Active Fault Scarps in Southern Malawi and Their Implications for the Distribution of Strain in Incipient Continental Rifts

The distribution of deformation during the early stages of continental rifting is an important constraint on our understanding of continental breakup. Incipient rifting in East Africa has been considered to be dominated by slip along rift border faults, with a subsequent transition to focused extension on axial segments in thinned crust and/or with active magmatism. Here, we study high‐resolution satellite data of the Zomba Graben in southern Malawi, an amagmatic rift whose topography is dominated by the west‐dipping Zomba fault. We document evidence for five subparallel fault scarps between 13 and 51 km long spaced ~10–15 km apart. The scarps consist of up to five segments between 4 and 18 km long, separated by minima in scarp height and river knickpoints. The maximum height of each fault scarp ranges from 9.5 ± 4.2 m to 35.3 ± 14.6 m, with the highest scarp measured on the intrabasin Chingale Step fault. We estimate that the scarps were formed by multiple earthquakes of up to Mw7.1 and represent a previously unrecognized seismic hazard. Our calculations show that 55 ± 24% of extensional strain is accommodated across intrabasin faults within the ~50 km wide rift. This demonstrates that a significant proportion of displacement can occur on intrabasin faults during early‐stage rifting, even in thick continental lithosphere with no evidence for magmatic fluids.


Introduction
The development and interaction of faults dominate deformation in the early stages of continental rifting and contribute to the eventual breakup of continental lithosphere (Cowie et al., 2005;Ebinger & Scholz, 2012). Numerical models of the early phase of continental rifting and nonvolcanic rifted margins suggest that faults grow in isolation and are distributed across a region of extension (Brune, 2014;Cowie, 1998;Cowie et al., 2000;Heimpel & Olson, 1996;Huismans & Beaumont, 2007, 2014. Because of changes in sedimentation, displacement rates, fault geometry, volcanism, and rift extension direction that can occur during later phases of continental extension (Gawthorpe & Leeder, 2000;Cowie et al., 2000;Buck, 2004;Ebinger, 2005;Brune, 2014;Philippon et al., 2015), patterns of distributed faulting generated during these early stages of rifting are commonly obscured. Consequently, there are also conceptual models, especially those based on the East African Rift System (EARS), which suggest that early-stage rifting (first~10 Ma) is dominated by activity on border faults (Corti, 2009;Ebinger, 2005). Observations from Lake Tanganyika suggest that up to 90% of early-rift extensional strain is accommodated on border faults (Muirhead et al., 2019) and that intrabasin faulting is limited to small-displacement faults that accommodate flexure (Turcotte & Schubert, ©2020. The Authors. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

10.1029/2019TC005834
Key Points: • We report five subparallel, previously unrecognized, active faults in the Zomba Graben, Malawi, with 10-50 km long, 10-35 m high fault scarps • Border faults are thought to dominate in incipient rifts, but we find intra-rift fault scarps accommodate 55 ± 24% of extensional strain • The Zomba Graben is a zone of relatively high seismic hazard linking Lake Malawi to the Urema Graben in Mozambique Supporting Information: • Supporting Information S1 • Table S1 • Table S2 • Table S3 • Table S4 • Table S5 • Table S6 • Table S7 • Table S8 • Table S9 • Table S10 • Table S11 To investigate the geometry and strain distribution within a youthful rift, we analyze the geomorphic signature of an active normal fault array in the Zomba Graben, which is at the southern end of the incipient, amagmatic Malawi Rift. Although the topography across the graben is dominated by a west-dipping border fault, we detect five previously unrecognized active fault scarps using field data and geomorphic analysis of high-resolution topographic data. Our analysis suggests that current rift-related deformation is distributed on both the border faults and faults within the rift interior. Our findings have implications for our understanding of the evolution of strain in the early stages of continental rifting and the associated seismic hazard.

