E, the total time-averaged extension rates in the intra-rift populations were calculated. Results show that intra-rift faults in the 7 Ma Magadi basin accommodate more extension (1.63 ± 0.10 mm yr–1) than those in the younger, 3 Ma Natron basin (0.50 ± 0.03 mm yr–1). The remaining extension for each basin is likely focused along border faults.

It is notable that some intra-rift faults in transect Y-Y’ across the Natron basin may be concealed by Holocene lake deposits within ∼16 km of the Natron border fault (Fig. 8A). These complications provide additional uncertainties in the presented extension rate estimates. Given that intra-rift faults in Magadi clearly accommodate more strain than those in Natron (Fig. 8B and D), extension rates within 16 km of the Nguruman border fault can perhaps provide an upper limit for intra-rift faults buried by Lake Natron sediments. Analyses show that well-exposed faults in Pleistocene lake sediments and Magadi Trachyte within 16 km of the Nguruman border fault in the Magadi basin (transect X-X’; Fig. 8) accommodate 0.36 ± 0.01 mm yr–1. Assuming that this is an upper limit for faults below Lake Natron, it is estimated that intra-rift faults in the Natron basin accommodate between 0.47 and 0.90 mm yr–1 of extension.

Estimating Flexure of the Border Fault Hanging Wall

Slip along a planar normal fault in the upper crust can be modeled by assuming that two vertical line loads are applied in equal and opposite directions to an elastic plate either side of the fault plane (i.e., where x = 0 in Fig. 9) (Vening Meinesz, 1950; Turcotte and Schubert, 1982; Stein et al., 1988; Bott, 1996). The footwall and hanging wall of the fault locally respond to this load by bending, with the subsequent deflection profile (i.e., hanging-wall flexural profiles in Fig. 9) dependent on the rigidity of the plate. Fault slip can thus generate extensional bending stresses that manifest at the near-surface in the flexing hanging wall and, in part, can drive the initiation of intra-rift faults (McClay and Ellis, 1987; Lavier et al., 2000; Imber et al., 2003; Ebinger, 2005).

Estimating extensional bending stresses from border fault activity in each basin first requires constraints on subsurface border fault throws that are provided by both gravity and seismic velocity data in the region (Birt et al., 1997; Ebinger et al., 1997). These studies give estimates of rift basin sediment fill of 3.6 km for Magadi and 1.6 km for Natron. Above the ground surface, the Nguruman and Natron escarpments exhibit throws of ∼1.4 km and ∼0.6 km, respectively. Summing values of subaerial and subsurface throws for each basin equates to total border fault throws of 2.2 km for Natron, and 5.0 km for Magadi. Assuming a ratio of upthrow to downthrow on the fault of 0.2 (e.g., Stein and Barrientos, 1985; refer also to Supplemental Materials, Section S2 [footnote 1]), the maximum downward deflection of the hanging wall (i.e., deflection at x = 0 in Fig. 9) is estimated to be 4 km for Magadi and 1.76 km for Natron. These values are within constraints provided by basins depths (Birt et al., 1997; Ebinger et al., 1997), which require minimum hanging-wall deflections of 3.6 km and 1.6 km in Magadi and Natron, respectively.

Based on hanging-wall deflections for each basin and a basin width of 50 km (Foster et al., 1997; Ebinger, 2005), the flexural profile and subsequent horizontal extensional strain can be approximated using the methods of Turcotte and Schubert (1982) and Billings and Kattenhorn (2005) (Fig. 9; refer to Supplemental Materials, Section S2 [footnote 2]). Using effective elastic thicknesses ranging 19–25 km for rift basins in north Tanzania, based on Ebinger et al. (1997), it is estimated that border fault flexure has produced ∼0.8%–1.1% and 1.8%–2.4% of extension in the Natron and Magadi basins, respectively. By making the simplifying assumption that strain was accumulated at a constant rate from border fault activation to the present day, time-averaged extension rates related to hanging-wall flexure range 0.34–0.45 mm yr–1 for Natron and 0.13–0.18 mm yr–1 for Magadi. These results have implications for how much regional extension is accommodated via intra-rift faulting in each basin (see DISCUSSION).

Distribution of Springs and Their Geochemistry

Hydrothermal springs are observed at the bases of some fault scarps in the study region, and are often aligned in fault-parallel clusters (Figs. 3E and 10A). In Magadi, springs occur in an ∼11-km-wide zone (rift-normal measurement) in the center of the basin, whereas springs in the Natron basin are present along an ∼23-km-wide zone extending normal to the rift from the border fault eastward across the hanging wall to an antithetic fault bounding the eastern side of Lake Natron. No springs are documented or observed along the Nguruman border fault of the Magadi basin, nor are any springs present in the center of the intra-rift fault population in Natron.

Spring-water samples were collected in the field using Giggenbach bottles, and dissolved gases were analyzed at the Volatiles Laboratory and Center for Stable Isotopes, University of New Mexico (Albuquerque) (methods described in Supplemental Materials, Section S4 [see footnote 1]). Sampled springs exhibit high pH (8.9–10.3) and warm temperatures, from 35.0 to 50.7 °C (Table 3). The N2-He-Ar systematics of spring samples were compared with fluids and gases from springs, gas vents, and fumaroles from the Rungwe Province (southern Tanzania) and Oldoinyo Lengai volcano of the EAR (Fischer et al., 2009; de Moor et al., 2013) (Fig. 10B). On an N2-He-Ar ternary diagram (Fig. 10B), fluids derived from the mantle should plot near the He apex, consistent with He/Ar and N2/Ar ratios of the upper mantle, which are shown to be ∼1 and 80, respectively (Burnard et al., 1997; Marty and Zimmermann, 1999). In contrast, both air and air-saturated water have N2/Ar ratios of 84 and ∼40 (varies slightly depending on temperature), respectively, and low proportions of He relative to Ar (<0.01) (de Moor et al., 2013). Consistent with magmatic gases and fluids from the Rungwe Province and Oldoinyo Lengai volcano, spring samples from the Natron and Magadi basins plot on a mixing line between air-saturated water and a mantle and/or crust end member (Fig. 10B). Values of δ13C (–3.1‰ to –5.7‰; Table 3) from CO2 gases dissolved in spring samples overlap previous values from Kenya Rift volcanoes (–1.7‰ to –6.9‰), which were interpreted as derived from the upper mantle (Darling et al., 1995) (Fig. 10C). These data are also consistent with isotope values from diffusely degassing CO2 in the Magadi and Natron basins that show a mixing relationship between air and mantle-derived CO2 released from degassing magmas (Lee et al., 2016). Plots of δ13C-CO2 and CO2 concentrations of springs in the region also fit on a mixing line between air and magmatic CO2 (Darling et al. 1995; Sano and Marty, 1995) (Fig. 10C).

