Bridging the Seismic Vulnerability Data Gap Through UAV and 360° Imagery: The Case of Nejapa, El Salvador

Featured Application

This article presents a remote sensing application using UAV and 360° imagery to build a 3D database of seismic exposure and vulnerability in Nejapa, El Salvador. By combining these techniques with census and land property data currently being collected by the government, the study shows how nationwide replication could position El Salvador as a regional leader in risk management.

Abstract

In Latin America, high seismic activity drives countries to develop disaster risk reduction policies based on seismic risk studies. This work demonstrates the feasibility of creating a seismic exposure and vulnerability database using remotely sensed data. In Nejapa, El Salvador, a drone flight and 360° photo capture were conducted to generate a 3D model of the city. Buildings were identified, characterised, and assigned a vulnerability model. This database was used to estimate seismic risk for a simulated Mw 6.7 earthquake on the Guaycume fault near the city. Results show that 71% of buildings would suffer complete damage and 68% of the population would be homeless, with losses exceeding USD 15 million. Findings were shared with relevant institutions in El Salvador through a dashboard. The country is currently collecting the same type of data used in the present study to update its cadastre and census. This is an opportunity to replicate this pilot experience in many other cities across the country and to provide open data access, positioning El Salvador at the forefront of civil protection in the Latin American region.

1. Introduction

Disasters related to natural hazards caused 86,473 fatalities in 2023, and losses reached USD 202.7 billion [1]. In particular, earthquakes affect dozens of countries around the world and are responsible for the majority of these fatalities (62,451). Policies to mitigate adverse effects of this type of catastrophe need to be supported by seismic risk studies, which estimate the damage that future earthquakes may cause in a region or city, as well as the human and economic losses associated with this damage. These studies require a large amount of data on the seismic sources and on the exposed elements (mainly housing, infrastructure, and population). The exposed assets must also be characterised by their seismic vulnerability and value.

Many countries have a long tradition of generating official digital mapping at different scales and have been working for years on the development of spatial data sharing policies. These countries allow free access to reliable and standardised official census and cadastral information, which is very useful for undertaking seismic risk studies (e.g., Reference [2] in Spain; Reference [3] in Portugal; References [4,5] in France; Reference [6] in Nepal; Reference [7] in Japan). There are also several initiatives to develop taxonomies and exposure databases related to earthquakes and other natural hazards at a national level [8,9,10,11,12], and therefore with a resolution that can be improved at the local level. However, even today, many countries lack such data or free distribution policies, making it very difficult for analysts to conduct risk studies.

In the specific case of Latin America, a group of scientists from the geological and mining services of eight countries in Central and South America, Spain, and Portugal [13] recently stated that all countries in the region have reported dramatic catastrophes in the last ten years. Latin America is a territory exposed to natural hazards, with floods and earthquakes causing the most damage. In that time, more than 11,000 lives have been lost and costs of some USD 50 billion have been incurred. In their report, the authors of Reference [13] explain that although most geological surveys have thematic cartography of their different geological hazards, there is a lack of detailed vulnerability studies that could be included in urban development plans. They recommend that geological surveys make an effort to transfer knowledge, favouring open access to their data, so that they can be used in disaster risk reduction (DRR)-oriented studies. The same would apply to other regions of the world.

Reference [14] analysed open data and open governance policies in the field of disaster risk reduction (DRR) in Japan. They note that data, once created, should be open access and well documented. Even if the custodians are local entities, they should be disseminated in such a way that they can serve different user profiles. They also reflect on the importance of synchronising the rapid advance of data capture technologies with the generation of derived products and policies to make them open access.

It is increasingly common to find risk studies in which, in the absence of government data, alternative ways of obtaining exposure and vulnerability information are being considered. In some cases, it is the community itself that generates the necessary data through crowdsourcing activities, such as Volunteered Geographic Information (VGI). This is a very interesting way of collecting geographic information, as a double goal is reached: (1) it is a way of making the population aware of the danger to which they are exposed, and (2) the data generated are specific to the intended application and therefore of high quality. To achieve this, different approaches are used. For instance, specific forms can be used to collect building data for seismic vulnerability analysis [15], or three-dimensional analogue models of hazard, exposure, and vulnerability can be created and the risk study completed [16].

