On 14th August 2021, at 0829 local time, a shallow earthquake of magnitude 7.2 struck the Tiburon Peninsula in the Caribbean nation of Haiti, approximately 150 km west of the capital Port-au-Prince. The main earthquake was followed by numerous aftershocks, with the largest to date being a magnitude 5.7 on 15 August 2021. Several thousand landslides (Martinez et al., 2021) were also triggered, by both the earthquake and subsequent rainfall from Tropical Cyclone Grace, which hit Haiti on 16 August 2021. The earthquake is estimated to have killed over 2,000 people, injured over 15,000, and damaged and destroyed 137,000 houses, affecting over 800,000 people in total (UN Office for the Coordination of Humanitarian Affairs (UN OCHA), 2021). According to the Post-Disaster Needs Assessment (PDNA), total damages and loss are estimated at US$ 1.62 billion, while recovery needs are estimated at US$ 1.98 billion.

The socio-economic and political system in Haiti is extremely complex (Oliver-Smith, 2010). Haiti is one of the poorest countries in the world, ranked 168 out of 187 on the human development index (UNDP, 2015) with almost 25% of the population living in extreme poverty on less than $1.25 per day (UNDP, 2013). At the time of the 2021 earthquake, Haiti was still recovering from the devastating earthquake of 2010, which killed over 150,000 and left 1.5 million homeless. The 2010 earthquake led to significant migration from the east to the west of Haiti, leading to a change in vulnerability and susceptibility in these areas to future earthquakes and other hazards such as Hurricane Matthew, which hit Haiti and the Tiburon Peninsula in 2016. Figure 1 shows the change in population between 2010 and 2021 in Les Cayes, a city in the southwest of Haiti. At the time of the 2021 earthquake, the Haitian President, Jovenel Moïse, had recently been assassinated, leading to significant instability, with relief efforts interrupted due to increased gang violence and insecurity.

The nature of structural damage and geotechnical failures after an earthquake are perishable data, yet provide crucial information to inform future disaster risk reduction (e.g. Bray et al., 2018). Normally such data would be collected through in-person reconnaissance missions. However, the instability and insecurity discussed above, combined with Haiti being on the UK government “red list” for Covid-19 and the UK Foreign and Commonwealth Office’s “do not travel” list, both at the time of the earthquake and in the immediate aftermath, prevented undertaking an in-person mission from the United Kingdom. Instead, a remote (or virtual) mission was the most appropriate approach to investigate the impact of the disaster. This paper discusses the team involved and motivations behind undertaking such a mission (Section 1), summarises the mission’s findings in terms of seismotectonics (Section 2), remote sensing (Section 3) and building damage assessment (Section 4), before discussing the main lessons learnt for undertaking hybrid post-earthquake reconnaissance.

The seismotectonics of Haiti, and Hispaniola, of which Haiti forms the western part, are controlled by the east-north eastwards convergence of the Caribbean plate, to the south, with the North American plate, to the north (Figure 3). The plates are converging at about 20 mm/yr (DeMets, 2001; DeMets et al., 2010). Moving from east to west along the plate boundary the convergence becomes progressively more oblique, transitioning from subduction sub-perpendicular to the plate boundary in the lesser Antilles, to highly oblique convergence in Hispaniola (Calais et al., 2016). This oblique motion is accommodated across a series of microplates and deformation zones, with the compressional component predominantly taken up by offshore thrust faults to the north and south of Hispaniola. The strike-slip component of the motion is taken up on two left-lateral strike-slip faults which dominate the geomorphology of Haiti; the Septentrional fault to the north, and the Enriquillo-Plantain-Garden fault (EPGF) to the south. The EPGF strikes east-west along Haiti’s southern peninsula and has been proposed to connect offshore with strike-slip faults in Jamaica. This fault is thought to have hosted a series of major historic earthquakes in the late 18th and early 19th centuries, with intensity magnitudes estimated between 6.6 and 7.5 (Bakun, Flores, ten Brink, 2012). Mann et al. (1995) interpreted the EPGF as a sub-vertical fault accommodating pure left-lateral strike slip, which is still the interpretation used in the Haitian seismic hazard map (Frankel et al., 2011) but has been questioned based on observations from a magnitude 7 earthquake which occurred on 12th January 2010, as discussed below.

