Seismic images of magmatic rifting beneath the western branch of the East African rift

Authors


Abstract

[1] We have performed a tomographic study using a joint data set that includes local and teleseismic events, recorded by a temporary network in the western branch of the East African rift system. From the travel time residuals, we derive a three-dimensional model of seismic P-wave velocity anomalies for the crust and upper mantle down to a depth of 80 km. Particular attention is paid to the verification of the inversion results by various resolution tests. The results show that the eastern rift shoulder is characterized by relatively high seismic velocities. Lower velocities are obtained beneath the entire length of the rift valley and the Rwenzori Mountains. A prominent feature is observed north-east of the mountain range: here we detected a vertically oriented, cylindrical low-velocity anomaly with maximum amplitudes in the middle crust and the upper mantle lithosphere. We suggest that this anomaly indicates reservoirs of molten material related to the ongoing rifting process within this segment of the rift. Just above this anomaly, at depths between 5 and 16 km, earthquake swarms exist. The observed reduction in P-wave velocity is used to provide constraints on the possible melt content and temperature anomaly in the uppermost mantle. The observed 3–5% P-velocity decrease can be explained by melt fraction up to 2%–3.3% or alternatively by a temperature increase of at least 248 to 376 K and even higher-temperature anomalies are possible if lower ambient temperatures in the reference mantle are assumed. Probably, the two effects act in combination.

1. Introduction

[2] The East African rift system (EARS) is the largest continental rift complex in the world, where strong extensional processes coexist with volcanic activity. The EARS extends from the Afar Triple Junction (ATJ) in the North to the lower Zambezi River in the South. In its central part, the EARS splits into two branches which surround the Tanzania Craton (TC) along its western and eastern border. The western branch comprises most of the East African Great lakes. The eastern branch, sometimes also called Gregory rift, includes the Main Ethiopian Rift and the Kenyan Rift Valley with Lake Turkana. It splays and peters out south of the North Tanzanian Divergence. In both branches, moderate manifestations of Cenozoic volcanic and thermal activity are observed. The deep structure beneath the EARS has been investigated by a number of studies [Ritsema et al., 1999; Montelli et al., 2004; Benoit et al., 2006a, 2006b; Park and Nyblade, 2006; Koulakov, 2007; Ebinger and Sleep, 1998]. Zeyen et al. [1997] have investigated styles of continental rifting in dependence of stress regime and presence or absence of a mantle plume. They have presented a model in which a large mantle plume is the principal source of the forces which lead to rift formation of eastern and western branches of the EARS. Achauer and Masson [2002] provided an overview of the results of tomographic studies in different continental rift systems and conclude that its development is controlled by a number of contributing factors: namely the rheological strength of the lithosphere, the current stress field, the velocity and the direction of plate motion, and the size and strength of local/regional convection. However, there has been no consensus on structural details such as the number of mantle plumes beneath East Africa. Chang and Van der Lee [2011] give a recent overview on this topic.

