The 2017 Regent Landslide, Freetown Peninsula, Sierra Leone

At 06:50 on Monday 14 August 2017, a hillslope on the Freetown Peninsula, Sierra Leone, collapsed, sending 300 000 m3 of debris into the flooded valley below. As this debris mixed with floodwater it became a sediment-laden flood which entered a drainage channel and travelled 6 km to the coastline. The event destroyed nearly 400 buildings, claimed the lives of an estimated 1100 people and affected c. 5000 people. The mechanism was a two-stage rainfall-triggered landslide followed by a channelized debris-laden flood. The processes were similar to the nearby 1945 event in Charlotte, which killed at least 13 people. Geomorphological mapping has identified evidence of hundreds of other large landslides that occurred before modern records, providing an appreciation of the slope processes affecting the Freetown Peninsula. Following the 2017 Regent Landslide, rehabilitation of the affected area involved a risk-reduction strategy that centred on reducing population exposure. These events are a reminder that the steep slopes and valleys across the Freetown Peninsula are highly susceptible to rainfall-triggered landslides which, given the topography, have a high propensity to generate high intensity landslides and debris-laden floods. Future urbanization must consider whole-catchment management, flooding and slope engineering issues to provide lasting landslide risk reduction.

tholeiitic intrusion, which intruded the West African Craton during the Early Jurassic c. 200 Ma (Chalokwu et al. 1995;Chalokwu 2001;Callegaro et al 2017). The intrusion has an arcuate outcrop towards the west and extends out under the Atlantic Ocean. It is composed of a layered complex of gabbro, norite, troctolite and anorthosite. The igneous layering dips SW and is transected by several steeply dipping late faults, with both NE-SW and WNW-ESE trends (Umeji 1983). The bedrock geology of Sierra Leone is dominated by Archean basement, formed between 3.5 and 2.8 Ga (Rollinson 2016). The western margin of that cratonic block, along the coast of Sierra Leone, is marked by Proterozoic terranes that were most probably accreted at the end of the Neoproterozoic (de Waele et al. 2015). The Freetown Layered Complex was intruded into this margin at c. 200 Ma (Callegaro et al. 2017) as part of the voluminous Central Atlantic Magmatic Province (CAMP). Finally, the Bullom Group marine and estuarine sediments were deposited unconformably on to the coastal strip of Sierra Leone during the Cenozoic (Dixey 1922;Williams 1978;Morel 1979;MacFarlane et al. 1979MacFarlane et al. , 1981Umeji 1983;Chalokwu 2001).
The present-day topography is the product of a combination of rock-weathering processes during the Quaternary when a thick saprolite soil mantle developed, and periods of erosion of soil and bedrock on slopes by landslides. Laterites consisting of ferricrete surfaces formed at the foot of slopes, with recent marine and river alluvium along the coast and in river tracts (Fookes 1990).

Geomorphology
The topography of the central part of the peninsula is dominated by steep hillslopes, ridges and valleys. Valleys in the mountain region are deeply incised and steeply sided, with a dense, though diminishing, tree cover. The drainage pattern is strongly influenced by a series of NE-SW-orientated parallel mountain ridges (Dixey 1922) and two main NW-SE-orientated valleys, probably associated with major unloading joint systems, and bedrock faults respectively. Rivers and groundwater pathways have exploited the mechanically weaker (more fractured and deeply weathered) materials that occur along these features. Where the rivers reach the coastal platforms or the flat embayments on the peninsula, they become sharply entrenched, but the hard lateritic crust confines the rivers to narrow, steep-sided valleys (Wells 1962). There are several artificial reservoirs (Guma Dam No. 1 and No. 2,Tacugama,Bathurst) that exploit the elevated valleys to supply water and hydropower to Freetown (Williams 1965). Dixon & Robertson (1970) provide a review of the weathered tropical soils encountered during the construction of the Guma Dam.