The epicentre of the 14th August 2021 earthquake was almost directly on the mapped trace of the EPGF near the town of L’Asile (Figure 4), with a shallow depth of ∼10 km (USGS, 2021). However, initial seismological results demonstrated that the earthquake was not a pure strike-slip event. Instead, about 17% of the movement was compressional and the earthquake occurred on a dipping plane (USGS, 2021). InSAR-derived ground surface displacements (discussed further in section 3) are consistent with the earthquake slip having a vertical component, and finite-fault inversions (Raimbault et al., 2021; USGS, 2021) demonstrate that the fault plane likely dipped to the north and was oblique, with both left-lateral and compressional slip. The fault dip is important, both because of its implications for the seismotectonics of Haiti, and because having a dipping fault means that the Boore-Joyner distances (the shortest distance between a building and the surface projection of the earthquake rupture, and a key parameter in ground motion prediction equations) will decay more slowly away from the epicentre than for a vertical fault i.e. that buildings might be expected to experience stronger ground motions over a wider area.

Limited on-the-ground access has made mapping and identification of surface ruptures challenging. Although many of the citizen science seismometers in the Ayiti-seismes, 2022 network (Calais et al., 2020) were not operational at the time of the earthquake, a single raspberry shake R50D4 provided a record of strong ground motions ∼15 km from the earthquake (on the edge of the seismic code’s definition of “near-field”; Ministère des Travaux Publics, Transports et Communications (MTPTC), 2012), and the station NQUSE in the American embassy provided an additional in-country record of strong ground motion (Courboulex et al., 2021). The availability of strong ground motion records is in stark contrast to the last major earthquake in Haiti, which occurred on 10 January 2010. At the time of that magnitude 7 earthquake, there were no strong ground motion instruments in Haiti, so ground motions had to be inferred based on building damage and aftershocks (Hough et al., 2010).

The magnitude of the 2021 earthquake was 7.2, meaning that it released about twice as much energy as that in 2010. However, the epicentre of the 2021 earthquake was ∼100 km west of that in 2010 (Figure 4), away from the major population centre in Port-au-Prince, which meant that the death toll was almost two orders of magnitude lower. Similar to the 2021 earthquake, the 2010 earthquake was initially thought to have occurred on the EPGF based on its epicentral location (e.g. Hayes et al., 2010). However, more detailed seismological and geodetic analysis demonstrated that the earthquake actually ruptured a previously unmapped fault (the Léogâne fault; Calais et al., 2010) and had a significant thrust component (Calais et al., 2010; Hayes et al., 2010). Subsequent GPS data have demonstrated compressional strain accumulation perpendicular to the eastern end of the EPGF but concluded that the fault-perpendicular motion was negligible further west, including in the region which ruptured in 2021 (Symithe & Calais, 2016; Saint Fleur, Klinger, Feuillet, 2020). The occurrence and nature of the 2021 earthquake, therefore, reinforces the message of the 2010 earthquake that the faulting in Haiti is more complex than has historically been assumed and that further work is needed to fully understand the seismic hazard in this region.

As well as highlighting previously unmapped faults, the 2010 earthquake led to the creation of new seismic hazard maps for Haiti (Frankel et al., 2011) as well as the creation of a national annexe to ASCE/SCI 7-05, 2005; (MTPTC, 2012); the spectra for the maximum credible earthquake and design spectra in this annex are shown in Figure 6). However, these maps were still dominated by strike-slip motion on the assumed vertical EPGF. Two other major caveats are: 1) that publicly available Vs30 measurements for most of Haiti (shear wave velocity at 30 m depth, a key control on ground motions) are based on topographic gradients (Wald & Allen, 2007 cited in; Frankel et al., 2011) rather than in-situ measurements and 2) that there are no specific ground motion prediction equations for Haiti. One of our aims for this mission, therefore, was to compare the recorded ground motions to the seismic code.

Peak ground accelerations at R50D4 and NQUSE were 0.36 and 0.023 g respectively (Figure 4C,E, compared to Figure 5 modelled peak ground accelerations). NQUSE is sited in the US embassy in Port-au-Prince, more than 100 km from the epicentre. Since the southern peninsula of Haiti is very narrow (<15 km wide in places) almost the whole width of the peninsula within the rupture region is near-field (Figure 4). Figure 6 therefore compares the spectral accelerations for near-field ground shaking for the maximum credible earthquake, and the design level in the current code to those observed at R50D4. At periods less than about 1 s, which are of relevance for Haiti since most buildings have 1 or 2 storeys, the accelerations from the 2021 earthquake are less than those for the maximum credible earthquake. However, they exceed the design level, meaning that even buildings built to code, which many are not, might be expected to have experienced plastic deformation. Although no single earthquake can invalidate a code based on a particular return period, the fact that the design accelerations have been exceeded less than a decade after the code’s introduction suggests further work is required to fully explore and understand seismic hazard in Haiti.