[3] The role and significance of magmatic activity and inclusions in active tectonic rift systems is a topic of ongoing debate. A model for melt generation by mantle decompression during uniform pure-shear extension of continental lithosphere at finite rates has been described by Bown and White [1995]. It was shown that melt generation strongly depends on rifting duration, the rate of the rifting, and the initial temperature condition of underlying mantle. In a similar model which includes the effect of melt extraction and redeposition. Schmeling and Wallner [2012] showed that magmatism may strongly affect the thermal structure and thus the strength of the shallow lithosphere. Bialas et al. [2010] using numerical modeling have investigated how the amount of melt influences the speed of rift opening and the maximum lithosphere thickness that can be rifted. They have shown that magma-assisted rifts may transition to tectonic rifts if rapid magma injection occurs over a short period of time (∼1 Ma) or with slow divergence rates over longer time periods (>10–20 Ma). Magmatic activity has previously been investigated in greater detail for the northern and eastern branches of the EARS. Kendall et al. [2005] have presented results of shear wave splitting observation in the Main Ethiopian Rift (MER), which together with recent geological data, indicate a strong component of melt-induced anisotropy with only minor crustal stretching, supporting the magma-assisted rifting model in this area of initially cold, thick continental lithosphere. The analysis of the lower-crustal earthquakes within the MER together with information on seismic structure of the crust and upper mantle, electromagnetic properties of the crust, rock geochemistry, and geological data, indicates that lower-crustal earthquakes are focused in mafic lower crust containing pockets of large fractions of partial melt [Keir et al., 2009]. Recently, Rychert et al. [2012] from the S-to-P receiver functions analysis have imaged the lithosphere—asthenosphere boundary beneath the Afar and the MER. Using geodynamic modeling they have shown that decompression melting of the mantle in the absence of a strong thermal plume is the best explanation of the increased velocities in the upper mantle beneath the rift. From the analysis of seismicity depth distribution, Albaric et al. [2009] have concluded that different parts of the EARS are characterized by significant variations in rheology. Especially, the crust in the western branch of the EARS appears to become stronger from north to south. Surface-wave studies in the MER [Bastow et al., 2010] further indicate that crustal magmatic inclusions exhibit a low aspect ratio (i.e., dykes and veins). These results suggest that the MER is an active magma-assisted rift and confirm the importance of magma intrusion during breakup. For other segments of the EARS, the role of magmatic inclusions has not yet been studied in detail.

[4] Several local-scale tomographic studies have focused on separate segments of the EARS and resolved the crustal structure of the MER [Daly et al., 2008], Lake Bogoria region in Kenya [Tongue, 1992], and upper mantle structure beneath Kenya rift [Park and Nyblade, 2006]. Jakovlev et al. [2011] have presented velocity structure beneath the Rwenzori Mountains obtained from local travel time tomography. This mountain range reaches the maximum altitude of more than 5000 m and, in contrast to other large mountains associated with the EARS, is located within the rift valley. They are composed of Precambrian rocks [Maasha, 1975; Link et al., 2010]. The nature of their extreme elevation is currently under debate. Recently, it was suggested that the uplift of the Rwenzori range is caused by a starting delamination of the lithosphere between two neighboring rift segments [Wallner and Schmeling, 2010, 2011]. Several additional studies on the structure and dynamic processes in the Rwenzori Mountain region were published recently: from the analysis of local seismicity, Lindenfeld et al. [2012] have found an indication that in this region is in the initial stage of rift segment development which may eventually lead to the complete detachment of the Rwenzori block from the surrounding rift flanks in the future as proposed by Koehn et al. [2008] from numerical modeling. From the results of the S receiver functions, Wölbern et al. [2012] have presented evidence for melt infiltration forming a midlithospheric discontinuity within cratonic lithosphere underlain by anomalously hot mantle.

[5] In this paper, we extend the investigation of the Rwenzori segment of the EARS by imaging the seismic structure beneath the rift. A previous tomographic study by Jakovlev et al. [2011] based on local earthquake data provided the seismic structure of only the upper crust. This model does not give much information on the major mechanism of rifting which are thought to be located at a depth that cannot be resolved by local earthquake tomography. To explore the crust and the uppermost mantle beneath the Rwenzori area, we perform a joint analysis of travel time data from local and teleseismic earthquakes, which allows for resolving seismic structures down to 80 km depth. The analysis of the obtained results provides constraints on the possible amounts of molten material and temperatures within the upper mantle beneath the rift.

2. Data and Methods

2.1. Data Set

[6] The data used in this study were collected by a temporary seismological network installed in the framework of the collaborative RiftLink Project (www.riftlink.org). The network consisted of up to 29 seismic stations covering an area of approximately 140 × 80 km2 (see the station distribution in Figure 1, left). The network was in operation from February 2006 to September 2007. Further details on the seismic network are given in Jakovlev et al. [2011] and Lindenfeld et al. [2012].

Figure 1.

(left) Overview map of the study region. Triangles show distribution of the network stations. Black dots indicate local earthquakes used for the tomographic inversion. Black solid lines denote the 1000 and 2500 m contours, 1000 m contour is thick. The thick black dashed-dotted line marks the Uganda/DR Congo border. Thick brown dashed line in the inset shows rough position of the EARS. (right) Location of the teleseismic events used for the tomographic inversion.