DISCUSSION

Data presented in this study are critical for investigating the development of rift basin architecture during continental rifting in the EAR, particularly the magmatic-tectonic processes that influence faulting in the first 10 m.y. of rift development (Figs. 2A, 2B). Results from this study provide insights into how and when strain migrates to the rift center prior to the development of magmatic segments, like those observed today in the Main Ethiopian Rift (Keranen et al., 2004; Fig. 2C). The evolution of fault systems in early-stage basins is discussed here by comparing the 7 Ma Magadi and 3 Ma Natron basins. Specific topics of interest are: (1) the volcanic-tectonic history of the Natron-Magadi region; (2) the importance of flexure of the border fault hanging wall for intra-rift faulting; (3) the amount of strain accommodated between border faults and intra-rift faults in each basin; (4) the evolving fault strain and basin morphology during early-stage rift development; and (5) the processes assisting the evolution of fault systems (e.g., kinematic, magmatic, hydrothermal).

Fault Evolution during Early-Stage Continental Rifting

Tectonic-Magmatic History of the Magadi and Natron Basins

Data presented in this study in context with previous work reveal that rift activity in the Natron and Magadi basins is characterized by persistent faulting and punctuated phases of volcanic activity (Baker and Mitchell, 1976; Crossley, 1979; Le Gall et al., 2008; Guth, 2015) (see Fig. 7; Supplemental Fig. S1 [footnote 2]). The earliest stages of rifting in the Natron-Magadi region were characterized by faulting and volcanism along the Nguruman escarpment of the Magadi basin. Faulting initiated on the rift border ca. 7 Ma, prior to the eruption of Lengitoto trachyte lavas (6.9–5.0 Ma) along the western rift edge (Crossley, 1979). Volcanism began on the eastern border of Magadi ca. 6.7 Ma with the eruption of Ol Eyaseti volcanics (Baker et al., 1971; Baker and Mitchell, 1976). The next major phase of volcanism involved eruption of Kirikiti lavas along the western side of the basin and the rift center from 5.1 to 2.5 Ma (Baker et al., 1971; Crossley, 1979). During this period, volcanism in the Natron region initiated with the development of Mosinik volcano at 4.05 Ma (Table 1), followed by faulting along the now-extinct Oldonyo Ogol border fault escarpment ca. 3 Ma (Foster et al., 1997; Le Gall et al., 2008; Stollhofen and Stanistreet, 2012). The Ngong-Turoka fault, which bounds the eastern side of Magadi basin, was active from 3.3 to 2.3 Ma (Baker and Mitchell, 1976), as were the Lenderut (2.7–2.5 Ma), Olorgesaile (2.7–2.2 Ma), and Ngong (2.6–2.5 Ma) composite volcanoes and related volcanics (Baker et al., 1971; Fairhead et al., 1972, 1977; Baker and Mitchell, 1976). Since 2 Ma, the Magadi and Natron basins have been subject to widespread, basin-filling lava eruptions, including the Limiru trachytes (2.05 Ma) and Singaraini basalts (1.96 Ma), with additional volcanism from Oldonyo Sambu (possibly 3.5 Ma to 1.8 Ma) and Shompole volcanoes [1.96 Ma–present(?)] (Fairhead et al., 1972; Isaacs and Curtis, 1974; Le Gall et al., 2008; Table 1). The final phase of tectonic-magmatic activity in the region was marked by a shift in border faulting in Natron from the Oldonyo Ogol escarpment to the Natron border fault between 1.2 and 1.0 Ma (Foster et al., 1997). Changes in the location of border faulting in Natron were both accompanied and preceded by voluminous axial volcanism in the rift center (combined minimum volume = 1263 km3; Guth, 2015), with eruption of Gelai volcanics (1.23–0.95 Ma) and Magadi Trachyte lavas (1.4–0.95 Ma) (Baker and Mitchell, 1976; Table 1). This volcanic resurfacing event filled the entire rift valley floor, erasing earlier faulted topography (Baker and Mitchell, 1976). Since that time, both rift basins have been characterized by an extended phase (ca. 1 Ma to present) of rift-parallel, intra-rift faulting (Baker, 1958; Crossley, 1979; Le Gall et al., 2008; Fig. 7), whereas volcanism has been largely absent (Muirhead et al., 2015). Intra-rift faults dissecting rift valley–filling lavas provide important information on the upper crustal extensional strain accommodated within the rift center since these final phases of volcanic activity (Fig. 8).