Climate
Sierra Leone has a humid, tropical climate and a seasonal rainfall pattern that peaks between July and September (McSweeney et al. 2010). Recently, periods of drought have occurred due to the delayed onset of the monsoon rains. When the heavy rain has arrived there has often been extensive flooding (Tarawalli 2012). This rainfall season is largely controlled by the movement of the tropical rain belt, also known as the Inter-Tropical Convergence Zone (ITCZ), which oscillates between the northern and southern tropics over the course of a year and affects Sierra Leone when in its northern position. The ITCZ produces the West African Monsoon, resulting in exceptionally high coastal rainfall during the wet season.

Natural hazards
The geology, topography, and physical location of Freetown make it prone to seasonally recurring natural hazards, chiefly flooding and landslides (Arup et al. 2018). Flooding in Sierra Leone is most common in the rainy season between May and November. While flooding occurs across Sierra Leone, flood events in Freetown can be particularly damaging as the steep terrain with narrow incised gulleys can cause rapid flash flooding. These high energy, rapid flows can cause substantial destruction and pose a significant threat to life.

Landslide hazard
Landslide hazards on the Freetown Peninsula are frequent and landslide hazard is widely recognized by government and, to a lesser extent, by the resident population (Freetown City Council 2014). Landslide types include deep-seated rotational slides, translational slides which typically initiate as debris slides and develop into debris flows, rock falls and single boulder rolls (Cruden & Varnes 1996;Thomas 1998). Some apparent rock outcrops are merely the projecting tip of large residual blocks embedded in the weathered mantle and these can be distinguished from displaced boulders by the orientation of the gabbro layering. Whilst these largely remain in situ, some creep is inevitable on steeper slopes (Thomas 1994).
Geomorphological evidence of old translational slides is widespread on the slopes around Freetown (Thomas 1994). Detachment of the slide mass along the bedrock-soil interface often leaves a relatively fresh bedrock scar, which is recognizable on high-resolution aerial photographs and digital elevation models by absent or sparse forest cover and a dark grey reflective surface. In the 1970s and 1980s, M. Thomas mapped several hundred slide scars ( Fig. 1) in the central highlands from 1:12 500-scale aerial photo interpretation with some field observations to ground truth remote assessments (Thomas 1998).
On the Freetown Peninsula, colluvium-covered slopes tend to fail at angles of >26° (Thomas 1983(Thomas , 1998. The thickness of soil and rock weathering profiles is locally affected by bedrock texture, geochemistry and mechanical discontinuities ( joints, faults and shears) and the degree/depth of laterization and saprolite development. The landslide material tends to consist of debris: a mix of laterite, fine and coarse soils and weathered rock corestones. Chinsman (1977) investigated the geotechnical implications of unfavourable kinematic relationships between bedding (igneous layering), joints/fractures and major faults in NE Freetown. Slopeparallel sheet joints are a commonly observed feature at landslide sites in other tropically weathered igneous terrains, such as in Hong Kong, and impart a structural control on slope instability (Parry 2016). Table 1 lists significant landslide events that have affected Freetown. One notable example is the 11 August 1945 Charlotte Landslide, which occurred in Charlotte Village near Bathurst, in the Orugu Valley. This landslide was triggered following significant rainfall amounting to more than 1000 mm over a period of five days, culminating in c. 400 mm of rainfall in the 24 h preceding failure. The landslide is noted to have resulted in at least 13 fatalities and partially blocked the Orugu Valley, causing flooding.

The Regent-Lumley Disaster
On the morning of Monday 14 August 2017, a major landslide occurred in the Regent ward, Western Area Rural, Freetown. The combined effects of the landslide and flooding resulted in one of the largest natural disasters to affect Sierra Leone in the last century. No natural disaster anywhere in the world claimed more lives in 2017. The event sparked an international response, which included the activation of the International Disaster Charter and a World Bankled Post-Disaster Damage and Loss Assessment (DaLA) (World Bank 2017).
The rock-debris slide in Regent and ensuing debris flow is referred to as the Regent Landslide, whilst the rock-debris slide, debris flow, flooding and ensuing effects were often referred to in

Sequence of events
The following presents a summary of the sequence of events which led to and included the Regent Landslide. The interpretation of events is compiled from eyewitness accounts, photographs taken immediately following the disaster, aerial photography, satellite imagery, and engineering geological and geomorphological fieldwork carried out between August 2017 and April 2018 by Arup and the British Geological Survey (BGS) for the World Bank and United National Office for Project Services (UNOPS). Figure 2 presents an overview of the 6 km long hazard footprint.