[7] For the tomographic inversion, we use a data set obtained by combining the local and teleseismic data. The local data set contains 9646 P-wave and 9536 S-wave travel times from 537 local events (Figure 1, left). Teleseismic data set includes 3341 P-wave travel times from 294 teleseismic and regional events (at an epicentral distance larger than 30° for teleseismic phases and between 5° and 30° for regional phases, respectively; see Figure 1, right). As the proper picking of S-wave arrival times is more difficult for teleseismic and regional data, we only included P-wave picks in the corresponding data subsets. For the local events, both P and S-wave picks were included, as the latter provide important constraints on earthquake location. However, here we present and discuss the P-velocities; the S-velocity model is only available for the upper crust and it is not significantly different of that presented in Jakovlev et al. [2011].

[8] A trigger algorithm was applied to identify and extract the numerous local events from the continuous data stream. However, all travel time picks were checked by visual inspection. The seismic analysis (SEISAN) software package [Havskov and Ottemoller, 1999] was used for the local earthquake data analysis and preliminary event location [Lindenfeld et al., 2012]. The local events picking errors are estimated to be less than 0.05 s for P-waves and 0.1 s for S-waves.

[9] While a total of 9812 local events are identified during the operating period, we limited the events used for the inversion process according to the following two criteria: (i) an event should be registered by more than 15 stations, and (ii) the distance from the event to the nearest station should less than 40 km. The number of the station in the first condition was chosen so that amount of the local and teleseismic events included in the inversion were comparable. The later condition is more liberal then the often-applied azimuthal gap of less than 180° criterion (see examples in Thurber et al. [1995]; Husen and Smith [2004]; and many others). This means that events outside of the station network are also included in the inversion, as the corresponding raypaths can provide important information on the velocity structure [Koulakov, 2009b]. The possible effects of location uncertainty on the estimation of crustal velocity structure in the Rwenzori region have been investigated in great detail in an earlier publication [Jakovlev et al., 2011]. Thus, in total 9646 P phases coming from the 537 local events were included to the local data subset.

[10] The seismic network includes instruments of different types, in particular short period sensors Mark L-4C3D and broadband sensors Guralp CMG-3T. Therefore, before the determination of arrival times from teleseismic events we applied the standard Mb filter of the SEISAN program to the seismograms. This filter simulates the response of the classic 1 Hz WWSSN (World Wide Seismographic Station Network) instrument. This allows a relatively precise determination of the differences in arrival times at different stations. For each event, teleseismic travel-time residuals are calculated with respect to the average arrival time at all recording stations, like in standard Aki, Christoffersson, and Husebye (ACH) method [Aki et al., 1977]. Note that the average residual for each event is zero. Finally, 3341 P phases coming from the 294 regional and teleseismic events are included into the second data subset.

2.2. Joint-Inversion Algorithm

[11] For the tomographic inversion of the combined data set, we modified the Local Tomography Software (LOTOS) code [Koulakov, 2009a], which was originally written for seismic tomography based on local earthquakes. The same algorithm for the joint inversion of local, regional, and teleseismic data has recently been used by Bianchi et al. [2013] to study the deep structure beneath the Puna Plateau in Argentina. The inversion procedure starts with the optimization of the 1-D shallow velocity model based on data from local sources with simultaneous determination of their coordinates. During entire inversion procedure the locations of teleseismic and regional events remain fixed, whereas local events are relocated. For the depths below 35 km, we use the global 1-D model AK135 [Kennett et al., 1995].