Intra-Rift Faulting in the Natron and Magadi Basins: Regional Extension versus Hanging-Wall Flexure

During early-stage rifting in the Eastern rift, strain was primarily accommodated through slip along border faults (Hayward and Ebinger, 1996), which accumulated displacements of thousands of meters over the lifetime of rifting. Large-scale fault slip (e.g., >1000 m) along border faults and subsequent flexure of the downthrown hanging wall generate extensional stresses in half-graben basins (Vening Meinesz, 1950; McClay and Ellis, 1987; Lavier et al., 2000; Imber et al., 2003; Cowie et al., 2005) (Fig. 9) that are thought to activate intra-rift faults in the border fault hanging wall (Ebinger, 2005) (Fig. 9). This style of faulting accommodates only localized bending strains, rather than regional extension (i.e., far-field extensional plate motions). The extensional strain measured within an intra-rift fault system (Fig. 8) represents strain accommodated by both hanging-wall flexure and regional extension. Therefore, hanging-wall flexural strains (Fig. 9) must be removed from the total intra-rift strain estimates to reveal the amount of regional extension accommodated via intra-rift faulting (Table 4).

Transects X-X’ and Y-Y’ in Figure 8 show the distribution of strain across intra-rift fault populations in the Magadi and Natron basins, as well as the cumulative time-averaged extension rate. Subtracting extension rates related to hanging-wall flexure (Fig. 9) from these estimates provides constraints on the amount of regional extension accommodated in the intra-rift fault systems (Table 4). Calculations from this study suggest that hanging-wall flexure in the Magadi basin has produced 0.13–0.18 mm yr–1 of extension. This hanging-wall flexure accounts for only a small fraction (<12%) of extension observed in this fault system (1.52–1.73 mm yr–1). Based on these estimates, we infer that intra-rift faults in Magadi accommodate 1.34–1.60 mm yr–1 of regional extension. Hanging-wall flexure in the Natron basin, however, can explain a significant portion of observed strain in the intra-rift fault systems (0.34–0.45 mm yr–1), accounting for between 38% and 96% of the 0.47–0.90 mm yr–1 time-averaged extension represented by intra-rift faults (Table 4). These results imply that intra-rift faults in the Natron basin accommodate 0.02–0.56 mm yr–1 of regional extension.

It is important to note that our results cannot resolve the proportion of extensional strain accommodated by dike intrusion, rather than normal faulting, at depth. Field, numerical, analog, geodetic, and geophysical studies illustrate a feedback between middle to upper crustal diking and normal faulting at the near surface, where static stress changes above the dike tip often initiate faulting and graben formation (e.g., Bjornsson et al., 1977; Brandsdóttir and Einarsson, 1979; Mastin and Pollard 1988; Rubin and Pollard, 1988; Chadwick and Embley, 1998; Buck et al., 2005, 2006; Behn et al., 2006; Keir et al., 2006; Wright et al., 2006, 2012; Rowland et al., 2007; Behn and Ito, 2008; Calais et al., 2008; Grandin et al., 2009; Ebinger et al., 2010; Belachew et al., 2013; Sigmundsson et al., 2015; Trippanera et al., 2015; Ruch et al., 2016). Based on these studies, extensional strain measured at the surface in the Natron and Magadi basins is expected to mirror that accommodated below the intra-rift fault system, regardless of what portion is accommodated through either dike intrusion or normal faulting in the upper crust.

Furthermore, length statistics in the Natron-Magadi fault population may have been affected by dike-fault interactions (i.e., Rowland et al., 2007). The low displacement-length coefficient (∼6.0 × 10–3) recorded in the Natron-Magadi region, as well as magmatic rifts elsewhere (e.g., Iceland, East Pacific Rise; Cowie et al., 1993; Angelier et al., 1997), suggest that these faults grow in length rapidly (Vetel et al., 2005) compared to faults in magma-poor extensional systems (e.g., displacement-length coefficients range 0.8–5.2 × 10–2 at Canyonlands and Volcanic Tablelands; Dawers et al., 1993; Cartwright et al., 1995; Dawers and Anders, 1995; Moore and Schultz, 1999). It is possible that these differences in displacement-length ratios reflect, in addition to varying mechanical strength of the host rock (Schultz et al., 2010), the effects of dike-imposed stresses along the length of the rift (i.e., Rubin and Pollard, 1988). In support of these assertions, the surface ruptures of dike-induced normal faults recorded in Afar are shown to have attained remarkable lengths (∼60 km) when considering their largest earthquake magnitudes (moment magnitude, Mw 5.6; Rowland et al., 2007; Grandin et al., 2009; Ebinger et al., 2010).

Border Fault versus Intra-Rift Faulting

Fault system analyses presented in this study do not provide direct constraints on border fault extension rates. However, in early-stage rift basins the extension accommodated by fault systems (border faults and intra-rift faults) should equal the far-field extension rate. Therefore, by assuming that modeled plate motion vectors for the EAR (i.e., Saria et al., 2014) accurately reflect extension rates over the past few million years, and assuming that all upper crustal strain is accommodated along faults observed in the 50-km-wide basins, the maximum border fault extension rates can be estimated by subtracting the plate motion vectors by the calculated time-averaged extension in the intra-rift fault populations (Table 4; see Supplemental Materials, Section S3 [footnote 1]). Geodetic models (i.e., Saria et al., 2014) make the simplifying assumption that strain is confined to a rigid plate boundary (i.e., within the ∼50-km-wide rift basin). With these limitations in mind, our results may also be tested and compared with time-averaged border fault extension rates inferred from basin depth estimates (see following).

Subtracting 1.34–1.60 mm yr–1 of regional extension measured across the Magadi intra-rift fault system in this study from the modeled 2.0 mm yr–1 extension rate for the basin, it is estimated here that the Nguruman border fault accommodates 0.40–0.66 mm yr–1 of extension. This is 20%–33% of the regional extension. Similarly, by subtracting the 0.02–0.56 mm yr–1 of regional extension measured across intra-rift faults of the Natron basin from the modeled 1.8 mm yr–1, the Natron border fault is estimated to accommodate 1.24–1.78 mm yr–1 of extension, which is 69%–99% of the regional extension. Therefore, border fault activity is a far greater contributor to regional extension in the 3 Ma Natron basin than in the 7 Ma Magadi basin. This is consistent with field observations of recent activity along subsidiary segments of the Natron border fault (Figs. 3A, 3B).