The Malawi Rift
The~900 km long Malawi Rift System is located at the southern end of the largely amagmatic western branch of the EARS (Ebinger et al., 1987; Figure 1). The rift along Lake Malawi is defined by a series of asymmetric half-grabens with segmented border faults that offset Proterozoic medium-to high-grade metamorphic rocks with fabrics formed in multiple Precambrian orogenic events (Fritz et al., 2013). The border faults related to the current rift regime are up to~120 km long and have throws of at least~7-8 km, and the hanging wall sediment thickness is at least 5 km (Contreras et al., 2000;Accardo et al., 2018;Shillington et al., 2020;Figures 1c-1e). Sediment sequences observed in the hanging walls of these border faults decrease in thickness from >5 km in the north of the lake to~1 km in the south (Scholz & Rosendahl, 1988;Shillington et al., 2020;Specht & Rosendahl, 1989). Faults have been mapped in Lake Malawi by a number of seismic reflection surveys carried out in the lake (Lyons et al., 2011;Mortimer et al., 2007;Scholz, 1995;Scholz & Rosendahl, 1988;Shillington et al., 2016) as well as a recent network of lake bottom and onshore seismometers in northern Malawi and Tanzania (Accardo et al., 2018;Shillington et al., 2016). These seismic surveys, along with the 2009 Karonga earthquake sequence that occurred within the hanging wall of the rift-bounding Livingstone fault, indicate that both rift border and intrabasin faults are currently active at the northern end of the Malawi Rift (Biggs et al., 2010;Gaherty et al., 2019;McCartney & Scholz, 2016). Active faulting is also occurring at the southern end of the lake, as demonstrated by the M w 6.1 1989 Salima earthquake (Jackson & Blenkinsop, 1993).
Rifting within the western branch of the EARS is thought to have initiated in the Oligocene (~25 Ma; Roberts et al., 2012). Low temperature apatite thermochronology from samples in the northern basin of Lake Malawi indicates that rifting commenced at~23 Ma (Mortimer et al., 2016). This is earlier than previously proposed age of~9 Ma based on radiometric dating of volcanic and volcanoclastic deposits from northern Malawi (Ebinger et al., 1993). The age of the rift in central and southern Malawi is poorly constrained. A 4.6 Ma age has been proposed for the onset of sediment accumulation in Lake Malawi's central basin (~350 km to the north of the Zomba Graben), from extrapolating the average rates of sediment accumulation iñ 1.3 Ma drill core (Lyons et al., 2015) to the entire sedimentary sequence (McCartney & Scholz, 2016). This would suggest a gradual southward propagation of the rift (Contreras et al., 2000;Ebinger et al., 1987), which is consistent with sediment thickness and footwall topography decreasing from north to south along the rift (Flannery & Rosendahl, 1990;Laó-Dávila et al., 2015;Specht & Rosendahl, 1989). An alternative hypothesis is that the onset of extension is uniform along the Malawi Rift, but the extension rate is faster in the northern part of the rift because the Euler pole of the plates is located south of the Malawi Rift (Calais et al., 2006;Saria et al., 2014 ;Stamps et al., 2018). Although there are currently few constraints on the age of the EARS to the south of Lake Malawi (Dulanya, 2017), we consider it unlikely that extension initiated in this region prior to the onset of sedimentation in Lake Malawi. This places a poorly constrained approximate maximum age of the rifting in the Zomba Graben as the mid-Miocene to early Pliocene (Delvaux, 1995).
The onshore rift south of Lake Malawi consists of three linked half-grabens ( Figure 1, Williams et al., 2019). The border faults have escarpment heights of <1,000 m and so are less prominent than those within the lake (Laó-Dávila et al., 2015;Figures 1c-1e). The Shire River, which has been the outlet of Lake Malawi sincẽ 800 ka, flows through the center of these grabens (Ivory et al., 2016; Figure 1). The NW-SE trending Makanjira Graben contains two known active intrabasin faults: the~55 km long Malombe fault and thẽ 110 km long Bilila-Mtakataka fault, a possible source of the 1989 M w 6.1 Salima earthquake (Jackson & Blenkinsop, 1993;Hodge et al., 2018Hodge et al., , 2019. The graben is bounded by the Chirobwe-Ncheu and  Craig et al. (2010) with the exception of the red focal mechanism which shows the location and CMT mechanism of the 17 March 2018 Nsanje earthquake. The Sh max direction is from Delvaux and Barth (2010). Rift border faults (in red) are adapted from Laó-Dávila et al. (2015). Foliation trend lines (in pink) are adapted from Williams et al. (2019). (c-e) Crustal-scale geologic cross sections and topographic profiles showing the major border faults and sedimentary basins across the Malawi Rift. Above 0 m elevation is to scale, below 0 m elevation is not to scale. Major border faults are marked in thicker red lines. The black line is the median topography in a 5 km either side of the lines shown in part b. The grey shading represents the maximum and minimum topography in that same swath. Geology has been adapted from Fritz et al. (2013). Basin and sediment depths for profiles a and b are adapted from Accardo et al. (2018) and Shillington et al. (2020), but intrabasin faults are not plotted. In the Zomba Graben, the thickness of hanging wall sediments is poorly constrained, but boreholes have suggested thicknesses of <50 m in places (Bloomfield & Garson, 1965;see also Figures S3 and S6). Note that the nature of the boundaries between the Ubendian and Irumide belts and the South Irumide Belt and the Unango Terrane are uncertain. (f) Overview map of southern Malawi showing the location of the Zomba Graben relative to other known faults in the region. Faults where scarps have been detected and measured are indicated in black. Faults which are suspected as active but with no measurements of throw or where no fault scarp has previously been detected are shown in blue. GPS vector relative to a stable Nubian plate is shown in red (Stamps et al., 2018). Foliation trend is shown in pink.
Mwanjage faults, although it is not known whether these are currently active or not. Immediately south of the Makanjira Graben lies the Zomba Graben, which is the focus of this study and described in more detail in section 2.2. To the south of the Zomba Graben, in the middle Shire Valley, the river drops by~380 m in elevation with no evidence of active faulting (Dulanya, 2017; Figure 1). The NW-SE trending Lower Shire Graben lies~60 km further south and is a reactivated Karoo-age basin bounded to the east by the 85 km long active Thyolo fault (Figure 1b; Hodge et al., 2019). The EARS continues south into the Urema Graben, the site of repeated seismic activity following the 2007 M w 7.0 Machaze earthquake (Lloyd et al, 2019;Copley et al., 2012).
Kinematic models of earthquake slip vectors and GPS measurements in the Malawi Rift indicate an extension direction of 086°± 5° (Saria et al., 2014;Stamps et al., 2018). The variation in the strike of the faults within the southern Malawi Rift varies between NW-SE and NE-SW trending grabens. Consequently, many of the faults that have reactivated within the rift are slightly oblique to the regional extension direction (Figure 1b; Williams et al., 2019). Despite this, the available earthquake focal mechanisms and fault slickensides indicate that the faults in the rift are reactivating as dip-slip structures in the current rifting regime (Williams et al., 2019).