[12] The following steps are then performed iteratively: (i) local source relocation in the updated 3-D P-velocity and S-velocity models, (ii) tracing of local and teleseismic rays through the entire model, and (iii) tomographic inversion of travel time residuals. The ray tracing is based on a bending algorithm [e.g., Um and Thurber, 1987]. For the ray tracing from local events, we used a version of the bending algorithm developed by Koulakov [2009a], which starts from a straight line and then deforms it to achieve the minimal travel time. For the teleseismic ray tracing, we use the same principle, but we start with the curve corresponding to the raypath in the 1-D model. This curve is bent along its entire length between source and receiver. The spacing of nodes used for bending is denser within the target volume than in the region outside. This enables stronger deformation of raypaths within the study volume, where the 3-D anomalies are computed. The maximum extent of the anomalous region (with denser raypath parameterization nodes spacing) beneath the network is given by a distance of 300 km from the receiving stations. Note that teleseismic data, which contain relative residuals, are not sensitive to absolute velocity values in the reference model. Based on local and teleseismic data, the inversion is performed simultaneously for several groups of parameters: (i) 3-D distribution of P-velocity anomalies, within the crust S-velocity anomalies are also considered [see Jakovlev et al., 2011]; (ii) coordinate and origin time corrections for local events; (iii) time corrections for teleseismic data. The inversion is performed using the LSQR algorithms [Paige and Saunders, 1982]. Smoothness of the obtained model is controlled by additional matrix block which minimize difference in the velocity perturbation values between all pairs of neighboring nodes in the parameterization grid. Similarly as in Koulakov [2009a], we used the parameterization grids constructed according to the distribution of rays (see example of the grid nodes distribution in the supporting information).1 To minimize any grid-related artifacts, we performed the inversions in four grids of different basic orientations and then computed the average. All of the real and synthetic models presented here are obtained after three iteration steps. Further iterations do not improve the data fit significantly.

3. Inversion Results and Verification

[14] The simultaneous use of the local and teleseismic/regional data sets for the tomographic inversion has several advantages. Local and teleseismic schemes cover different depth intervals and thus complement each other. The local scheme provides information only on the uppermost crust which cannot be studied by teleseismic tomography. However, the derived velocity structures in the upper part of the model allow for a correction of the travel times for teleseismic rays which has an important effect upon the entire inversion. The total depth of the resolved area in the case of teleseismic inversion is roughly equal to the value of the network aperture, e.g., ∼80 km in our case, which is not achievable from local earthquake tomography.

[15] The resulting P-velocity anomalies after joint data inversion are presented in Figure 2 in five horizontal sections at depths from 15 to 75 km. This model corresponds to the inversion results after three iterations which enables the reduction of the average travel time residual from 0.593 to 0.287 s.

Figure 2.

P-velocity anomalies in horizontal sections obtained from the real data inversion. A1 and A2 indicate anomalies that are described in the text. The color scale indicates the velocity anomaly as a percentage. The top left subplot shows the local seismicity and position of the vertical section of Figures 3, 5, and 6. For orientation elevation contours are plotted (in black).

[16] Generally, at all depth levels, velocities directly beneath the rift, including the Rwenzori range, are significantly lower. Higher velocities are found beneath the eastern rift shoulder. Velocity anomalies beneath the western rift shoulder cannot be resolved. Results for the horizontal section at 15 km depth agree well with tomographic images obtained from the local study by Jakovlev et al. [2011]. Within the crust, up to a depth of about 35 km, positive and negative velocity anomalies exhibit average values of about 5%. The maximum negative anomalies occur south of the Rwenzori beneath Lake Edward/Lake George segment of the rift. At greater depth, below 35 km, amplitudes of the anomalies are diminished. An isolated low-velocity anomaly (labeled A1 and A2 in Figures 2 and 3, respectively) is observed in the center of the study area (approximately located at 0.55°N and 30.4°E) beneath the region, where the Rwenzori range is connected to the rift shoulder. This anomaly is visible at all depth levels, however, it is least pronounced within the lower crust.

Figure 3.

Tomographic image of the P-velocity anomalies beneath the Rwenzori region (left) at a depth of 15 km and (right) in a vertical section (the position is indicated). A1 and A2 indicate anomalies that are referred to in the text. The color scale indicates the relative velocity anomaly. The thick black line denotes the 1000 m elevation contour. The dashed ellipse marks the area of observed crustal earthquake swarms; the small gray ellipse indicates the epicenters of mantle earthquakes, see section 4.