Border Fault Slip Rates from Subsurface Throw

Border fault extension rate estimates presented in Table 4 may be further tested by comparing these values with estimated extension rates based on the age, throw, and assumed dip of the rift basin border faults (Table 5). Using a range of possible ages for the Natron border fault (1.2–1.0 Ma; Foster et al., 1997) and an assumed fault dip of 50°–60°, consistent with focal plane mechanisms in the region and normal faults in general (Foster et al., 1997; Lambert et al., 2014), the 2.2 km of throw accumulated on the Natron border fault corresponds to an extension rate of 1.06–1.85 mm yr–1. This compares well to the 1.24–1.78 mm yr–1 from the fault strain analysis summarized in Tables 4 and 5. Similarly, a range of reasonable ages for the Nguruman fault (7.1–6.9 Ma; Crossley, 1979) and an assumed fault dip of 50°–60° corresponds to an extension rate of 0.41–0.61 mm yr–1, which also compares favorably to the 0.4–0.66 mm yr–1 from the fault strain analysis (Tables 4 and 5). This simple approach provides comparable results, which lends confidence to the fault-strain analyses in this study (Fig. 11; Table 4).

Extension rates across the Natron border fault, however, do not match results from age dating and fault-throw analyses collected along border fault segments in the field (e.g., samples 2, 3, and 12 in Table 1; Table 5). The border fault segments analyzed in the field in this study are subsidiary structures of the main escarpment (Sherrod et al., 2013). Due to the border fault’s overall size and the substantial basin-fill sediments concealing its vertical dimensions (Ebinger et al., 1997; Sherrod et al., 2013), analyses of extension rates along the primary fault escarpment can only be constrained from subsurface information or indirect techniques, such as the fault-strain analysis applied in this study.

Border fault extension rates shown in Table 5 correspond to a range of possible slip rates that are consistent with rates along major normal faults on Earth. Slip rates are well constrained for normal faults in extensional terrains such as the Apennines (Italy), Taupo Rift (New Zealand), and Basin and Range (United States). Field studies show that normal fault slip rates are typically ∼0.1–4 mm yr–1 (Villamor and Berryman, 2001, 2006; Hetzel and Hampel, 2005; Kent et al., 2005; Hampel et al., 2007; Berryman et al., 2008; Blakeslee and Kattenhorn, 2013; Wilkinson et al., 2015), although they may be as high as tens of millimeters per year during periods of covolcanic slip associated with caldera-forming rhyolite volcanism in magmatic rifts (e.g., the Paeroa fault of the Taupo Rift, New Zealand; Rowland et al., 2010; Downs et al., 2014). Compared to normal faults elsewhere, the Natron border fault exhibits a relatively high slip rate, with estimates from this study ranging from 1.93 to 3.56 mm yr–1. The slip rate of 0.62–1.32 mm yr–1 estimated for the Nguruman fault is typical of normal faults at many extensional settings.

By contrast, the highest slip rates on individual intra-rift faults (as much as 0.45 mm yr–1) in both the Natron and Magadi basins are lower than the surrounding border faults (cf. Tables 1 and 5). These low slip rates are, however, consistent with a rift system where extensional strains are spread over a wide area and accommodated on many faults (e.g., >100).

Evolution of Fault-Related Strain and Structural Relief in the Natron-Magadi Region

We provide the first quantitative analysis of the spatial and temporal distribution of fault-related strain over the course of early-stage rift evolution in the Magadi and Natron basins (Fig. 11). These data illustrate how fault-related strain is distributed within rift basins after ∼3 and 7 m.y. of rifting in the Eastern rift. Accounting for the various sources of uncertainty, there are clear differences in how strain is distributed between Natron and Magadi. Specifically, in the Natron basin, the majority (>69%) of regional extension is accommodated along the border fault, whereas in Magadi the greatest strain is primarily accommodated in the rift center in the intra-rift fault population (>67%). Furthermore, during early-stage rifting in the 3 Ma Natron basin, a significant amount of intra-rift faulting (38%–96%) is driven by hanging-wall flexure of the border fault. After 7 m.y. of rifting, however, flexure of the border fault hanging wall has less influence on the formation of intra-rift faults, accounting for only 8%–12% of the extensional strain (refer to Table 4). Intra-rift fault populations instead accommodate the majority of regional extension as border faults become less active, as illustrated by the Magadi rift basin.

Evolving fault system dynamics in continental rift basins are accompanied by changes in structural relief that affect water and sediment catchments (Scholz et al., 1990; Gawthorpe and Leeder, 2000; Ebinger and Scholz, 2012). For example, zones of maximum subsidence in half-graben basins in the Malawi and Tanganyika rifts of East Africa occur adjacent to border faults, where more than 4 km of synrift sediments have accumulated (Scholz et al., 1990). Similarly, border fault subsidence in the Natron and Magadi region has clearly been a major control on the position of ephemeral lakes (Ebinger and Scholz, 2012). However, the transition to intra-rift faulting in the Magadi basin also appears to have affected the position of Lake Natron and the development of Lake Magadi (Crane, 1981) (Fig. 12). As shown in Figure 1C, the lowest relief in the Magadi basin occurs in the rift center, consistent with the focusing of strain into the intra-rift fault system. In the Natron basin, however, the lowest relief is observed adjacent to the border fault escarpment (Fig. 1D), where the greatest amount of strain is also accommodated. The occurrence of Pleistocene lake sediments from Lake Natron northward along the base of the Nguruman escarpment indicates that paleo–Lake Natron extended further north into the Magadi basin during the Pleistocene (Crossley, 1979). The transition to intra-rift faulting in the Magadi basin and subsequent reduction in border activity may explain the southward retreat of the northern shoreline of Lake Natron, as areas of structural relief evolved in response to changes in fault system dynamics (Fig. 12). However, during warmer and wetter climatic periods in the late Pleistocene, increasing water levels have also occasionally resulted in Lakes Natron and Magadi forming a single, contiguous lake (Roberts et al., 1993). Similar patterns in the distribution of lake systems in the Natron-Magadi region are observed in basins of different ages across south Kenya and north Tanzania (Fig. 13). For example, in the 1 Ma Manyara and 3 Ma Natron basins, lakes occur adjacent to border fault escarpments, whereas in more evolved basins in the Kenya Rift (15–7 Ma) lakes are situated along the rift axis. The positions of these lakes likely reflect the locus of upper crustal strain (rift border versus rift axis) within these basins (Ebinger and Scholz, 2012).