An example of a globally successful VGI is OpenStreetMap (OSM). It is a freely accessible online platform that allows the creation and download of detailed urban cartography (buildings, streets, shops, points of interest, etc.). OSM has made it possible to carry out numerous risk studies, such as those of [17], dedicated to industrial buildings; References [18,19] for residential buildings in Valparaíso, Chile; or Reference [20] in Vrancea, Romania. Sometimes, it is necessary to integrate several sources of VGI, as in the study by [21], who used OSM and an online database created by volunteers in the city of Cologne, Germany, which collects the year of construction of the building. Reference [22] built an exposure database after a field campaign to complete an ad hoc form designed by stakeholders and implemented on mobile devices. Furthermore, OSM is used to integrate building-level attributes to the GED4ALL Global Exposure Database for Multi-Hazard Risk Analysis [9].

An alternative approach to generate exposure and vulnerability databases when governmental data are not available or accessible is the use of modern geomatics technologies. Among these, unmanned aerial vehicle (UAV) data collection in urban environments is becoming increasingly widespread, as UAVs are becoming more affordable and easier to use. They are equipped with sensors that record images and allow the creation of orthophotos, digital terrain models (DTMs), and digital surface models (DSMs). These data are very useful in identifying the most relevant elements of built-up areas, such as buildings (e.g., [23,24]), vegetation, or vacant parcels [25]. In addition, the building geometry can be enriched with semantic attributes related to their seismic vulnerability, such as area, number of floors, or construction materials.

In this context, this study presents a viable option to obtain information on exposure and vulnerability in unmapped urban areas, which can be useful in risk studies. Taking into consideration the report by [13], focused on Latin America, we have taken Nejapa, in El Salvador, as a pilot city in which to carry out a rapid data collection survey to generate a seismic exposure and vulnerability database, with an application in seismic risk assessment.

El Salvador is a country subjected to high tectonic deformation, which leads to high seismic hazard [26]. Most of the population and economic activity are concentrated in densely populated cities that are often located very close to faults with a high seismic potential. Such is the case with Nejapa, a city located a few kilometres from the Guaycume fault, an active structure capable of generating earthquakes of up to Mw 6.8 [27,28].

Like many other countries, El Salvador has high-quality official digital cartography, but, to date, this is not available to the general public. Access to information is mainly limited to government entities with specific agreements. Most geospatial data involve usage restrictions and expensive prices. To use cadastral or census data, a bureaucratic procedure is necessary, which requires a monetary payment and usually takes longer than the time available to carry out the study of interest.

The country has a photogrammetric flight that can be consulted through a web viewer, but it does not offer the possibility of downloading photograms or orthophotos. Spatial data layers can sometimes be accessed through personal contacts in research institutions, but they are often outdated or lack metadata, which makes them difficult to use. In terms of open access information sources in Nejapa, Google Earth imagery was of low resolution at the time of this study. Google StreetMap does not exist in this area, and OpenStreetMap is not entirely reliable or does not provide the detail needed to build a building-by-building database.

El Salvador has recently started a series of activities aimed at updating the national cartography and cadastral system. A high photogrammetric flight is being carried out for the whole country, with a resolution of 20 cm and recording in both the visible and the near-infrared ranges of the spectrum. In addition, a low flight is being carried out for the departmental capitals, with a resolution of 8 cm and LiDAR capture; and 360° images are being taken in different streets of its territory. The orthophoto mosaic can be viewed in a web visualiser [29]. As for the census, in May 2024, data collection started for its update, which could be ready in 2025. It is therefore expected that the country will have modernised its data and its census and cadastre system by the near future; however, there is no news about possible open publication of these data.

The objective of this study is to offer an alternative to the current situation of lack of access to data that makes it possible to create a database of seismic exposure and vulnerability by remote sensing and which is of sufficient quality to be used in a seismic risk study aimed at proposing DRR measures. For this purpose, a UAV flight was conducted and 360° photographs were taken in the historic centre of Nejapa. With these data, an orthophoto, a DTM, an SDM, and a 3D model of the city were generated. This was the basis for identifying and characterising the buildings and assigning them a vulnerability model. In the analysis of the images, both manual techniques and machine learning were used to classify objects. The exposure and vulnerability database generated was used in a seismic risk estimation, simulating an earthquake on the Guaycume fault. Finally, a dashboard has been created that could be a platform for sharing the data generated in this study, as well as their results, both with governmental entities and with the general public, promoting data sharing.

As can be seen, the data collection in this methodological proposal has been designed to align with the updates to the census and cadastral cartography currently being carried out in El Salvador. The primary goal is to enable the methodology to be widely replicated in other cities, should the government decide to make its data publicly available in the future. The exposure and vulnerability database could be made to cover the entire national territory, thus complying with the proposals of the report of the Latin American geological survey specialists [13]. As far as the authors are aware, this would be a novelty in Latin America, which would position El Salvador at the forefront of citizen protection and of the production and dissemination of geospatial information at the urban scale, by applying these data to the study of its natural hazards and risks.