Earth-Observation satellites can provide frequent imagery of very large areas worldwide, enabling the observation of locations that are difficult or dangerous to access with traditional survey methods (Milillo et al., 2016), and providing a wide-angle view of major disasters and their impact on the built environment. Satellite remote-sensing data is typically released between a few hours and few days after an earthquake (Yun et al., 2015), and can provide sub-centimetre accuracy (Bürgmann et al., 2000), sub-metre resolution and daily-to-weekly revisit times. Such data enables a close-up view of buildings and structures while capturing large-scale movements and providing a broad overview of the extent of damage caused by the disaster. In the context of post-earthquake reconnaissance, the most frequently used space-borne remote sensing techniques are Synthetic Aperture Radar Interferometry (InSAR) and Optical/Multispectral imagery.

InSAR uses radar images acquired by satellites to measure deformation of the Earth’s surface. The satellite radar antenna emits an electromagnetic signal in the microwave frequency band toward the ground surface and receives signals scattered back from the Earth to the satellite (Bamler, 2000). The backscattered signal is used to generate a SAR image which contains information on the phase and amplitude of the signals. As the phase is recorded as a function of the satellite positions and image acquisition time, the phase difference between two radar acquisitions can be used to estimate ground deformations along the look direction (LOS) of the satellite radar antenna (Bürgmann et al., 2000). Pre- and post-event SAR images can be used to estimate earthquake-induced ground deformation (Stramondo et al., 2005; Chini et al., 2008). InSAR deformation measurements can also be used to identify relationships between co-seismic ground deformations and building damage (Yun et al., 2015; Barba-Sevilla et al., 2018; Li et al., 2021).

Optical/multispectral images are collected by satellite sensors operating in the optical part of the electromagnetic spectrum, which includes visible, near-infrared and short-wave infrared wavelengths. Thanks to the use of optical wavebands, optical/multispectral imagery is very similar to common photographs, making its interpretation easier than InSAR data. However, in contrast to InSAR techniques, optical/multispectral sensors cannot penetrate dense clouds. In geographic regions with frequent cloud coverage, this can result in limited observation capabilities. In the context of post-earthquake assessment, a wealth of literature has investigated the possibility of using optical imagery to assess structural damage based on the extraction of building characteristics, such as texture, geometric shape, and spectral reflectance (Anniballe et al., 2018; Ji et al., 2018), or to detect secondary-induced hazards, such as landslides (Zekkos et al., 2017).

Two main types of methods are traditionally used to analyse optical data: visual interpretation and image classification techniques (Bai et al., 2018; Duarte et al., 2018; Rathje et al., 2005). Visual interpretation can involve a manual superimposition of pre- and post-earthquake images to detect damaged structures, or the evaluation of only post-event images to map the occurrence of secondary hazards, such as landslides. Image-enhancement techniques can involve automated change detection between two pre- and post-event images, or automated classification, such as Convolutions Neural Networks (CNN) and Normalised Difference Vegetation Index (NDVI), of post-event data.

The following two sections show the results of using InSAR and optical/multispectral images for the characterization of co-seismic displacements and the identification of landslides caused by the 2021 magnitude 7.2 Haiti earthquake.

This section illustrates the results of the remote damage assessment we carried out following the 14th August 2021 earthquake. In total, an estimated 137,500 buildings were damaged or destroyed as a result of the earthquake (Haiti: Earthquake Flash Update, 2021). The EEFIT team worked with the dataset provided by the Structural Extreme Events Reconnaissance (StEER), who coordinated the collection of building damage descriptions and photographs by local Haitians (Kijewski-Correa et al., 2021). This dataset was paramount to understanding the structural damage distribution. The database includes a map of Haiti with the recorded buildings and their assigned damage level (Figure 10A). The EEFIT team surveyed a total of 2062 out of 11,669 buildings (data available in the Fulcrum App), and grouped them in different functional categories: schools, hospitals, churches and residential buildings. A total of 836 schools, 78 hospitals, 319 churches and 829 other buildings were assessed. The following structural classification was used for each assessed building: Reinforced Concrete with infill masonry shear walls (RC), Confined Masonry (CM), Unreinforced Masonry bearing walls (URM), Reinforced Masonry bearing walls (RM), Wood Light frames (WL), Wood with Stone infills (WS), and Unknown (UN). The damage level for each case-study building was assigned according to a rating system comparable to the European Macroseismic Scale (EMS-98), as defined in the StEER assessment manual (Miranda, 2021). The damage levels are subsequently abbreviated as: DS1 – No Damage, DS2 – Minor Damage, DS3 – Moderate Damage, DS4 – Severe Damage and DS5—Total or Partial Collapse. Figure 11 shows the percentage of each structural typology experienced and the damage distribution for buildings in different functional categories. As an example of the observations resulting from the remote assessment, the rest of this section presents some detailed insight into the assessment performed on schools (Figure 10B).