Source of Spring Waters in the Natron and Magadi Basins

Diffuse soil CO2 flux surveys and carbon isotope analyses from the Natron and Magadi basins reveal that rift-wide fault systems in the region act as pathways for the rise of mantle-derived CO2 (Lee et al., 2016). This diffuse CO2 is likely sourced from large magma bodies, which have been detected geophysically in the lower crust and upper mantle (Birt et al., 1997; Roecker et al., 2015; Lee et al., 2016). Is the chemical signature of springs in the Natron and Magadi basins also consistent with a magmatic source?

Spring samples from this study exhibit N2/Ar ratios (15–67) within the range of, or slightly less than, air-saturated water (Fig. 10B). High He contents shown in Figure 10B, however, suggest additional inputs from either mantle or crustal sources. Similar N2-He-Ar abundances have been observed previously in the EAR for springs in the Rungwe Province and fumaroles from Oldoinyo Lengai, which were interpreted to be sourced from magma bodies (Fischer et al., 2009; Barry et al., 2013; de Moor et al., 2013). A magmatic interpretation for the observed N2-He-Ar abundances is also consistent with δ13C-CO2 values in the Natron and Magadi springs (Fig. 10C), and isotope investigations of diffusely degassing CO2 (Lee et al., 2016). In line with this assertion, helium isotope data from springs in Lake Magadi (reported in McNitt et al., 1989) are consistent with significant contributions (∼25%–50%) of mantle-derived fluids. These geochemical data, combined with geophysical studies supporting the presence of lower crustal and upper mantle magma bodies in the region (Birt et al., 1997; Roecker et al., 2015), suggest a magmatic, rather than crustal, source for dissolved gas species in Natron and Magadi springs.

Controls on Fault System Evolution

Fault strain analyses in this study demonstrate that the Magadi and Natron basins are in a transitional stage of continental rift evolution; the locus of strain is actively transitioning from border faults to intra-rift fault populations (e.g., Figs. 2A, 2B). Previous studies in the Main Ethiopian Rift have attributed these transitions to the influence of magmatism, oblique rifting, and crustal inheritance (Chorowicz et al., 1994; Bonini et al., 1997; Kendall et al., 2005; Corti, 2008). We discuss here how each of these processes applies to the Natron and Magadi basins, as well as the role of magmatic fluids.

Crustal Inheritance

Chorowicz et al. (1994) attributed intra-rift faulting in the Main Ethiopian Rift in the late Pleistocene to the reactivation of preexisting Pan-African basement structures. This interpretation was challenged in later studies (e.g., Bonini et al., 1997; Corti, 2009) given the difficulties in explaining why optimally oriented structural fabrics would not be activated immediately during the initial stages of rifting. The influence of crustal inheritance in the Kenya Rift is illustrated by orientations of transfer structures between border faults, as well as the broad geometry of central Kenya Rift basins, which follow major northwest-southeast–striking shear zones mapped in Mozambique basement rocks (Smith and Mosley, 1993; Mosley, 1993; Smith, 1994; Katumwehe et al., 2016; Robertson et al., 2015). These observations support previous assertions that border faulting in the EAR often initiates on preexisting basement fabrics (McConnell, 1972; Daly et al., 1989; Foster et al., 1997; Katumwehe et al., 2015; Laó-Dávila et al., 2015). However, as basins evolve and favorably oriented faults develop orthogonal to the regional extension direction, the earlier formed, unfavorably oriented and/or located faults are eventually abandoned (Ebinger, 2005). Structural investigations in the Kenya Rift show a number of transverse, northwest-southeast–striking intra-rift faults, which probably initiated on preexisting basement fabrics (Le Turdu et al., 1999; Atmaoui and Hollnack, 2003; Kattenhorn et al., 2013). However, the majority of intra-rift faults in the Natron and Magadi basins, as well as intra-rift faults farther north in the Naivasha-Nakuru basin (Muirhead et al., 2015), trend orthogonal to the east-southeast–west-northwest regional extension direction (Fig. 7), and oblique to the northwest-southeast–striking basement structures. These observations suggest that, although some border fault segments and intra-rift faults (e.g., Kordjya fault of the Magadi basin; Le Turdu et al., 1999; Kattenhorn et al., 2013) may have initiated along inherited crustal fabrics (e.g., Smith and Mosley, 1993; Foster et al., 1997), the transition to an intra-rift dominated faulting regime in the study area cannot be attributed to the reactivation of basement weaknesses.