2. Materials and Methods

2.1. Seismic Risk Analysis

Seismic risk is defined as the potential loss of life, injury, or damage to assets which could occur to a system, society, or community in a specific period of time, determined probabilistically as a function of earthquake hazard, exposure, vulnerability, and capacity (adapted from [30]). In this section, these concepts and the methods applied to perform the corresponding estimations are explained.

Figure 1 presents the workflow diagram followed to estimate risk in Nejapa, with particular emphasis on the exposure and vulnerability components, which constitute the main focus of this research. The minimum set of building attributes that the exposure database should contain is proposed, along with several alternative methods to obtain them, depending on the freely available data sources.

Three color-coded blocks are distinguished in the diagram: green for population; pink for building footprint area and roof material (data captured from above); and orange for the number of stories and construction materials (ground-based data collected at street level). The paths marked with thicker arrows within each block correspond to those followed in this study, due to the lack of available datasets that had to be generated. The path highlighted with thick blue arrows represents the procedure that could be followed if El Salvador were to make publicly available the geospatial data it is currently producing. As can be observed, this path is the most direct and efficient, which would enhance the generalisability of the proposed methodological approach.

The calculation methods applied to estimate ground motion (brown) and seismic risk (grey) are described in more detail in Section 2.1.1 and Section 2.1.3.

2.1.1. Seismic Action

The seismic action represents the seismic ground motion expected at the site of interest (i.e., at the base of each building). It integrates the combined effect of the energy released at the source (where the seismic rupture occurs), the attenuation of this energy with the source-to-site distance, and the ground motion amplification effects caused by local conditions due to lithological characteristics and subsurface structure at the site. In this study, the seismic action at each site corresponds to the specific response spectrum consistent with the controlling earthquake derived by hazard disaggregation. The procedure proposed by [31] with four phases is followed:

First, the probabilistic area-source method is applied at regional level, integrating all the seismic sources with their recurrences and seismic activity parameters, namely activity rates, b-values, and maximum expected magnitude for each zone [32]. The result obtained is a set of hazard curves, which give the probability of exceeding different acceleration levels at each calculation site.

Second, hazard disaggregation is performed, which consists of determining the magnitude–distance pairs that have the greatest contribution to the seismic hazard at the site of interest for a given probability level [33].

Third, an active source capable of generating earthquakes consistent with the controlling magnitude and located at a source-to-site distance consistent with the controlling distance is sought.

Finally, the seismic action considered for the risk study is determined by applying the deterministic method. For this purpose, a strong motion model is applied using the controlling magnitude and distance values.

2.1.2. Exposure and Vulnerability

Exposure includes all property and persons to be protected in seismic hazard-prone areas. In this study, only conventional residential buildings and their occupants are considered, since the study site is a residential area. And vulnerability is a measure of the likelihood of exposed assets to withstand earthquake shaking and to experience damage. To assess exposure and vulnerability, it is necessary to create a spatial database with the inventory of buildings and their most relevant attributes related to their seismic performance, such as height and construction materials, as well as the distribution of occupants (Figure 1). This inventory is usually carried out by integrating data from various sources (census, cadastre, etc.), in addition to those taken in the field. All these data are incorporated and managed within a Geographic Information System (GIS).

Once all possible data are integrated, buildings are classified into groups with similar characteristics, so-called Model Building Types (MBTs). All buildings in the same MBT are expected to show a similar seismic performance, assessed by the fragility and vulnerability model assigned to each MBT. The fragility models used in this study are taken from the analytical method of [34], since the description of their building typologies is comparable to the Nejapa buildings. They are composed of two curves: the capacity spectrum of the structure and the fragility function. The capacity spectrum relates the spectral acceleration (Sa) corresponding to the shear force generated at the base of the building (intensity measure) to the maximum spectral displacement (Sd) that the structure will suffer from the shaking (engineering demand parameter) [35]. The fragility functions give the probability that the structure will suffer different degrees of damage.