The Zomba Graben
The Zomba Graben is a NE-SW trending segment of the onshore Malawi Rift where the dominant structure is the NW-dipping Zomba fault. The Zomba Graben has traditionally been considered a half-graben similar to those in Lake Malawi (e.g., Ebinger et al., 1987), but Laó-Dávila et al. (2015) map it as a~50 km wide fullgraben, bounded to the west by the SE-dipping Lisungwe fault ( Figure 1e). Other than the disputed mapping of the rift border faults at the edges of the graben, little is known about the pattern of faulting in this region or the distribution of strain (Figure 1). A set of escarpments in the center of the graben that offset fluvio-lacustrine sediments have been variously mapped as "terrace features" (Bloomfield, 1965; Figure 2), active fault scarps (Dixey, 1926), and inactive Late Jurassic or Early Cretaceous faults (Dixey, 1938).
The topography of the Zomba Graben is influenced by the structure and composition of the basement complex, which comprises Proterozoic metamorphic rocks of the Southern Irumide Belt and subsequent intrusions ( Figure 2; Manda et al., 2019). At the regional scale, the rift follows the strike of the Southern Irumide foliation (Figure 1f). On the western side of the rift, the Kirk Range is composed of metasedimentary schists, paragneisses, and granulites and contrasts with the lower elevation eastern side, which is composed of meta-igneous charnockitic granulites ( Figure 2; Bloomfield & Garson, 1965). Syn-and post-kinematic intrusions form local regions of high topography. The most notable of these are the Proterozoic Chingale Ring Complex and the Upper Jurassic-Lower Cretaceous Chilwa Alkaline Province, which formed during a phase of NE-SW extension (i.e., orthogonal to the current extension direction; Bloomfield, 1965, Eby et al., 1995Castaing, 1991). Although the Zomba Massif, a quartz-syenite and granite intrusion, dates from this period, there is no evidence for contemporaneous Cretaceous-Jurassic age faults or sediments.

Active Faults Within the Zomba Graben
Active faults in Malawi are typically identified by the surface exposure of a continuous, steep scarp at the base of a footwall escarpment (Jackson & Blenkinsop, 1997;Hodge et al., 2018Hodge et al., , 2019. Supplementary evidence is provided by observations of soft, unconsolidated hanging wall sediments, preserved uplifted river terraces in the footwall, and knickpoints in streams within <100 m of the fault scarps (Jackson & Blenkinsop, 1997). These criteria have been used prior to this study to identify three active faults south of Lake Malawi but outside the Zomba Graben: the Bilila-Mtakataka, Malombe, and Thyolo faults (Figure 1c; Hodge et al., 2019). As no active faults had previously been documented in the Zomba Graben, geological maps were used to first identify faults where Bloomfield (1965) mapped Tertiary-Recent hanging wall sediments (Figure 3a). The detailed topographic analysis described below was then used to determine whether a fault has formed an active fault scarp, in which case it would be considered currently active. These satellite-based observations were used to target a 6-week field campaign in 2018, which confirmed the results and enabled more detailed analysis of the scarp morphology in a few selected locations.

Topographic Analysis to Assess Fault Activity
We used a TanDEM-X digital elevation model (DEM) of the Zomba Graben to identify and analyze fault scarps and the river channels that cross them ( Figure 3). TanDEM-X DEMs have a horizontal resolution of 12.5 m and an absolute vertical mean error of ±0.2 m (RMSE < 1.4 m; Wessel et al., 2018), which is sufficient for measuring the meter-scale fault scarps in the Zomba Graben (see Hodge et al., 2019, for discussion of DEM morphology and scarp height in this region). We considered the presence of a linear scarp coinciding with changes in channel incision and width and hanging wall sediment deposition, as evidence for active faulting during the current rifting episode ( Figure 3).
We produced slope maps from the DEM by calculating the scalar magnitude of the slope derivative using the grdgradient tool in the Generic Mapping Tools routines (Wessel & Smith, 1998;Figure 3b) and use these maps to identify the location of the active fault scarps. We noted any gaps in the scarps, or peaks and troughs in the displacement profiles, that may be indicative of fault segmentation (following Hodge et al., 2018). We extracted 500 m long fault-perpendicular topographic profiles every 12 m and stacked them at 100 m intervals. The stacking has the effect of removing short-wavelength topographic features not related to the fault such as local sedimentation and erosion, or human settlements and vegetation close to the fault. The scarp height was measured by fitting regression lines to the footwall and hanging wall topography of the stacked profiles and calculating the vertical difference between the extrapolated regression lines at the point of maximum steepness on the scarp (Avouac, 1993;Figure 4a). To estimate the uncertainty, we applied a Monte Carlo approach by selecting 10,000 random subsets of points from the footwall and hanging wall and allowing the exact location of the fault to vary. The variation between profiles is due to a combination of differential geomorphic degradation of the fault scarp and variation in fault offset that forms during an earthquake. Example profiles for each fault are shown in Figure 4. We filter the resulting measurements along strike using a 3 km wide moving median (Hodge et al., 2018(Hodge et al., , 2019. Rivers that cross normal faults record information about the timing and magnitude of active faulting in their long profile (Boulton & Whittaker, 2009). River long profiles, which plot the change in elevation of the river channel with distance along the course of the river, can therefore be used as an indicator of fault activity in regions where the location of active faults is poorly constrained (Boulton & Whittaker, 2009). For example, on rivers that cross active normal faults, a change in uplift rate in the footwall of the fault leads to the river channel responding transiently by becoming steeper and more incised upstream of the fault . The knickpoints or knickzones that form as a result of this process propagate upstream through time, with their vertical rate of propagation a function of the magnitude of uplift rate change (Attal et al., , 2011Whittaker et al., 2008). We extracted long profiles of rivers in the Zomba Graben with a drainage area greater than 6,000 m 2 using TopoToolbox (Schwanghart & Scherler, 2014). We identified changes in channel steepness that are evidence of perturbations to the power-law relationship, S = k sn A −θref , between local channel gradient, S, and upstream drainage area, A, in detachment limited rivers (Whipple & Tucker, 1999). We used a reference concavity index, θ ref = 0.45 to calculate a normalized channel steepness index value, k sn , using the 12.5 m resolution DEM, thus facilitating a comparison between streams with a large range of drainage areas (Wobus et al., 2006). Relatively high k sn values reflect knickpoints or knickzones where river channels are transiently adjusting to perturbations to the channel gradient-drainage area relationship. Thus, we look for areas where linearly aligned increases in k sn values coincide with our other morphological indicators of active faulting ( Figure 3). Consistent along-strike increases in k sn upstream of a fault suggest a perturbation of the stream power law caused by changes in uplift rates by active normal faulting (Wobus et al., 2006;Figure 3d). In these cases, relatively lower k sn values further upstream reflect the channel conditions prior to tectonic disturbance.