Among all surveyed schools, 39.5% (330 out of 836) and 26.2% (219) were constructed out of confined masonry and concrete frame with masonry walls, respectively (Figure 11A, left). In terms of damage level, 24.9 and 38.4% of the assessed schools experienced no or minor damage, respectively (Figure 11A, right). However, a significant proportion of schools demonstrated severe damage (11.1%), or partial or total collapse (6.1%). For most building typologies, the greatest percentage of buildings experienced only minor damage. However, the most common damage level for schools made of wood light frames (WL) was no damage, and that for wooden frames with stone infill (WS) was severe damage. Damage levels by building typology are presented in Figure 12.

In-plane and out-of-plane masonry failure were the most common failure mechanisms for the damaged schools. Vertical cracks were evident along the corners and walls of the buildings, indicating out-of-plane material failure. Horizontal and diagonal “x” shaped cracks were also present, typically around openings and along walls in-plane with the earthquake slip direction. Whilst these failure mechanisms were primarily evident in the masonry and concrete structures, failure of the connections, either from poor material strength, detailing or workmanship was evident in all building typologies. On many occasions, openings along the perimeter walls provided areas of high stress concentration where the seismic load had no clear load path to the wall and ground below. Figures 13C–F show the typical damage mechanisms experienced by the schools graded from moderate to total collapse.

Schools which performed well had regular plans and provided adequate strength and ductility. Light timber roof structures performed particularly well, experiencing little or no damage. Strong foundations and sufficient connections between the shear walls and supporting corner columns acted to dissipate the seismic forces effectively and thus reduced the damage experienced by these schools. Evidence of this is presented in Figures 13A–B.

With the ongoing discussions around climate emergency and sustainability, there have been advocates for the review of in-person missions, versus utilising remotely-collected data to undertake assessments and evaluate the impact of disasters such as the Haiti Earthquake (Whitworth M. R. et al, 2020). The aim of this paper is to present the 2021 EEFIT Haiti mission in terms of the lessons it held for conducting remote missions. We now outline some of our observations about the (dis)advantages of carrying out such post-earthquake reconnaissance remotely. In doing so we note that some of these observations are likely to be mission-specific, depending on the geographic setting, team membership and pre-existing connections to the affected community inter alia. We nonetheless hope that these insights will be instructive for future missions, remote or otherwise.

Although the mission was remote, in the sense that the team didn’t travel to Haiti, it relied heavily on data collected in person. The StEER dataset which underpinned our building damage assessments was collected by Haitian volunteers, mostly non-engineers. Their pictures, notes and recordings were critical to our assessments of building damage, particularly for identifying minor and moderate damage, such as cracking, which cannot be detected remotely. It is therefore important to recognise that this mission was not truly remote, and that, at least at the moment, conducting a genuinely remote mission without the involvement, hard work and commitment of locals, often working under challenging conditions, is not possible. Calling such missions remote has the potential to privilege work done by trained engineers, and more generally workers from the Global North, over that carried out by local communities. It is important that future teams conducting “remote” missions recognise and consider the hierarchies of knowledge implicit in their choice of terminology.

Local reporting also provided the only human connection EEFIT team members had to the disaster. As well as showing the damage to buildings, many of the audio recordings gathered made clear the trauma of the loss of homes and livelihoods, which extends far beyond structural damage. Showing damaged buildings as coloured dots on a map, without the experience of going to, and assessing, the situation in person has the potential to disconnect built infrastructure from the human experience it represents. However, the pictures which Haitian volunteers had collected did bring this experience into focus. In many cases, it was challenging to distinguish whether features of a building, such as tarpaulined walls, represented a temporary repair to earthquake damage or pre-earthquake living conditions. Such challenges highlighted to team members the idea, which has long been recognised in disaster studies, that the conditions of vulnerability which enable a hazard, such as an earthquake, to become a disaster, also mean that the normal living conditions of vulnerable people are frequently indistinguishable from what might be identified as a disaster in another context (e.g. pp.10, Hewitt, 1983; Aronsson-Storrier and Dahlberg, 2021).

We expect that this understanding, of the socio-cultural context of the disaster, would have been more evident had the team been in Haiti. Similarly, though the StEER dataset provided an invaluable resource, there were frustrations in using it associated with not having collected the data ourselves (such as pictures taken at awkward angles or not capturing key structural elements). Perhaps more importantly, however, it was hard to recognise the challenges which the people capturing the data had faced in doing so, both as non-engineers and due to the unstable political situation in Haiti at the time.