Precursors to failure
Precipitation totals in Sierra Leone were above average during the 2017 rainy season. The US National Weather Service's Climate Prediction Center states that 1040 mm of rain fell in Sierra Leone between 1 July 2017 and 14 August 2017three times more than expected for the same period during a typical rainy season. Overnight on Sunday 13 August-Monday 14 August 2017, severe rainfall caused significant flooding in Regent and in the uppermost parts of the Babadorie Valley. The US National Oceanic and Atmospheric Administration (NOAA) Africa Rainfall Climatology Version 2.0 (ARC2) model reported total rainfall anomalies of up to 100 mm more than normal for the period 10-16 August 2017. This equates to almost 200% of the normal rainfall totals for the period. On 14 August 2017 the NOAA ARC2 satellite recorded 25-50 mm rainfall across the area encompassing Freetown (Arup et al. 2018). Accurate rainfall time-series data are not available for the Freetown Peninsula area, making the identification of location-specific values not possible. Flooding in Regent and along the Babadorie Valley at this time was such that some people were already being forced to abandon their properties for their own safety. Many properties in Regent are constructed across or near to natural drainage lines and many have concreted yards and driveways. There is little to no provision for surface water management and hence such surface runoff here is erosive and can be rapid.

Initiation
The Regent Landslide initiated on a north-northeasterly facing slope, at c. 06:50 on Monday 14 August 2017 (Fig. 3). During postfailure inspection, water seepages were observed draining from open joints on the upper parts of the landslide scar. The compound effect of softening of clay-infill within discontinuities, loss of pore suctions due to saturation under an extreme rainfall event, and increased porewater pressures on key joint sets within the gabbro rock mass are considered to be key factors that led the slope to fail. The landslide is understood to have occurred in two phases, as indicated by eyewitness accounts.

First stage translational rock-debris slides
Two clear failure mechanisms can be identified from the upper part of the landslide scar ( Fig. 4): (a) in the upper scar area, the fracture orientation facilitated wedge-type sliding failure mechanisms (recognized in the immediate aftermath of the landslide by Usamah 2017), along weathered joints within the gabbro, which released large angular blocks of variously weathered gabbro (Fig. 5); (b) on the lower part of the upper scar the joint pattern is different, with highly persistent and broadly undulating slopeparallel sheet jointing lined with softened clay infill within partially weathered gabbro being the principal discontinuity set, favouring a more planar translational sliding mechanism.
The exact sequence of the reported two-stage failure remains unclear as it was not caught on camera and visibility was low due to the morning mist. It is difficult to infer which of the above processes occurred first, or indeed if there was some preliminary failure, evidence of which was subsequently eradicated by the catastrophic main failure of the slope. Broadly, the failure mechanism is most accurately characterized as a deep-seated translational rock slide which displaced a mantle of variously weathered gabbro, weathered saprolite and residual soil and trees with it. Unlike many of the historical failures observed on the Freetown Peninsula, the main failure surface was, in places, below the bedrock-soil (regolith) interface, which was a contributing factor to the large volume of debris released by the landslide.
The discovery of in situ remains of a house foundation in the midslope area, below a prominent bench feature located at c. 350 mAMSL (Fig. 6) points to a slope failure contained entirely within the upper slope area, and rules out a basal or toe failure. Quite why the failure broke out at mid-slope and formed a bench feature rather than continuing to some deeper level remains unclear. Broad c. 10 m wavelength undulations with an amplitude of c. 1 m along the controlling surface-parallel gabbro sheet joints may have brought the failure plane closer to the ground surface here, causing it to daylight in the slope. The debris then ran out and stripped off the slope vegetation and topsoils from the lower slope but did not fail through it (Fig. 7).