. Description of Fault Scarps
Based on the topographic analysis, we found evidence for five active fault scarps within the Zomba Graben that we subsequently confirmed during a 6-week field campaign in 2018. In this section, we describe the tectonic geomorphology and field observations of two border faults, Zomba and Lisungwe, and three intra-rift faults, Chingale Step, Mlungusi, and Mtsimukwe. For each fault, we report the dimensions of the fault scarps discovered at the base of footwall escarpments (all measurements are listed in supporting information Tables S1-S6) and describe the other features that led us to conclude they are active. Examples of the fault scarps measured on each fault are shown in Figure 4. The measurements are shown in Figure 5 and summarized in Table 1.
The NW-dipping Zomba fault borders the eastern side of the Zomba Graben and has a~50 km long escarpment with an active fault scarp at the base (Figures 5a, 5b, and S2). The mean height of the fault scarp is 15.6 ± 5.2 m (Figures 4a and 5b). The scarp height measurements show evidence for five 4-18 km long segments. Segment 3 has an asymmetric scarp height profile, with the highest scarp height measured at the northern end of the segment. The maximum scarp height in each segment varies between~17 m and~31 m, with the highest scarp height observed in the southernmost segment. Normalized channel steepness values (K sn ) increase where rivers cross the fault ( Figure S2d). Higher K sn values 3-4 km into the hanging wall at the northern end of the fault are associated with the distal edges of alluvial fans ( Figure S2d). The scarp is noticeably steeper adjacent to the Zomba Plateau, and a~2 km step to the northwest occurs between Segments 1 and 2, at the northern end of the fault in front of the Zomba Plateau ( Figures S2b and S2c). At the northern end of the fault, triangular facets were observed in the field (Figure 6a). Large alluvial fans in the hanging wall are composed of material derived from the Zomba Plateau and appear to have been offset by the fault (Figure 6b). In the middle of the fault, a~17 m high fault scarp ( Figure 6c) contains a zone of highly fractured charnockite at its base ( Figure 6d).
The NW-dipping Chingale Step fault (Figures 4a, 5a, 5c, and S3) is located approximately 10 km into the hanging wall of the Zomba fault and has formed a~40 km long scarp. The fault scarp shows evidence of multiple ruptures, with two scarps identified at numerous sites along the fault (Figure 4a). The lower scarp is steepest and is separated from the upper gentler dipping scarp by a break in slope (see also Figure S3). The steeper lower scarp suggests this rupture is younger than the upper scarp. We measured the height of

10.1029/2019TC005834
Tectonics both the lower scarp and the total height of the multiple scarp. The average height of the total scarp was 19.6 ± 12.1, composed of three segments (Figure 5c). Segment lengths vary between 10 and 18 km, with a maximum total scarp height of 35.3 ± 14.6 m on the northern segment. In contrast, the lower scarp is not segmented, and the scarp height is approximately constant along the entire length of the fault (mean height of 5.7 ± 2.5 m; red line in Figure 5c).  (Figure 7a). The vertical difference between the fault and the top of the oversteepened reaches shows two segments, separated by a zone of linkage, which follow the shape of the fault scarp displacement profile (Figure 7). The boundaries between segments observed in the knickpoint elevations also correspond to one of the segment boundaries identified in the scarp height profile. As the vertical movement of a knickpoint is controlled by the magnitude of the perturbation experienced

10.1029/2019TC005834
Tectonics by a river channel (Attal et al., , 2011Whittaker et al., 2008), the correspondence between the knickpoint elevations and the along strike scarp height suggests that a change in fault uplift rate is controlling a transient channel response (Figure 7). Furthermore, the river channel within the linkage zone shows two knickpoints, which is likely related to the separate initiation of faulting of the southern and northern segments (Figure 7c). The consistent offset measured across the lowermost fault scarp (red line in Figure 5c) suggests that these segments have linked in the time since the active fault scarp begun to be preserved and are now operating as a single fault. A scarp was visible along all sections visited, with a zone of fault gouge and fractured basement rocks at the base of the fault scarps in exposed stream beds (Figure 8). At the Kalira River site, the fault surface itself was visible with a polished surface and slickensides (Figure 8). The multiple scarp was not visible in the field, although most accessible locations were on the southern end of fault where the difference between the total scarp and the lowermost scarp was smaller than along the northern segment of the fault. The SE-dipping Mlungusi fault has formed a~20 km long scarp in the center of the Zomba Graben (Figures 4a, 5a, and 5d) and has a mean scarp height of 6.9 ± 3.1 m. The slope map shows a prominent scarp along the entire length of the fault ( Figure S5), except in two locations where rivers flow along the fault (Figure 5d). In contrast, K sn values are only elevated in the center of the fault ( Figure S5d). The Shire River crosses from the hanging wall into the footwall at the southern end of the Mlungusi fault, where it changes from a wide, meandering channel with a floodplain in the hanging wall to a narrow, incised channel with an associated set of rapids as it crosses into the footwall (Figure 9). Along strike from this location, a steep fault scarp was observed in the field (Figure 9c) which has offset fine-grained fluvial-lacustrine deposits found on the floor of the graben. The fault scarp height profile (Figure 5d) possibly shows evidence of a segment boundary in the center of the fault, although obvious segmentation is obscured by erosion of the fault scarp by the Shire River (Figure 5d). In the center of the fault, the scarp is covered by clast-supported rounded to subrounded pebbles within a sandy matrix, which we interpret as lacustrine beach deposits (Figure 9c). Long-term footwall uplift has led to drainage reorganization as rivers draining into the axial Mtsimukwe River in the footwall of the Mlungusi fault are restricted to the western side of the channel, distal to the fault ( Figure S5d).
The E-dipping Mtsimukwe fault has formed a~13 km long scarp trending~N-S (Figures 4a, 5a, and 5e). The mean scarp height is 3.6 ± 0.7 m. The slope map shows that the fault scarp is best preserved in the central part of the fault except a~2 km long section where the fault intersects a road ( Figure S6b). The streams that cross the fault show a small increase in K sn value, with the largest increase in center of the fault, but there is no evidence of segmentation ( Figure S6).
The E-dipping Lisungwe fault is the western border fault of the Zomba Graben (Figures 2, 5a, and 5f).
High slope values have a linear trend aligned with an increase in K sn in the streams that cross the escarpment ( Figure S7). The mean offset across the fault scarp is 10.0 ± 6.7 m (Figures 4a and 5f).
No fieldwork was conducted on the Mtsimukwe and Lisungwe faults due to access reasons, but the remote sensing observations described above and seen in Figures S6 and S7 are similar to the observations on faults (e.g. Zomba, Chingale Step and Mlungusi faults) where fieldwork was able to confirm the evidence gained remotely.