[17] In Figure 3, we show results of the inversion for a vertical section extending approximately EW across the rift and cutting through the isolated anomaly. From this figure, it appears that the anomalous regions in the crust (A1) and mantle (A2) are separated and that the strength of the lower anomaly (A2) increases with depth. The boundary between the low and high velocities is dipping at an angle of about 65° toward the east. In the western part of this section, we observe a pronounced vertically extending low-velocity anomaly, for which the amplitude increases toward the surface. This structure is located directly beneath the rift center. Its lateral extent toward the west cannot be resolved.

[18] In order to estimate the resolution and robustness of the results in different parts of the study area, we performed a number of synthetic tests. The travel times for the synthetic tests were computed by ray tracing between sources and receivers corresponding to the configuration of the real system. To simulate picking errors, we assume a normal distribution. In our case, we estimate an average picking error of 0.05 and 0.1 s for P and S arrivals, respectively. The positions of local sources are assumed to be unknown and are determined during the inversion. As in the real case, the global 1-D model AK135 [Kennett et al., 1995] is used as a starting model. The values of the inversion parameters used in all synthetic tests are the same as those for the real data inversion.

[19] To evaluate the horizontal resolution, we performed reconstructions of several models with different sizes of synthetic patterns. Figure 4 presents the results for two synthetic models defined for the whole area with anomalies of 25 × 25 and 50 × 50 km2 of lateral size. The velocity anomalies, with amplitudes of ±5%, remain unchanged at all depths. The tomographic reconstructions show that the best horizontal resolution is achieved in the central northern part of the study area, where the density of the rays is greatest. The ability to resolve small-scale features decreases with depth. Within the crust, anomalies of 25 km width (or larger) can be resolved in all parts the study area. Larger features (>50 km) can be resolved down to a depth of 75 km.

Figure 4.

Horizontal checkerboard models and their reconstructions. The top plot shows the model and the results of the tomographic inversion for 25 km wide horizontal velocity anomalies. The width of the anomalies in the bottom plot is 50 km. The color scale indicates the velocity anomaly as a percentage. The thick black line denotes the 1000 m elevation isoline.

[20] It is well known that in the case of teleseismic inversion, the vertical resolution is always much poorer than the horizontal resolution. In order to assess the realistic capacity of the inversion to resolve the vertical patterns, we have performed a synthetic test with a model defined in a vertical section (in the same location as the vertical section presented in Figure 3). The synthetic anomalies are composed of two layers of periodical anomalies with the size of 25 km. The results of reconstruction (Figure 5) show that anomalies (its shape and position) within the upper layer are correctly resolved. At 25 km depth, we observe a clear change of signs; however, the amplitudes of the retrieved anomalies below 25 km depth are significantly lower than in the true model. We cannot detect the lower limit of the second layer at 50 km depth as the anomalies are smeared downward. All these artifacts should to be taken into account when interpreting the results of the real data inversion.

Figure 5.

Vertical checkerboard model and reconstruction. The location of the vertical section is indicated by a line in Figures 2 and 3. The color scale indicates the velocity anomaly as a percentage with respect to the reference model.

[21] We performed additional tests (Figure 6) to show that two negative anomalies (A1 and A2) observed in the middle of the study area (see vertical section in Figure 3) are separate. The tests simulate three possible situations: (1) there is only one isolated anomaly in the upper crust; (2) there is only one isolated anomaly beneath the crust; (3) there are two separate anomalies, one in the upper crust and another in the upper mantle beneath it. The results indicate that the third possibility is more likely.

Figure 6.

Reconstruction results of three synthetic tests. In the top, middle, and bottom plots present reconstruction results of synthetic models which simulate three possible situations: (1) there is only one isolated anomaly in the upper crust, (2) there is only one isolated anomaly beneath the crust, and (3) there are two separate anomalies, one in the upper crust and another in the upper mantle beneath it, respectively. The location of the vertical section is indicated by a line in Figures 2 and 3. The color scale indicates the velocity anomaly as a percentage with respect to the reference model.

[22] More details about the tests can be found in the supporting information.