Second stage transition to channelized debris flow
From about halfway down to the foot of the slope, material was no longer falling and sliding from the hillside by wedge and translational mechanisms, but processes shifted to sliding, rolling and flowing over the ground surface. Once this vast volume of material reached the lower parts of the slope, it mixed with the gathering floodwaters at the foot of the slope. The flooding here would have been locally static, accumulating in the natural topographic depressions in this part of Regent. Some of the floodwaters would have continued to flow rapidly from east to west down the Babadorie Valley and out to sea.
At this point along its path the landslide began to display predominantly flow-like behaviour with debris and large boulders entrained in a dense slurry. Internal deformation (abrasion) within the slurry further disaggregated weak weathered rock particles and water content was sustained by incoming surface water. This sort of debris flow commonly develops when landslide debris mixes with surface runoff (Iverson 1997). The transition from sliding to flow-like behaviour can occur rapidly (Santi et al. 2011).
As the debris flow became channelized it gained mass and greater mobility due to scour erosion of river bank deposits and entrainment of debris picked up from the river bed. The velocity of such debris flows can be as high as 15 m s -1 (Hungr et al. 2001). At this point, the path of the debris flow was controlled by the alignment and confinement of the east-west-trending Babadorie Valley.
Due to the presence of natural pinch-points in the Babadorie Valley, it is probable that the channel may have, at times during the event, become partially blocked or 'choked' by boulders, trees or other entrained debris. The occurrence of temporary dams would have permitted the build-up of material until such a time that the blockage burst suddenly, releasing a further surge of fast-flowing debris along the channel. These natural pinch-points would also have facilitated acceleration of the flow in places. These factors may account for the extent and scale of building damage and devastation as far as 3 km downstream along the channel in Kamayama.

Third stage: debris-laden flooding
The destructive force of the debris flow would have gradually reduced from east to west as its load was deposited, the topographic gradient reduced, and the floodplain widened. By the time the debris flow reached Lumley, it would have resembled a large, relatively fast-flowing flood (compared to what might be a typical seasonal flood in the channel). The floodwaters then flowed out into the
Atlantic Ocean, depositing their remaining load as a sediment plume, which can be clearly seen in post-event satellite imagery (Fig. 2).

Causes and mechanisms
While there were a significant number of properties built into the pre-failure slope profile, and these were generally facilitated by the cutting of a construction platform into the hillslope, it is believed that these works were not a major contributing factor to triggering the landslide. This is because the landslide has liberated much debris of intact, partially weathered gabbro, indicating a deeper failure surface than could be linked to shallow building cuts on the slope. Construction was limited to the lower half of the slope, which is not the source of the majority of the landslide debris as some foundations remain in situ.

Discontinuities within the rock mass
The bedrock geology in the landslide source area is olivine-rich layered gabbro, with layers dipping between 20 and 45°to the SW. The rock mass also contains basalt dykes, some oriented parallel to the slope. The bedrock has been altered in the upper 10 m, producing fresh gabbro core stones typically surrounded by a rind of extremely weak to very strong rock. This rind disaggregates to loose soil with little mechanical agitation. Material between corestones contains concentric layers of strong red iron cements and low strength (stiff ) clay.
The first stage landslide mechanism involved a combination of translational (planar) and wedge sliding along discrete pre-existing bedrock discontinuities, principally weathered sheet joints. The gabbro layering is not a significant structural control on the landslide mechanism on this north-facing slope as the rock is tight along layers which are differentiated by slight mineralogical

439
The 2017 Regent Landslide, Sierra Leone differences (olivine content). The main rupture surface for the Regent Landslide was 5-10 m below ground level, making it a deep-seated movement. This main failure surface follows a persistent undulating joint set with variable dips of between 20 and 45°to the NNE, parallel to the slope face (Fig. 6). This controlling basal discontinuity may have effectively daylighted at the mid-slope, forming the wide bench feature evident in the landslide scar. The joint characteristics largely control rock mass strength in high intact-strength rocks such as gabbro and provide an indication of the weathering processes. The characteristics of the important joint sets were inspected in the main backscarp. The rock materials along the joint surfaces are heavily altered by weathering and display both red and white discoloration. The presence of kaolinite may indicate that the gabbro along these joints was initially altered by hydrothermal flow when the Freetown Layered Complex was emplaced. Subsequent supergene weathering exploited and further weakened these joint planes. The rock joint material is low strength, such that 1-20 mm thick slabs can be readily dislodged from exposed surfaces by hand, and easily broken by heavy hand pressure and disaggregated to dust between thumb and forefinger. Joints located 5-10 m below the pre-slide ground surface are filled with clay (Fig. 8) and several types of secondary mineralization occur as soft white kaolin and hard red iron oxide. The geological origin and history of all the fracture sets and their infill materials is unclear and the role of previous phases of slope movement, or possible tectonic and/or hydrothermal activity could be important for engineering geology across the Freetown Peninsula.