. Fault Kinematics and Extension Direction
Slickensides along the Chingale Step fault (Figures 8b-8d; plunging 52°towards 301°) and kinematic indicators on faults elsewhere in the Zomba Graben (Bloomfield, 1965;Chorowicz & Sorlien, 1992) indicate an approximately NW-SE extension direction (Williams et al., 2019). This extension is orthogonal to the surface traces (015 ± 11°) of the five faults, indicating a generally dip-slip displacement (Williams et al., 2019). This contradicts previous rift-wide estimates of extension (Delvaux & Barth, 2010; Figure 1). However, these estimates likely reflect that normal faulting events in the Malawi Rift are approximately purely dip-slip despite regional changes in fault strike (Williams et al., 2019), as also indicated by the 2009 Karonga earthquakes (Biggs et al., 2010), the 1989 Salima earthquake (Jackson & Blenkinsop, 1993), and a M w = 5.6 earthquake in March 2018 (red focal mechanism in Figure 1b). An extension azimuth of 072°is recorded geodetically at Zomba (Figure 1; Stamps et al., 2018) in contrast to the NW-SE extension in the Zomba Graben. This variation in extension direction is consistent with previous observations of deviation between local and regional extension directions in Malawi (Delvaux & Barth, 2010;Hodge et al., 2018;Williams et al., 2019).

Age of the Fault Scarps
The faults in the Zomba Graben have offset the sedimentary cover on the floor of the graben, which includes coarse-grained sandstones of the Matope beds, as well as sandy, pebbly, fluvio-lacustrine deposits in the center of the rift (Figure 2). Both of these sediment packages were likely deposited in the Late Quaternary (sincẽ 800 ka) reflecting fluctuations in the level of Lake Malawi. Prior to the Mid-Pleistocene Transition (MPT; 800 ka), Lake Malawi was much shallower than the present day and likely drained through an outlet at the northern end of the lake (Ivory et al., 2016). Since the MPT, Lake Malawi experienced a significant base level rise and now fluctuates between high-stand and low-stand conditions (up to~500 m lower) and now outlets through the Shire River valley into the Zambezi River (Ivory et al., 2016;Lyons et al., 2015;McCartney & Scholz, 2016;Owen et al., 1990). Since~75 ka, high-stand conditions have prevailed, during which lake levels may have been up to~150 m higher than present (Lyons et al., 2015;McCartney & Scholz, 2016). The floor of the Zomba Graben lies~10 m above the current lake level; therefore, during high stands, lacustrine sediments will have been deposited in the graben, whereas low stands were accompanied by the deposition of coarse-grained sediments (Lyons et al., 2011(Lyons et al., , 2015. We conclude that the observed fault scarps have formed during the Late Quaternary time period: <800 ka and maybe as recently as <75 ka as they offset and have been draped by both fineand coarse-grained sediments. A number of observations are consistent with this interpretation: (i) The fault scarps are generally steep (Figures 3b and 4a); (ii) scarp diffusion modelling on the Bilila-Mtakataka fault, to the north of the Zomba Graben, suggests a~20 m high fault scarp formed within the last 6.4 ± 4 ka (Hodge et al., 2020). Although this age represents the age of the scarp, not the fault itself, and is heavily dependent on the diffusion age used in the modelling, which is poorly constrained, it nonetheless demonstrates that the fault scarps observed in southern Malawi are active in the Late Quaternary. (iii) The Mlungusi fault is draped by a layer of clast-supported, subrounded to rounded cobbles in a sandy matrix, consistent with a lacustrine paleobeach deposit, which was likely deposited during a Late Quaternary high stand within Lake Malawi (since the MPT~800 ka); and (iv) optically stimulated luminescence (OSL) ages for surficial paleo-fluvio-lacustrine deposits near Lake Chilwa, 20-30 km to the east of the Zomba Graben, are <50 ka (Thomas et al., 2009; Figure 1).
The measured scarp heights (section 3.2) are generally greater than empirically derived single event displacement values for normal faults (e.g. Leonard, 2010). Therefore, and based on the relative age constraints above, the scarps should be viewed as composite scarps representing the cumulative offset from multiple earthquakes during the Late Quaternary (see Hodge et al., 2020 for a discussion on this).