4. Discussion and Conclusions

[23] From the results of the joint tomographic inversion, we obtained images of the P-wave velocity structure in the crust and upper mantle beneath the Rwenzori Mountains and the eastern rift shoulder. The latter is characterized by relatively high seismic velocities. Lower velocities are observed beneath the entire length of the rift valley and the Rwenzori Mountains.

[24] Before combining the new and previous results into a tectonic model, we quantitatively address the question of possible anomalous temperatures and melt fractions associated with the low-velocity anomalies. First, we estimate the possible anomalous temperature at upper mantle conditions (50 km depth) responsible for a 3–5% Vp-decrease in the absence of melt to test whether the solidus temperature may be exceeded. We use the formalism discussed in Kreutzmann et al. [2004] which accounts for the pressure-dependent and temperature-dependent anharmonic and anelastic effects reducing seismic velocities at a given frequency. We choose 0.1 Hz, which is a typical order of frequency for teleseismic P-waves. Other rock physics parameters are assumed as in Kreutzmann et al. [2004]. For the grain size which influences the anelastic Vp-reduction, we choose 0.01 m typical for olivine mantle xenoliths. Because the anelastic effect is important only at high temperature, the temperature dependence of Vp is stronger at high temperature, or in reverse, at high ambient temperature a smaller temperature anomaly is needed to explain an observed Vp-decrease. Thus, the interpretation of a given Vp-decrease requires an assumption about the ambient reference temperature at 50 km depth which—per definition—would be associated with 0% Vp-anomaly. As we do not have this information, we arbitrarily test different reference temperatures, namely 900°C, 1200°C, and 1337°C, the latter being equal to the solidus temperature of peridotite at 50 km depth. For these reference temperatures, a 3–5% Vp-decrease may be explained by a positive temperature anomaly of 362–552, 286–432, and 248–376 K, respectively. Adding these anomalous temperatures to the assumed reference temperatures clearly shows that supersolidus conditions are met for almost all of these cases. Only for ambient temperatures below 700°C at 50 km depth the anomalous region remains at a subsolidus condition. It should be noted that these results on anomalous temperatures depend on various rock physics parameters, and have to be taken with care. Varying those parameters within reasonable ranges may change the estimated anomalous temperature estimates by about ±70 K.

[25] As the above analysis strongly suggests that supersolidus temperatures are reached within the anomalous region at 50 km depth, we will now constrain possible maximum amounts of partial melt. The effect of partial melt on seismic velocities depends on the geometrical distribution of the melt phase within the solid matrix and the melt fraction. Schmeling [1985] calculated effective elastic moduli of a composite elastic material containing connected fluid (i.e., melt) inclusions assuming various idealized inclusion geometries in the limits of high and low seismic frequencies. The high-frequency limit corresponds to the unrelaxed state, in which fluid flow between inclusions of different shape and orientation has not yet occurred, i.e., the fluid pressure has not equilibrated within the melt phase. Melt squirt [see Schmeling, 1985, and references therein] equilibrates the fluid pressure, and in the relaxed state or low frequency limit the fluid pressure is equal in the melt phase. This derivation of relaxed and unrelaxed elastic moduli has been revisited by Schmeling et al. [2012] to allow for calculation of P- and S-velocities as a function of melt fraction and various combinations of idealized melt inclusion geometries. These geometries include thin films assumed to wet grain boundaries, melt pockets idealized as ellipsoidal inclusions with short to long half-axes ratios (aspect ratios) of 0.1–1, the latter case corresponding to spherical inclusions, and additionally tubes with tapered cross sections representing melt distributed along grain edges. Assuming elastic properties typical for 50–60 km depth (bulk modulus of 130.7 GPa, shear modulus of 67.7 GPa, bulk modulus of melt of 27.4 GPa) and a solid and melt density of 3377 and 3056 kg/m3, respectively, Vp-decreases as a function of melt fraction are shown in Figure 7 for different melt geometries for both the unrelaxed and relaxed (dashed) cases. The observed P-anomalies at a depth of 50 km range between −3% and −5%. In the limiting case that this decrease is entirely due to the presence of melt, the horizontal −3% and −5% lines in Figure 7 allow to assess the corresponding melt fractions. Thus, if we assume a zero P-anomaly for zero melt fraction, the velocity decrease can be explained by melt fractions between 0.36% and 6.2% depending on melt geometry.