Engineering properties and behaviour
Soil classification and mineralogical testing was carried out on bulk samples collected from the debris of the Regent Landslide at the BGS laboratories in Nottingham. Initial particle size distribution analyses on the landslide debris (matrix, excluding cobbles and boulders) returned 31% gravel, 36% sand, 19% silt and 14% clay. Particle size distribution analyses were done to British Standards (BS1377:1990) and with X-ray sedigraph for the fine fraction. Given the 33% fines content the landslide debris should resemble a fine-grained, well-graded engineering soil, but inclusion of extremely strong gabbro boulders and cobbles affects its fieldscale properties. The accumulated soil has reasonably poor drainage properties and is prone to development of excess porewater pressures when saturated and sheared (undrained loading), explaining the 'quicksand-like behaviour' described by local geologists on the ground in the hours after the landslide.
Mineralogical testing of the clay fraction of the landslide debris by XRD identified 14% halloysite group clays with natural moisture contents of 29-31%. Halloysite clays have a hollow, needle-like morphology that is microstructurally metastable and linked with 'collapsible soil' behaviour, whereby saturated soils can yield very suddenly when rapidly loaded (Moon 2016). This may account for some aspect of the apparent rapid loss of shear strength that occurred along surface-parallel joints during the first stage translational and wedge failures and the second stage transition to flowing behaviour of the slide mass. These values should be treated with due caution as the data are based on only a very small number of samples. Furthermore, hydrated halloysite clay morphologies are highly sensitive to environmental and storage conditions that can occur between the field (Sierra Leone) and laboratory (UK). Other thick clay seams were identified in the bedrock 2 km to the west of the Regent Landslide in the Babadorie Valley. Atterberg Limit tests found these to be high to very high plasticity clays with natural moisture content of 48-55%. XRD returned 28% halloysitic clays. Further XRD, SEM and geotechnical characterization work is being carried out by BGS to further ascertain the material properties and behaviour.
The identification of clay-filled fractures in the gabbro host rock is highly significant to slope stability and ground engineering projects on the Freetown Peninsula, due to the marked reduction in shear strength and rock mass strength these features introduce. Local sources suggest the nearby Sugar Loaf Mountain gets its name from a colloquial understanding of this problematic geotechnical behaviour, whereby hard granular soils rapidly disintegrate to a 'syrup-like slurry' on wetting.

Deforestation
Comparison of 2005 and 2017 Google Earth imagery highlights the marked reduction in the density of slope vegetation in Regent. Burnt tree stumps and long grass were observed at the top of the ridge behind the rear scarp, indicating that the ridge slope has undergone recent deforestation. Significant deforestation is also apparent on the southerly side of the slope behind the Regent Landslide (Usamah 2017). Deforestation increased the susceptibility of the slope to failure in two ways: firstly, by reducing the canopy protection and water-uptake associated with dense tropical forest cover, meaning increased seepage of rainwater into the ground; and secondly, by removing the mechanical stabilizing effect offered by tree roots, which can be particularly beneficial for slope stability in weathered tropical soils. The stabilizing effect of tree roots is evidenced by the retention of a large block of ground on the eastern flank of the landslide scar, which coincides with the elevation at which dense forest cover is present on the adjacent hillslope. This area is referred to as 'The Return' (marked on Figs 3 and 4).