Strain Distribution Across the Zomba Graben
We found evidence for five Late Quaternary active faults within the Zomba Graben. The across strike spacing between the active faults is approximately 10-15 km, and the maximum height of each fault scarp ranges between 9.5 ± 4.2 m and 35.3 ± 14.6 m ( Figure 5). The distribution of fault scarps within the graben is a notable feature. The highest fault scarp was found along the Chingale Step fault (Figure 5c), which is 10.1029/2019TC005834 Tectonics located in the hanging wall of the border fault ( Figure 5a). Thus, the strain within the Zomba Graben over the time period that the active fault scarps have formed is not localized on a single major border fault but instead distributed across the width of the rift (Figure 5a).
We estimate the cumulative strain since the formation of the fault scarps, assuming an approximately uniform age of the fault scarps. We calculate finite strain across the Zomba Graben using the England and Molnar (1997) adaptation of the Kostrov (1974) expression of strain rate: where a is the surface area of the region, L k is the length of fault k in that region, s k is the slip of fault k in that region, ϑ k is the dip of fault k, b u k is a unit vector of fault k in the direction of slip, and b n k is a unit vector of fault k normal to the fault plane. The advantage of this adaption is that it is independent of seismogenic thickness or shear modulus. We assume pure dip-slip faults that strike parallel to each other (see section 3.3), such that ε 22 is equal to zero, and ε 11 is perpendicular to the mean strike of the faults. The only unknown term is fault dip, for which we assume an Andersonian value of 60°, although using a randomly selected fault dip, within the usual range for normal faults (45°-60°; Collettini & Sibson, 2001), produces similar results ( Figure S8). We calculated the strain within 5 km (across strike) by 50 km (along strike) rectangular boxes with long axes orientated 200°, using the maximum scarp height measured along each fault as a proxy for fault throw (following Hodge et al., 2018).
The strain associated with the fault scarps has dominantly occurred across the Chingale Step fault (42 ± 17%) and the Zomba border fault (40 ± 17%), both west-dipping faults on the eastern side of the graben (Figure 10c). The proportion of strain across both east-dipping intrabasin faults (Mtsimukwe: 6 ± 6% and Mlungusi: 7 ± 1%) and the east-dipping border fault (Lisungwe: 6 ± 3%; Figure 10c) is all the same within error. The proportion of strain accommodated across the east-dipping intrabasin faults (13 ± 6%) is within the error of the proportion of strain accommodated on the border fault (6 ± 3%). On the west-dipping faults, there is no significant difference between strain on the intrabasin (42 ± 17%) and border faults (40 ± 17%). Across the whole graben, the proportion of strain that has occurred across the

10.1029/2019TC005834
Tectonics intrabasin faults (55 ± 24 %) and the proportion of strain across the border faults (45 ± 20 %) is roughly equal, although we note the large error bars. This equates to 18.1 ± 7.1 m of finite extension on the intrabasin faults and 14.9 ± 6.7 m of finite extension on the border faults (Figure 10c). The east-dipping faults have accommodated 6.1 ± 3.1 m of finite extension whereas the west-dipping faults have accommodated 26.9 ± 11.4 m.

Strain Migration and Fault Evolution During Rifting
Current conceptual models suggest that following an initial stage of distributed deformation, the transition from rift border faults to intra-rift deformation occurs over timescales of 10-15 Ma in the East African Rift as lithospheric thinning leads to asthenospheric upwelling and production of magmatic fluids (Buck, 2004;Ebinger, 2005;Ebinger & Scholz, 2012;Huismans & Beaumont, 2014). Here we compare our observations from the early-stage Zomba Graben with those from elsewhere in the East African Rift and discuss the implications for the evolution of strain distribution during rift development.
In the Zomba Graben, we observe that 55 ± 24% of the strain is accommodated on intra-rift faults ( Figure 10). While this level of intra-rift strain is commonly observed in magmatic rift segments (e.g., Bilham et al., 1999;Ebinger & Casey, 2001), this is far higher than other amagmatic rifts. In the~10 Ma, magma-poor Lake Tanganyika Rift, the border faults accommodate 90% of extensional strain (Muirhead et al., 2019). Within the northern basin of Lake Malawi, Mortimer et al. (2007) inferred that activity on intra-rift faults diminished over the lifetime of the rift. In contrast, seismic reflection profiles from the central basin of Lake Malawi show that intra-rift faults have been active over the lifetime of the amagmatic rift (McCartney & Scholz, 2016), but these same profiles were unable to constrain the activity on border faults over the same time period.
There are two possible explanations for the roughly equal distribution of deformation between the intra-rift faults and the rift border faults that we observe in the Zomba Graben ( Figure 10): (1) The current distribution represents a period of time prior to the localization of strain across the rift border faults or (2) deformation has migrated from the border faults onto intra-rift structures. Topographic profiles across the northern and central basins of Lake Malawi show that the largest topographic expression is associated with the major border fault(s) at the edge of the rift (Figures 1c and 1d). Within the Zomba Graben, although the border faults are less developed, they still have the largest topographic expression (Figure 1e). This suggests that that over the lifetime of the rift, the Zomba fault, and possibly the Lisungwe fault, have experienced more cumulative displacement than the intra-rift faults ( Figure 1e). Thus, we prefer an interpretation where the approximately even distribution of active strain between border and intra-rift faults has developed after a phase of heightened border fault activity but before observable magmatism. We discuss four mechanisms that may explain this pattern of strain across a distributed network of faults in an apparently nonvolcanic rift: (1) cessation of border fault activity, (2) lithospheric flexure, (3) a cryptic fluid phase, and (4) transient changes associated with fault network evolution and fault linkage. We then discuss whether the pattern of strain can be explained by the lithospheric structure, including crustal heterogeneities.

Cessation of Border Fault Activity and Lithospheric Flexure
The maximum amount of displacement a normal fault can accumulate, before it becomes more favorable to form a new fault, is thought to be controlled by a combination of effective elastic thickness, seismogenic thickness, and surface processes, such as footwall erosion and hanging wall deposition (Olive et al., 2014;Scholz & Contreras, 1998). The exact mechanism of fault abandonment is debated, but it generally requires large total displacements (>5 km) that lead to an increase in the flexural restoring force and/or rotation of the fault dip to unfavorable angles (Goldsworthy & Jackson, 2001;Scholz & Contreras, 1998).
When rift border faults are abandoned, migration of fault activity into the hanging wall of the previously active fault is widely observed (see Goldsworthy & Jackson, 2001, for further discussion). Accardo et al. (2018) propose that the >7 km of throw on the border faults in the northern and central basins of Lake Malawi indicates that these faults are approaching their maximum size and that the migration of strain into the rift interior is imminent or now occurring (e.g., the Karonga earthquake sequence; Biggs et al., 2010;Kolawole et al., 2018). However, this mechanism is unlikely to cause the intra-rift faulting in the Zomba Graben, because both the border and intra-rift faults have been active during the Late Quaternary and the topographic relief is relatively small (<1 km). Similarly, bending forces associated with flexure of the 10.1029/2019TC005834 Tectonics border fault hanging wall can induce strain in the intra-rift region within the upper crust, and although this has been proposed for Lake Malawi (Kolawole et al., 2018), the low throws and thick elastic crust (~30 km) in the Zomba Graben would generate negligible flexural strain (Jackson & Blenkinsop, 1997;Muirhead et al., 2016).