A lack of social relations and human connections to Haiti was a more generalised problem of the mission. Not being in-country made it harder to connect directly with appropriate stakeholders, making it more difficult to ensure that the lessons learned have tangible benefits for the affected communities. Local information and knowledge about the impacts of the earthquake, and locally specific factors which might have contributed to building damage, were also missing from a remote mission. These difficulties are partly mission-specific. The political situation in Haiti at the time and EEFIT’s lack of pre-existing in-country contacts meant that it was challenging to build connections and talk to people with appropriate local knowledge, irrespective of the team’s physical location. However, we think that being present in-country would have provided more opportunities for developing such connections and directly interacting with regional stakeholders. This lack of connection also impacted our assessment and understanding of WS4 Disaster Response and Recovery (Figure 2). It is important to note that whether in person or remote, field missions need to be carefully planned in order not to impact ongoing relief and recovery efforts. The status of the relief and recovery effort at the time of the mission may have impacted people’s availability to engage.

There were also limitations to the team being remote in terms of our assessment of the geophysical impacts of the earthquake. InSAR and seismological data are now routinely available shortly after an earthquake, along with models of how and when the causative fault slipped during an earthquake (e.g. USGS, 2021). These data provided a useful guide to understanding the physical properties of the earthquake, its seismotectonic implications (section 2) and the associated large-scale ground motions. Recent studies have also demonstrated the possibility of using optical image correlation to map surface ruptures at scales as small as ∼20 cm (Milliner & Donnellan, 2020). However, in practice, smaller-scale features, such as surface ruptures, remain challenging to map remotely due to the limited availability of affordable high-resolution optical imagery. It may also be difficult to interpret such features without an on-the-ground context. For instance, even for the Ridgecrest earthquake (2019) many of the identified surface features might have been related to cracking rather than the direct surface expression of faulting (Milliner and Donnellan, 2020). These limitations aside, the ready availability of remote sensing data, and its utility in providing a first-order overview of what has happened in an earthquake, leads us to suggest that the type of remote sensing analysis conducted in this mission might provide a useful pre-mission exercise for future teams who will travel to an earthquake-affected region. Such pre-mission remote sensing might also allow areas of particular interest or concern to be identified in advance, and give team members a broader context for the, inevitably, smaller region it is possible to visit whilst in country.

Remote sensing data has long been used to map landslides and understand triggers, mechanisms and susceptibility factors (Jones et al., 2019; Jones et al., 2021). However, these approaches rely on 2D-surface mapping, rather than understanding the phenomena in 3D, by determining depth of failure and volumes. Zekkos et al. (2017) highlight the benefit of field mapping to understand aspects such as depth of failure and entrainment. Similar issues with using remote sensing data on both a large and smaller scale were evident in the Haiti mission, further compounded by cloud cover limiting the identification of landslides and the automation of the identification process. However, there is still significant benefit to rapid assessments to support the recovery and relief efforts.

The larger team for this “remote” mission relative to previous in-country missions was another benefit which might provide a useful addition to future missions. One of EEFIT’s purposes is to train earthquake engineers in building damage assessment. However, for an in-country mission numbers are limited by funding and potentially ethical concerns about bringing relatively inexperienced engineers to a region which has recently experienced a disaster. This mission was able to include a relatively large team and thus to give more people an opportunity to learn about post-earthquake reconnaissance.

We suggest that future missions could benefit from a hybrid approach where some team members provide remote support, whilst a smaller group travel to the affected area. This approach would allow EEFIT to fulfil its aims of training and to capture the large-scale regional context which remote sensing data allows, whilst also having the benefits of an in-person mission described above.

We have presented preliminary results and observations of the Earthquake Engineering Field Investigation Team remote mission to Haiti, following the August 2021 earthquake. The aims of the mission were to use openly available data to understand the pre-existing tectonics, and the deformation and motions which occurred during the earthquake, and to carry out damage assessments on affected buildings in the region. These damage assessments were only possible because of data collected locally by Haitians, often under challenging conditions. Although important insights can be gained into the large-scale deformation and regions of potential damage using only remote sensing data, we conclude that missions investigating structural damage cannot be truly remote at present, and should acknowledge the role of local stakeholders. We also conclude that the connection to the human aspects of disaster is poorer for lack of in-person experience. However, hybrid missions have the potential to reduce environmental and local impacts, expand the number of engineers who can be trained in damage assessment and also allow for the in-person engagement with local stakeholders, which is crucial to ensuring that the results of post-earthquake reconnaissance missions contribute to disaster risk reduction.