441
The 2017 Regent Landslide, Sierra Leone Despite the obvious negative effects of deforestation on slope stability at the Regent Landslide, it should be noted that several other smaller landslides also occurred at around the same time in forested areas. One of these was located on the back-slope of the Regent Landslide and is shown schematically on Figure 7. Historically, landslides have occurred across the Freetown Peninsula in areas unaffected by deforestation (Table 1; Fig. 1). This serves to illustrate that landslides are a natural part of terrain evolution on the Freetown Peninsula and that, although more recent deforestation unquestionably has a negative effect on slope stability, it is not the only factor that increases the landslide susceptibility. The answer to the question of whether the Regent Landslide would have occurred were it not for deforestation on the steep hillslopes near Regent is not a straightforward one. Quantitative investigation of this potential effect by experiment or test, and process modelling, is necessary to check whether this assessment is correct or not.

Impacts and losses
The Regent-Lumley Disaster claimed the lives of c. 1100 people and directly affected more than 5000 people. Those who survived the disaster were, in many cases, left without family, homes, possessions, savings, transport, jobs, schools and general livelihoods.
The impacts of the disaster varied along the length of the Babadorie Valley. In Regent, despite larger and apparently more structurally robust buildings, the number of fatalities and proportion of buildings destroyed was highest. This can be attributed to the devastating high-energy nature of the landslide that affected this area. To the west, although further from the source of the landslide, damage to properties in the channel near Kamayama was still extensive (Fig. 9), which demonstrates the debris-laden nature of the flooding here. In Lumley, 6 km from the Regent Landslide, the effects of the disaster were still significant due to increased vulnerability of the population in these areas, many of whom live in informal buildings.
In total, the disaster was assessed to have caused c. $US31 M damage and loss (World Bank 2017), the majority of which was in the housing sector (c. $US14 M) followed by both social protection and healthcare sectors (each c. $US5 M). It has been further estimated that a total of c. $US82 M is required to support short-, medium-and long-term recovery efforts following the Regent-Lumley Disaster.

Rehabilitation
Rehabilitation works at the site of the Regent Landslide have been ongoing since April 2018 and have largely comprised the regrading and management of the slopes and watercourses at the foot of the Regent Landslide in Regent (Fig. 10). Moving forward, landslide risk in this area will be managed by reducing exposure by preventing rebuilding and resettlement in the Babadorie Valley, much of which, including the affected areas in Regent, is already designated as a National Forest Reserve. Managing risk by more traditional hard-engineering approaches, which typically seek to reduce hazard (by rock-bolting or netting etc.), was not appropriate given the urgent need to implement risk-reduction measures at the site prior to the 2018 rainy season and other logistical and financial considerations. The site has also been much re-planted which will act to both deter resettlement and promote the local stability of the reprofiled channels. Much of this rehabilitation work and replanting was done by the local population, many of whom were affected by the disaster.

Conclusions
The Regent-Lumley Disaster comprised a three-stage hazard event: first stage translational failures through partially weathered gabbro in mountainous uplands; second stage transition to channelized debris flows on the lower slopes above partially flooded valleys; and third stage debris-laden channelized flooding. The large volume of material involved may be attributed to a deeper-seated failure through partially weathered gabbro rather than at the soil-bedrock interface as has historically been observed in Freetown. This mechanism liberated extremely large mega-boulders that contributed to the destructive force of the first and second stage processes. The transition to debris-flowing may have been further encouraged by the presence of highly collapsible halloysitic clays; however, this remains the subject of further investigation. Downstream from the Regent Landslide, the energy of the flow may have been maintained and surged by the formation of temporary debris-dams at natural pinch-points in the river valley. This event shares similarities with the 1945 Charlotte Landslide, which resulted in 13 fatalities. Perhaps the most striking difference between the two events is not the nature of the hazard, but the increase in population exposure within the landslide footprint due to urbanization and population growth. Looking forward, climate change may well bring more intense seasonal rainfall patterns, and population exposure is set to rise in Freetown, increasing landslide risk further. Future work must focus on the implementation of practical, effective and implementable risk-reduction strategies across the Freetown Peninsula, including improved understanding of how and where fractures affect slope stability on natural and artificial slopes.