Influence of Fluids
Magma-assisted rifting, where magmatic fluids and volatiles enable extension at lower stresses than the available tectonic forces, is thought to play an important role in facilitating rifting in thick continental lithosphere (Buck, 2004;Ebinger et al., 2017). There is no evidence of active magmatic activity or more than a few kilometers of crustal thinning in the Zomba Graben (Reed et al., 2016;Wang et al., 2019). Furthermore, the geochemistry of hot springs in southern Malawi does not suggest a magmatic influence (Dulanya et al., 2010), and the nearest active volcano, Rungwe, is located~700 km to the north (Figure 1). However, decompression melting or lower crustal magmatic intrusion might not lead to perceptible surface effects, and there is evidence for non-zero crustal thinning beneath southern Malawi (Wang et al., 2019). In an example further south and west in the EARS, low seismic velocities suggest decompression melting in the upper asthenosphere of the Okavango Delta, despite a very low level of extension and no surface volcanism or evidence for mantle upwelling (Yu et al., 2017).
Earthquakes occur in Malawi throughout the 38-42 km thick crust (Craig et al., 2011;Jackson & Blenkinsop, 1993;Tedla et al., 2011), consistent with estimates of effective elastic thickness in excess of 30 km (Ebinger et al., 1999). This can occur in magmatic rift zones if border faults penetrate into the lower crust or if melt and volatile migration into the lower crust causes localized weakening, which can also trigger seismicity (Ebinger et al., 2017). However, in northern Malawi, there is little difference between the focal depth of earthquakes on the intrabasin and border faults (Ebinger et al., 2019). Current imaging of the lower crust in Malawi does not allow us to discriminate between these mechanisms, and alternative explanations for the deep seismicity include (1) a dry, strong, granulite facies lower crust (Jackson et al., 2004), (2) above average lithospheric thickness (Chen & Molnar, 1983), or (3) localized zones of weak rheology within a strong, elastic lower crust (Fagereng, 2013).

Transient Changes Associated with Fault Evolution
The process of fault segment linkage can lead to an increase in fault slip rates over timescales of~100 ka (Taylor et al., 2004). This can subsequently enable a newly linked fault to accommodate a greater proportion of the regional extension rate (Cowie et al., 2005;Taylor et al., 2004). The cumulative scarp on the Chingale Step fault is segmented but the steepest, most recent scarp on the fault cuts across the segment boundaries and is continuous and~6 m high along the length of the fault ( Figure 5b). As well as indicating that faults can rupture multiple segments simultaneously, it suggests that segment linkage has occurred. This implies that faults in this region grow through linkage that occurs before they have accumulated significant (i.e., >1 km) offsets and on short timescales of <800 ka (and possibly <~50 ka). This rapid linkage, before continued slip accumulation, is consistent with the recent hybrid fault growth model where slip accumulates at a constant fault length after an initial growth phase (Rotevatn et al., 2019).
Observations from the Livingstone fault, northern Lake Malawi, suggest that permanent segment linkage that occurred during the past~1.6 Ma has caused displacement rates to become greatest in the locations that were previously between segments (Mortimer et al., 2016). The segment linkage we observe along the Chingale Step fault is not associated with higher displacements at the previous segment boundary locations compared with locations along the rest of the fault (Figure 5b). When faults are closely spaced, as they are in the Zomba Graben, across-strike co-seismic elastic stress changes can drive transient variations in slip rates on individual faults over the timescale of a few earthquake cycles (Cowie et al., 2012). With only one example, we cannot tell whether the displacement that has accumulated on the Chingale Step fault scarp across the segment boundaries is a long-lived feature of the rift or a transient feature only representative of the time window we sampled (i.e., since the fault scarps have been preserved in the Late Quaternary). Nonetheless, although elastic fault interactions are likely contributing to the temporal evolution of strain across the network of faults in the Zomba Graben, they cannot explain the initial formation of a distributed fault network.

Lithospheric Structure
The roles of crustal thickness, lower crustal rheology, and lithospheric thermal structure in developing end-member narrow or wide rifts have been well studied (e.g., Buck, 1991;Huismans & Beaumont, 2007). Typically, strain in narrow rifts concentrates in the weakest part of the lithosphere while in wide rifts,