Oblique Rifting

Strain localization into intra-rift faults in the Main Ethiopian Rift has also been proposed to be the result of oblique extension along the border fault system (rift obliquity as great as 45°; Corti, 2008; Agostini et al., 2011b). These assertions are based on the timing and kinematics of faulting observed from analog experiments (Corti, 2008; Agostini et al., 2011b; Corti, 2012), which compare favorably with field and remote-sensing studies in the region (Corti, 2009; Agostini et al., 2011a). Rifting in the Natron and Magadi basins, however, has been inferred to occur primarily under orthogonal extension (Le Gall et al., 2008). Accordingly, intra-rift faults (008° mean trend) and border faults (006° mean trend) trend approximately normal to the 090°–110° extension direction inferred from analyses of fault plane solutions, volcanic vent alignments, neotectonic joints, and GPS plate motion vectors (Ibs-von Seht et al., 2001; Atmaoui and Hollnack, 2003; Brazier et al., 2005; Stamps et al., 2008; Saria et al., 2014). Focal mechanisms for earthquakes in the central parts of the Natron basin (Lambert et al., 2014) show dip-slip motions along north-northeast–trending normal faults. It is therefore unlikely that the evolution to intra-rift faulting in the study area has been assisted by oblique-style rifting along the border system.

Role of Magma

The development of intra-rift fault systems in the Natron and Magadi region occurred with accompanying magma intrusion (Baer et al., 2008; Calais et al., 2008; Biggs et al., 2009; Ebinger et al., 2013) and magmatic volatile release in the rift basins (Lee et al., 2016). An axial gravity anomaly has been traced along the Kenya Rift into the Magadi basin (Baker and Wohlenberg, 1971; Mechie et al., 1997; Ibs-von Seht et al., 2001), and interpreted as lower crustal intrusions. Ibs-von Seht et al. (2001) attributed high seismic velocities in the center of the Magadi basin to a zone of cooled middle to upper crustal dikes, and Birt et al. (1997) observed a low seismic velocity zone in the upper mantle that they interpreted as partial melt below the central rift axis. The presence of magma along the rift center is consistent with asthenospheric upwelling and enhanced melt production below the zone of maximum lithospheric extension in the center of the rift (Ebinger and Casey, 2001; Ebinger, 2005; Rooney, 2010). Furthermore, asthenospheric melts are likely guided (upward) along steep gradients in the lithosphere-asthenosphere boundary produced by thinning and flexing of the lithosphere, resulting in magma migration at depth from the rift margins into the rift center (Rooney et al., 2011, 2014; Keir et al., 2015). Geological, geophysical, and numerical modeling studies suggest that during early-stage rifting, when faulting is concentrated along rift margins, magmatism may also be focused along the margins of the rift (e.g., Bastow et al., 2005, 2008; Corti, 2009; Maccaferri et al., 2014). The rise of magmatic fluids along the western side of the 3 Ma Natron basin, from the border fault outward ∼20 km toward the rift center, supports these previous assertions. In all, data from this study and previous work support the notion that fault-controlled strain plays a key role in controlling along-axis upper mantle segmentation and aiding melt focusing at depth (Bastow et al., 2005, 2008). Magma and related volatiles ascending from melt zones at depth can also be guided along the border fault and into the rift center along intra-rift faults (Casey et al., 2006; Le Corvec et al., 2013), some of which formed from flexure in the border fault hanging wall.

Magma rising into the center of these rift basins is likely to help localize deformation in intra-rift fault populations and drive the transition from border fault to intra-rift fault-dominated strain accommodation (Ebinger and Casey, 2001; Keir et al., 2006; Beutel et al., 2010). For example, dikes emplaced at middle to upper crustal levels enhance extensional stresses in the area above the dike tip (Rubin and Pollard, 1988). These static stress changes can initiate faulting in the shallow crust, as has been demonstrated during dike rifting events in Iceland and Ethiopia (Bjornsson et al., 1977; Rowland et al., 2007; Belachew et al., 2013; Wright et al., 2012; Sigmundsson et al., 2015; Trippanera et al., 2015). Similarly, a diking event in south Natron in 2007 was accompanied by slip along several intra-rift faults, based on seismicity, focal mechanisms, InSAR (interferometric synthetic aperture radar) displacements, and measured surface ruptures (Baer et al., 2008; Calais et al., 2008; Biggs et al., 2009; Wauthier, 2011). However, this event was confined to a volcanic field at the southern end of the Natron basin, and may not represent a rift-wide process of upper crustal diking from the center of the Natron basin northward along the length of Magadi (Muirhead et al., 2015). Despite the presence of underlying magma at the rift center (Birt et al., 1997), there is no evidence (i.e., volcanic cones, lava flows) of significant surface-breaching dike intrusions since the eruption of Magadi Trachyte lavas (Muirhead et al., 2015), suggesting dike arrest at predominantly lower to middle crustal depths, with upper crustal extensional strain (i.e., upper few kilometers) accommodated almost entirely by normal faulting.

Numerical modeling studies show that cooling magma intrusions transfer heat to the surrounding country rock, causing thermally driven variations in crustal rheology (Daniels et al., 2014). Dike intrusions can therefore locally weaken lithosphere and assist in crustal thinning and strain localization in the basin center (Bialas et al., 2010; Daniels et al., 2014). Extensional stresses are also shown numerically to localize around cooled axial intrusions emplaced within continental crust (Beutel et al. 2010), thereby focusing extensional strain into the rift center even during episodes of low magma flux. In the Magadi basin in particular, geophysical and geochemical observations (e.g., Baker and Wohlenberg, 1971; Birt et al., 1997; Mechie et al., 1997; Ibs-von Seht et al., 2001; Lee et al., 2016) suggest that the transition to intra-rift faulting shown in this study (Fig. 11) has been accompanied by axial magma emplacement in the lower crust and upper mantle. These observations provide support for the potential role of magmatism in assisting the transition to focused intra-rift faulting in the study region.