Figure 7.

Seismic P-velocity as a function of total melt fraction for various geometries of melt inclusions based on a theoretical poro-elastic melt—solid model of Schmeling [1985] and Schmeling et al. [2012]. Favored curves are pockets0.3 (see text). Horizontal lines give the observed velocity decrease at around 50 km depth. Subscripts u (solid curves) and r (dashed curves) refer to the unrelaxed (high frequency) and relaxed (low frequency) cases, respectively. Inset figure: simultaneously allowing for melt fraction and temperature-related Vp-decrease using the pocket0.3 assumption and anharmonic and anelastic temperature dependence of seismic velocities (see Kreutzmann et al. [2004] for details). Assumptions for the temperature dependence: seismic frequency 0.1 Hz, depth = 50 km, reference temperature 1337°C (= solidus temperature at 50 km depth), grain size 0.01 m, all other parameters as given in Kreutzmann et al. [2004].

[26] Laboratory experiments show that the equilibrium melt distribution in partially molten peridotite is a combination of tapered melt pockets and tubes [Faul, 1997]. Based on such observed melt geometries, we determined the effective modulus decrease utilizing finite element models. In these models [Zippel, 1996], realistic melt geometries have been taken from photographs of thin sections of partially molten peridotite with various melt fractions [Faul, 1997]. These photographs have been transformed into 2-D purely elastic finite element grids, assuming different bulk moduli (66, 20 GPa) for matrix and melt, as well as different shear moduli (40, 0 GPa, respectively). These setups have been subjected to external compression and shear, and the effective moduli of the unrelaxed state have been calculated for the different melt fractions. From these Kreutzmann et al. [2004] derived the P-decrease as (∂ln Vp/∂φ)T = −1.23, where φ is the melt fraction. This agrees well with models with an idealized melt geometry of tubes or with aspect ratio 0.1 pockets assuming identical elastic moduli. Transferring this result to our 50 km deep conditions with different elastic moduli (see values above) the tube model leads to (∂ln Vp/∂φ)T = −1.49. While these decreases are smaller than those of Hammond and Humphreys [2000], they agree well with the model and laboratory results of Takei [1998, 2000], if a dihedral angle typical for the olivine–melt system is used. Applying our model to the observed P-decrease of 3–5% results in melt fractions between 2% and 3.3% (Figure 6).

[27] While the two previous paragraphs assumed either the temperature or the melt effect separately, it is likely that both effects superimpose in nature. Combining the two effects, both of which decrease seismic velocities, Figure 7 (inset) shows how a given Vp-decrease may be partitioned between partial melting and elevated temperatures assuming a reference temperature equal to the solidus temperature (see above). In case of batch melting (i.e., all the melt stays where it is generated), most of the Vp-decrease will be due to melting, as the degree of melting of 4% corresponds to only about 5–10 K supersolidus temperatures in a partially molten peridotite. If, however, a considerable fraction of melt has been extracted out of the source region, the melt we see is residual, retained melt, and any combination of melt fraction and anomalous temperature along a constant Vp-decrease curve in Figure 7 (inset) may explain the observations. For example, the combinations 1–2% melt and 130–180 K anomalous temperatures are well suited for the observed 3–5% Vp-decrease. Further information, such as S-wave velocity anomalies or seismic attenuation, is needed to resolve the nonuniqueness. As this figure assumed a reference temperature equal to the solidus temperature, what would be different for a smaller reference temperature? In this case, the solidus temperature will be intersected at some higher anomalous temperature in Figure 7 (inset), e.g., at 137 K for a 1200°C reference temperature and a 1337°C solidus temperature. Above that temperature lines of constant P-velocity anomaly can be constructed in a similar way as shown here, but with a smaller maximum amount of possible melt near the solidus and smaller slopes because the maximum anomalous temperatures are higher (c.f. discussion above).