10.1029/2019TC005834
Tectonics lower crustal viscous flow drives distributed deformation in the upper crust (Buck, 1991). Southern Africa has unusually thick continental lithosphere (140-180 km; Craig et al., 2011) and crust (38-42 km thick; Tedla et al., 2011;Wang et al., 2019), resulting from multiple episodes of orogenic thickening (Fritz et al., 2013). The available constraints on the lithospheric properties of the Zomba Graben suggest that conditions are similar to the very young (120-40 Ka) Okavango Rift (Craig et al., 2011). However, we have demonstrated that strain within the narrow (<50 km) Zomba Graben is distributed whereas the Okavango Rift is >150 km wide and deformation is localized to <50 km wide zones at the edges of the rift (Ebinger & Scholz, 2012). Thus, the currently available constraints on lithospheric thickness and thermal structure cannot explain the differences between these two rifts.
In the Okavango Rift, early localization onto a single border fault has been inferred to arise from deformation along a localized, preexisting weakness (Kinabo et al., 2007). This early localization differs from predictions by numerical simulations, which demonstrate that a strong, ductile lower crust promotes distributed faulting due to enhanced upper crustal fault interactions during rifting (Heimpel & Olson, 1996). The crust in southern Malawi is made up of high-grade metamorphic rocks that lack hydrous minerals with low friction coefficients (Hellebrekers et al., 2019), and exposed fault rocks are not demonstrably different in composition from these high-grade basement rocks (Williams et al., 2019). At some local scales, for example, along the Chingale Step fault, the faults cross-cut the foliation orientation. Thus, although the Zomba Graben follows the trend of the regional foliation, distributed faulting cannot be attributed to continued reactivation of preexisting frictionally weak planes in the brittle regime. However, lateral heterogeneity in the lower crust, such as an anastomosing shear zone system, would enable strain localization in lower viscosity zones in an otherwise cold, strong layer (Fagereng, 2013). Such localization at depth may guide deformation patterns in the overlying brittle crust, such as the oblique strike of faults in southern Malawi, relative to the regional extension direction (Hodge et al., 2018;Williams et al., 2019).
The hypothesis of a lower crust with rheological heterogeneity (i) satisfies the requirements for a dominantly strong lower crust; (ii) can, with sufficient velocity-weakening material (Hellebrekers et al., 2019), explain the deep seismicity observed in the Malawi Rift; (iii) facilitates the formation of a distributed fault network at the surface; (iv) does not require preexisting frictional weaknesses in the upper crust; and (v) can explain the pure dip behavior of the faults in an oblique rift (Philippon et al., 2015;Williams et al., 2019). This suggests that patterns of strain distribution in amagmatic rifts may not be solely controlled by rift maturity; instead, scale, geometry, and degree of rheological heterogeneity in the lower crust as well as elastic fault interactions in the upper crust may also be important.

Implications for Seismic Hazard
Our analysis of high-resolution satellite topography has identified five active faults in the Zomba Graben, but current assessments of seismic hazard in Malawi do not extend south of Lake Malawi (Hodge et al., 2015;Midzi et al., 1999). Based on standard scaling laws (Leonard, 2010), these 15-50 km long faults could host earthquakes of M w 6.3-7.1, assuming that faults do not fail in smaller individual segments and that multiple separate faults do not slip in the same event. The administrative regions within 30 km of the Zomba Graben contain~22% of the Malawian population (~4 million people), including the major population and administrative centers of Blantyre (population:~800,000) and Zomba (population:~100,000; Malawi National Statistics Office, 2018). The proximity of these newly identified faults to this number of people presents a significant challenge for both seismic risk and regional development. Furthermore, the path of the Shire River is strongly affected by rift topography, and a large earthquake on the Mlungusi fault in particular could affect both irrigation and flooding on a regional scale as well as dam stability on the Shire River. This is particularly important as~80% of Malawi's electricity is generated by hydroelectric dams within the middle Shire Valley (Taulo et al., 2015).
In slowly deforming regions and/or regions with poor seismic detection infrastructure, instrumental catalogues are unlikely to record a representative enough sample of seismicity to sufficiently assess seismic hazard. Including fault maps in probabilistic seismic hazard analysis can overcome this problem (e.g. Chartier et al., 2017;Hodge et al., 2015;Pace et al., 2018), but this requires estimates of both earthquake magnitude and recurrence interval. In Lake Malawi, where fault slip rates have not been measured, Hodge et al. (2015) assigned the plate motion to individual faults in their hazard assessment. However, the discovery of multiple, subparallel fault scarps in the Zomba Graben suggests longer earthquake recurrence times and lower 10.1029/2019TC005834 Tectonics probabilities of peak ground acceleration for a given return period. Furthermore, increases in the number of faults within a region also leads to more variable fault slip rates, which in turn can result in earthquake clusters (Cowie et al., 2012). The 2009 Karonga earthquake sequence in northern Malawi illustrates that such clusters occur in the East African Rift (Biggs et al., 2010), but instrumental records are not sufficient to determine how widespread this behavior is. Earthquake clusters present an additional challenge for seismic hazard assessment as they are not captured by time-independent probabilistic seismic hazard assessments.

Conclusions
We use high-resolution TanDEM-X topography to find and measure active faults scarps in the Zomba Graben in southern Malawi. We show that the active fault scarps occur at the base of the rift border faults and also on the rift floor between them. The strain calculated from the heights of the fault scarps is roughly equally distributed between the rift border faults and the intrabasin faults. This presents new insights into the behavior of rifts during the incipient stages of continental extension in a region of thick lithosphere and no active volcanism. In contrast to the prevailing paradigm, it suggests that during the early, amagmatic stages of continental rifting, a significant portion of the extension can occur on intra-rift faults.
Active intrabasin faults have previously been documented during early rifting (McCartney & Scholz, 2016), but they have been inferred to be subsidiary to the border faults (Muirhead et al., 2019). The strain distribution across the rift in the Zomba Graben show that early-rift intrabasin faults can accommodate a similar proportion of extension to the border faults. Therefore, while the overall mode of rifting is likely to be controlled by the rheology of the lithosphere, we suggest that upper-crustal fault interactions and strength variations within the lower crust can lead to spatially distributed and temporally transient faulting within early-stage rifts. Finally, we find that the onshore rift in southern Malawi represents a significant, but previously unappreciated source of seismic hazard within the East African Rift.

Acknowledgments
This work was funded by the EPSRC project "Prepare" (EP/P028233/1), funded under the Global Challenges Research Fund. We thank Kondwani Dombola for his assistance with fieldwork planning and logistics. TanDEM-X data were obtained via DLR proposal DEM_GEOL0686. We thank Donna Shillington, an anonymous reviewer, and the associate editor for reviews that improved this manuscript. Fault scarp profiles and scarp height data are available in the supplementary material and have also been archived within the Zenodo data repository at: http://doi.org/10.5281/ zenodo.3582282.