Role of Magmatic Fluids

An important factor controlling lithospheric weakening and the localization of rifting of cratonic lithosphere is the presence of CO2-rich fluids (Geissler et al., 2005; Lindenfeld et al., 2012; Albaric et al., 2014). In continental rift settings exhibiting high fluid flow (e.g., Taupo Volcanic Zone, New Zealand), the magnitude of extensional tectonic stresses required for normal faulting decreases locally in areas exhibiting high pore-fluid pressures (Reyners et al., 2007; Rowland and Simmons, 2012). Similarly, lower crustal earthquakes observed in the EAR may require anomalously high pore-fluid pressures at depth (Seno and Saito, 1994; Lindenfeld et al., 2012), with some of these fluids likely exsolved from lower crustal magma intrusions (Keir et al., 2009; Lee et al., 2016). The rheology of lithosphere, particularly within fault zones, will vary cyclically depending on the presence or absence of fluids (Yardley et al., 2010). Deeply derived magmatic fluids (e.g., CO2, H2O) are redistributed en masse during earthquake slip events (Sibson, 2000; Brodsky et al., 2003; Shelly et al., 2015). Depending on oxygen fugacity, the presence of these magmatic fluids (i.e., CO2 and H2O) may enhance intracrystalline deformation in quartz and other crystalline minerals (Chen et al., 2006; Chernak et al., 2009; Holyoke and Kronenberg, 2013). During periods of seismic quiescence, fluids present in fault zones are gradually consumed during hydration reactions to produce weaker, fine-grained, hydrous minerals (Moore and Rymer, 2007; Yardley et al., 2010).

Geochemical data from spring systems throughout the Natron and Magadi basins (Fig. 10) and recently published seismicity and diffuse soil CO2 data support the release of volatiles into the rift from lower crustal and/or upper mantle magma bodies (Lee et al., 2016). The high output of magmatic gases (>1 Mt yr–1 of magmatic CO2; Lee et al., 2016) along rift-wide fault networks in the Magadi-Natron region suggests a significant supply of deeply circulating fluids that may be able to weaken faults and assist in strain localization. In line with these assertions, the locations of hot springs in the Magadi-Natron region coincide with areas of high strain (Figs. 8 and 10; Table 2). For example, time-averaged extension rates in the Magadi basin are highest within faults dissecting the Magadi Trachyte lava group (0.93 ± 0.28 mm yr–1) in the center of the rift. Within this area of high strain (unit 3 in Table 2), hot springs manifest at the base of intra-rift fault scarps (Fig. 3E), whereas no springs were observed along the Nguruman border fault escarpment. In the Natron basin, however, hydrothermal systems are present along the western border fault of the rift and ∼20 km outward into the border fault hanging wall. In contrast to Magadi, the border fault accommodates significantly more regional extension (1.24–1.78 mm yr–1) than intra-rift faults (0.0.2–0.56 mm yr–1). Therefore, we assert that magmatic fluids in the Natron basin likely ascend along the border fault, and then possibly outward into nearby faults forming in the flexing border fault hanging wall (Fig. 14). As flexure imposes compressional stresses in the lower half of an the elastic plate, and extensional stresses in the upper half, fluid migration along subvertical extensional structures (e.g., normal faults and dilational joints) within the hanging wall is expected to be enhanced in the upper ∼10 km of the crust in Natron. The close associations between areas of high strain and volatile release suggest that magmatic volatiles may also play an active role in weakening lithosphere and assisting the transition to focused intra-rift faulting in the study region.

Conceptual Model: Evolving Fault-Fluid Systems in the Natron and Magadi Basins

By comparing the 3 Ma Natron and 7 Ma Magadi basins we provide a revised conceptual model for evolving fault and magmatic volatile systems during early-stage rifting. Results from the Natron basin show that, within the first 3 m.y. of rifting, border faults accommodate the majority of strain (>69%) related to regional extension. In Natron specifically, the depth of synrift basin fill suggests that the border fault system accrued ∼2200 m of throw during this period, which caused significant flexure of the hanging wall in the ∼50-km-wide half-graben basin. Horizontal extensional strain related to this flexure drove the formation of intra-rift faults, which acted to accommodate regional extension as well as localized stresses in the flexing hanging wall. Evidence for diking in south Natron (i.e., Calais et al., 2008; Biggs et al., 2013; Muirhead et al., 2015), lower crustal seismicity (Roecker et al., 2015; Lee et al., 2016) and rift-wide release of magmatic volatiles at the surface (Lee et al., 2016; Fig. 10) support the development of intra-rift fault systems coevally with the emplacement of crustal magmas. Fluids released from magmas interact with these fault systems (Figs. 10 and 14), and may explain the presence of lower to mid-crustal earthquake swarms documented in these basins (Seno and Saito, 1994; Lee et al., 2016). The focusing of magma toward the rift center and concomitant release of magmatic fluids into the flexing hanging wall provide a previously unrecognized mechanism that may help to weaken crust and assist the transition to intra-rift dominated strain accommodation.

Findings from the Magadi basin show that, by 7 m.y., regional extension was primarily accommodated in the intra-rift fault population. By this stage of rifting, the border fault became relatively less active (accommodating 0.40–0.66 mm yr–1 of extension, compared to 1.24–1.78 mm yr–1 in Natron), and hanging-wall flexure played only a minor role in driving intra-rift faulting. The transition to intra-rift faulting was accompanied by widespread magma emplacement below the rift axis (Baker and Wohlenberg, 1971; Birt et al., 1997; Ibs-von Seht et al., 2001; Lee et al., 2016). Furthermore, the migration of strain away from the border fault into the rift center is followed by a similar migration in the locations of magmatic volatile release. These observations highlight a strong association between strain localization and the release of magmatic fluids in the rift.