2.1.3. Damage and Losses

In the analytical method used in this study, the capacity spectrum of each MBT is crossed with the seismic demand spectrum at the location of the buildings to obtain the performance point, i.e., the maximum spectral displacement that the building suffers from the earthquake represented in the demand spectrum. To calculate this performance point, different methods have been developed and here we propose the use of the Hazus I-DCM (Improved Displacement Coefficient Method) as described in FEMA-440 [36,37]. In FEMA-440, Chapter 3.4 describes the Displacement Coefficient Method (DCM) proposed in FEMA-356, presenting the estimation of the target displacement at the building’s roof level, ⁠δ⁠, as a function of coefficients C0, C1, C2, and C3 (Equation (3.9)). Chapter 5 introduces improvements to the DCM intended to simplify the computation of these coefficients (Equations (5-1) and (5-2), and Section 5.4).

Once the performance point is obtained, the spectral displacement value is introduced into the fragility functions to obtain the physical damage to the buildings on a scale of five degrees described in [34]: null, slight, moderate, extensive, and complete.

Subsequently, the human losses associated with these damages were estimated using the method proposed in Risk-UE [38]. The authors present an empirical model for estimating casualties based on seven parameters (pp. 13–17), including the number of collapsed buildings (C), the average number of inhabitants per building (M1), and the occupancy rate at the time of the simulated earthquake (M2). For the latter, [38] provides occupancy graphs from which the rate can be derived for different hours of the day. In this study, three extreme scenarios were considered: at 2:00 a.m. (the most unfavourable, as it presents the highest occupancy), at 10:00 a.m. (the lowest occupancy), and at 7:00 p.m. (intermediate occupancy). In addition, the model includes three other parameters that account for the percentage of people trapped in collapsed buildings (M3), the severity of injuries (M4), and the percentage of fatalities depending on the capacity of emergency services (M5). For these parameters, the authors provide reference values according to the building typology.

Furthermore, Risk-UE (p. 17) provides a model for estimating the number of homeless people as a function of the damage grade of each building, the number of residents, and the building category (single- or multi-family).

Finally, the reconstruction and repair costs were estimated according to the damage level of each building ([38], p. 24). It was assumed that buildings with complete damage would require full reconstruction (100% of their cost), whereas those with extensive, moderate, or slight damage would require repair, estimated at 50%, 10%, and 2% of the reconstruction cost, respectively.

2.2. Study Area and Data

The city of Nejapa is located in the north of El Salvador and has more than 30,000 inhabitants. The study area is the historical centre of the city (Figure 2). Its delimitation is taken from the Territorial Development Plan for the metropolitan sub-region of El Salvador and has an area of approximately 100,000 m². It has an orthogonal road layout, with two main streets running north–south crossed by secondary streets running east–west. Most of the houses are one-storey, with some exceptions of two-storey houses; only a few have three stories.

About 5 km north of the city is the Guaycume fault [28], which runs northwest–southeast.

The data for the seismic risk study have been taken from several sources. The Hazards and Natural Resources Observatory of El Salvador (DGOA-MARN) has collaborated by providing data on the fault, such as geometry and maximum magnitude, and a map of continuous values of seismic wave shear velocity in the first 30 m of crust (Vs30), necessary to consider the effect of the ground in the calculation of the seismic action. The values of Vs30 in the study area range from 451.7 to 504.8 m/s. Data on the fault mechanism are taken from [28].

For the exposure assessment, a vector layer of points in shapefile format containing the location of the buildings has been used (Figure 2). It comes from a population census conducted in 2006 (Abel Argueta, Universidad Nacional de El Salvador, personal communication), so it is considered out-of-date or incomplete and should be updated in future work.

In addition, from the National Drinking Water and Sanitation Plan it has been possible to obtain a layer of census sections with population estimates for the year 2022, and another with the geometry of the parcels. However, no metadata came along with these layers.

An initial data quality check was conducted, comparing these data with open repositories, such as Google Earth, Google Street View, and OpenStreetMap. However, they either were non-existent, had insufficient resolution, or did not have digitised buildings. Therefore, given that the available data are clearly insufficient or of uncertain quality, a field campaign had to be conducted to collect data on which to base the exposure and vulnerability assessment.

2.3. In-Field Data Collection

A field campaign was carried out on the streets of the historic centre of Nejapa to collect building data. In the preparatory phase, we used ArcGis to calculate the optimal route to cover the entire study area on foot in the shortest time possible. For the study of exposure and vulnerability, fieldwork is always recommended, as it allows access to interiors, measurements, and data collection with minimum uncertainty. However, it is not always possible for reasons of safety, excessive size of the area to be covered, time available, or budget [39]. It is recommended that visits are made with specialised local staff and as safely as possible for surveyors and measuring equipment.