[28] Several points should be noted: if the tomography underestimates the seismic anomalies at that depth of 50 km, the melt fractions and/or anomalous temperature will be higher accordingly. Other effects such as volatiles (water) and compositional variations (depletion) also influence P-velocities. As these two effects work in opposite directions it is difficult to judge whether our estimates on melt fraction and anomalous temperature are upper or lower bounds. Without any further observational data our present model cannot resolve these issues.

[29] In Figure 8, the regions associated with the predicted melt fractions and anomalous temperatures are indicated by “melt inclusions.” They have interesting relations to other observed features in the Rwenzori Mountains region. One of these is observed north-east of the Rwenzori Mountains (Figure 3). Here, we detected a vertically oriented, cylindrical low-velocity anomaly with maximum amplitudes in the middle crust and the upper mantle lithosphere, surrounded by relatively high-velocity material of the rift shoulder. Synthetic tests have shown that this feature is most likely produced by two separate anomalies. A possible interpretation (see Figure 8) is that these anomalies indicate reservoirs of molten material and fluids at different depth levels, which could be connected by a fracture system. This hypothesis is supported by the existence of earthquake swarms that were detected by Lindenfeld et al. [2012] in the Rwenzori region. The clusters are located within a restricted area north-east of the mountains, in the upper crust at depths between 5 and 16 km—just above the observed velocity anomaly (Figure 3, left, dashed ellipse). The spatiotemporal characteristics of the recorded earthquake swarms show numerous similarities in comparison to swarms observed in other continental rifts, like the Kenya rift, the Eger rift, or the Rio Grande rift. From petrological considerations, the authors argue that the recorded earthquake swarms are presumably caused by crustal fluid migrations and CO2 emanations rising from a magmatic body in the upper mantle. The transport of these volatiles concentrates along vertical oriented intersections of preexisting crustal fault systems accompanied by pressure perturbations which are able to trigger seismic energy release.

Figure 8.

Possible interpretation of the obtained results. Question marks indicate not well-resolved regions. Moho depth and location of the S-velocity discontinuity are taken from results of receiver function studies [Wölbern et al., 2010, 2012]. See text for details.

[30] Additional evidence for a magmatic body in the lithospheric mantle is provided by the detection of mantle earthquakes between 53 and 60 km depth [Lindenfeld and Rümpker, 2011]. These events are located close to the northern border of the observed low-velocity anomaly (Figure 3, left, yellow ellipse). The existence of deep earthquakes within the hot and weak material of a rifting regime is explained by magmatic impregnation of the lithosphere as proposed by Foley [2008]. Rapid magma movements, fed by the flanking magmatic body, and associated with high transient strain rates may cause seismic deformation and energy release at depth beneath the rift. Therefore, the detected mantle earthquakes provide a linkage between the observed upper mantle anomaly and the postulated melt transport in the crust (Figure 8).

[31] If our interpretation of the low-velocity regions being due to the presence of melt inclusions and elevated temperatures is correct, it may provide new constraints for quantifying the dynamic process of magma-assisted rifting. Such constraints include (a) the observed depth ranges within the lithosphere where magma is believed to be intruded, and (b) the lateral extent of these melt rich zones. Both these parameters have been shown to be essential in controlling the melt-induced weakening during rifting [Schmeling and Wallner, 2012].

[32] In conclusion, the newly detected seismic low-velocity anomalies within the lower crust and upper mantle lithosphere in the western branch of the East African rift System, combined with seismicity observations and petrophysical interpretations estimating melt fractions and anomalous temperatures suggest that magmatic processes are presently active within the two parallel rift segments surrounding the Rwenzori Mountains and particularly where the Rwenzori block is still connected to the rift shoulder.

Acknowledgments

[33] Funding for this study was provided by the Deutsche Forschungsgemeinschaft (DFG). The study was partly supported by Integration Project SB RAS No. 14. We are grateful to the Geophysical Instrument Pool of Deutsches GeoForschungsZentrum Potsdam (GFZ) for providing the seismological equipment. GEOFON (also at GFZ) is thanked for archiving the data. We gratefully acknowledge support from the Geology Department of Makerere University, Kampala, the Ugandan National Council for Science and Technology (UNCST), and the Ugandan Wildlife Authority (UWA).

Advertisement