Our results provide important insights into the processes governing strain localization into the rift center that may also apply to other parts of the Eastern rift. Competing hypotheses for the development of intra-rift faults in the Main Ethiopian Rift (Fig. 1B) focus on the influence of oblique rifting versus the role magmatism (see Corti, 2009, and references therein; Fig. 2). The Natron and Magadi basins are currently in this transitional phase, and our results support the role of not only magmatism, but also magmatic volatiles in establishing intra-rift fault systems and focusing strain in the rift center. Once this fault system is developed, crustal thinning within the rift center may enhance magma production from decompression melting (Ebinger, 2005; Corti, 2008; Rooney, 2010) and assist in the development of magmatic segments like those present in the Main Ethiopian Rift (Ebinger et al., 2001; Keranen et al., 2004; Maguire et al., 2006; Fig. 2). Magmatic volatile release may have therefore been a key process for weakening lithosphere to localize strain in the rift center prior to the establishment of magmatic segments in Ethiopia.

CONCLUSIONS

This study is the first to investigate the evolving distribution of upper crustal deformation in the 3 Ma Natron and 7 Ma Magadi basins from quantitative fault-strain analyses. These data are examined alongside spring geochemistry and new 40Ar/39Ar ages of rift volcanics to elucidate the timing of volcanic and tectonic activity. Results demonstrate for the first time a clear link between zones of high upper crustal strain and magmatic fluid release within early-stage rift basins (younger than 10 m.y.) in the East African Rift.

New 40Ar/39Ar ages provide revised constraints on the timing of activity in the Natron basin from Mosonik (as old as 4.05 Ma), Gelai (1.23–0.95 Ma), and Oldonyo Sambu volcanoes (as young as 1.97 Ma). Results also suggest that the Singaraini basalts in the Magadi basin were emplaced by 1.96 Ma. Intra-rift faults measured and dated in the field in the Natron basin exhibit minimum slip rates that range from 0.06 to 0.09 mm yr–1, whereas analyzed (subsidiary) border fault segments range from 0.09 to 0.16 mm yr–1. Time-averaged minimum slip rates from intra-rift faults measured in the field in Magadi range from 0.01 to 0.13 mm yr–1, but likely reach as high as 0.45 mm yr–1 on the largest faults. Basin-wide strain analyses give estimated slip rates of 1.93–3.56 mm yr–1 for the Natron border fault and 0.62–1.32 mm yr–1 for the Nguruman border fault, which are in good agreement with estimates based on rift basin depths and border fault ages.

Results from the Natron basin in this study show that, within the first 3 m.y. of rifting, border faults accommodate the majority of strain (>69%) related to regional extension. Flexure of the border fault hanging wall may have been an important driver of the initial formation of intra-rift faults, which ultimately accommodated both regional extension as well as localized stresses in the flexing plate. Within the 7 Ma Magadi basin, regional extension is primarily accommodated in the intra-rift fault population (67%–80%), revealing a transition from border fault- to intra-rift fault–dominated strain accommodation sometime between 3 and 7 m.y. after rift initiation. The transition to intra-rift faulting in the Magadi basin also affected the position of Lake Natron and the development of Lake Magadi, where a reduction in activity along the Nguruman border fault resulted in the southward retreat of the northern shoreline of Lake Natron.

The localization of strain into intra-rift faults in the Magadi basin did not require the reactivation of preexisting structures (Chorowicz, et al., 1994) or oblique-style rifting (Corti, 2008), as has been asserted previously for the Main Ethiopian Rift. Instead, the transition to intra-rift faulting in Magadi has been accompanied by recent magmatism and magmatic volatile release in the rift center. Areas of high strain in each basin also coincide with the presence of hydrothermal spring systems, and geochemical analyses of dissolved gas species support a magmatic source for these fluids. These observations highlight a strong association between strain localization and the introduction of magmatic fluids into the rift, likely sourced from magma bodies in the upper mantle and lower crust. The ascent of magmatic volatiles within rift basins may therefore help to weaken lithosphere and localize strain in the rift center prior to the establishment of magmatic segments.

This study was supported by National Science Foundation grants EAR-1113677 (to Kattenhorn) and EAR-1113066 (to Fischer). Additional support to Muirhead was provided by Fulbright New Zealand and the Ministry of Science and Innovation. We thank the Tanzania Commission for Science and Technology (COSTECH) and the Kenyan National Council for Science and Technology for granting research permits, and Majura Songo and Norbert Opiyo-Akech for assistance. Aerial photographs were provided by the Polar Geospatial Center, University of Minnesota (St. Paul). We are grateful to Remigius Gama, Matt Blakeslee, Benson Onguso, Nicole Thomas, and Melania Maqway for assistance in the field. We thank Cindy Ebinger in particular for valuable discussions on continental rifting, and Eric Mittelstaedt, Catherine Cooper, John Watkinson, Dennis Geist, Alexandria Guth, Christelle Wauthier, Julie Rowland, Pilar Villamor, Matt Pendleton, Darin Schwartz, and Alexa Van Eaton for helpful discussions and comments. We also thank two anonymous reviewers, who provided constructive and insightful comments that improved the manuscript.

1Supplemental Materials. Description of supplemental geological map for the region (Figure S1), as well as methods for calculating flexural strain in a fault hanging wall, estimating the basin-scale extension rate in study region, spring sampling and analysis, and 40Ar/39Ar dating of volcanics. Please visit http://dx.doi.org/10.1130/GES01375.S1 or the full-text article on www.gsapubs.org to view the Supplemental Materials.
2 Supplemental Figure S1. Geological map of the Magadi-Natron region. Please visit http://dx.doi.org/10.1130/GES01375.S2 or the full-text article on www.gsapubs.org to view the Supplemental Figure.
3 Supplemental Table S1. Incremental-heating 40Ar/39Ar data for newly analyzed volcanic rocks in the Natron and Magadi basins. Ages reported in bold and highlighted in yellow are considered most reliable, and take into account the isochron determined 40Ar/36Ar ratio for the “trapped” component. Please visit http://dx.doi.org/10.1130/GES01375.S3 or the full-text article on www.gsapubs.org to view the Supplemental Table.