The aim of this on-site data collection is to create the footprint of the buildings and calculate the attributes related to their seismic performance: area, height (or number of stories), main construction materials, and roof material.

The following sensors were used:

• Panoramic camera: The Ricoh Theta V 14 Mpx camera was used, costing around USD 500. A total of 500 photographs were taken in 3 h, covering all the streets in the study area (Figure 3B). The software provided by the camera allowed for creating continuous imaging. By photo-interpreting these ground perspective images, it was possible to obtain the attributes related to the materials of the structure and the exterior walls, as well as the number of floors of each building.
• UAV flight (Figure 3): a DJI MAVIC pro drone was used, at a cost of approximately USD 1500. We used the Ground Station pro application to set a flight over the area at an altitude of 100 m in 11 passes, with 90% longitudinal and 60% transverse overlap between passes. Two flight sessions of about 2 h each were conducted, according to the battery lifespan. We created the orthophoto and the digital surface model of the study area with Metashape Version 1.7 software. Both were used to derive the footprints (contours) of the buildings and the construction material of the roofs.

The data taken with the camera and the drone have been processed with a computer equipped with an Intel core i9 processor, 3.4 Ghz, with 64 GB of RAM and a 2 TB solid-state drive. The cost of the computer is approximately USD 2500.

3. Results and Analysis

3.1. Ground Motion Estimation

We consider the seismic source model and the ground motion model used in the regional seismic hazard study by [40], which provides regional expected ground motion estimates for different return periods in rock conditions. To particularise these estimates to the local setting of Nejapa, we perform a hazard disaggregation analysis with these specific constraints: First, we consider the reference coordinates at a site in Nejapa. Second, we define a target ground motion parameter for hazard disaggregation consistent with an intensity measure (IM) suitable for assessing the performance of typical local structures e.g., [41,42]. As shown below, the largest proportion of the building stock in Nejapa consists of one- to two-storey masonry buildings, whose performance is most appropriately evaluated using a spectral acceleration of 0.1 s, SA (0.1), as the IM. Third, we adopt an SA (0.1) value consistent with the seismic hazard corresponding to a 475-year return period, which is suitable for normal-importance dwelling structures such as those in the study area. Finally, we obtain the resulting site-specific controlling parameters (controlling magnitude and controlling distance) for Nejapa in rocky conditions.

Figure 4 shows the results. It is observed that the controlling earthquake, which is associated with the maximum contribution to the seismic hazard, corresponds to a local, cortical event, of magnitude between 6.0 and 7.0, located at an r_rup source-to-site distance between 20 and 40 km, which is consistent with an r_JB (Joyner and Boore) distance of between 0 and 20 km.

We analyse the active faults that are capable of generating earthquakes of a size consistent with the controlling magnitude and located at a source-to-site distance compatible with the controlling distance. As a result of this analysis, we identify Guaycume fault as the best candidate. It is a crustal strike-slip dextral fault located at a r_JB distance of 0 to 20 km from Nejapa and it is capable of generating earthquakes of magnitudes within the range of values obtained in the disaggregation analysis [27,28]. It is one of the most prominent active structures and has the highest deformation rates in the area [43].

We carry out a deterministic hazard scenario considering this seismic source. The extreme earthquake simulated in this study was a Mw 6.7 event. As a complement, an event of Mw 5.0, more frequent than the previous one, is also considered to assess its impact.

Table 1. Guaycume fault data.

Fault Parameters
Length (km)	25.3
Depth (km)	12.7
Max. Magnitude	7.0
Strike (°)	108
Dip (°)	80
Rake (°)	180
Paleoseismic Slip Rate (mm/yr)	5.3
Geodetic Slip Rate (mm/yr)	8.0
Earthquake Scenario Rupture Data
Hypocentre Latitude (° N)	13.87
Hypocentre Longitude (° W)	89.28
Hypocentre Depth (km)	10

summarises the seismic source data considered to compute the seismic scenarios.

In this application, we have selected the ground motion model of [44] for cortical sources. This is one of the strong motion models considered most suitable for cortical sources in El Salvador, according to the analysis of residuals carried out by [45]. This model also requires introducing the site effect through the Vs30 value, which ranges from 450 to 500 m/s in the study area (Figure 5, left), according to the analysis of [46].

For Nejapa, Figure 5 shows the response spectra obtained for the two ruptures considered including site effects. Median values, together with the median plus-one-sigma spectra and median minus one-sigma spectra, are included to reflect the ground motion variability expected in the city. The Mw 5.0 earthquake presents much lower accelerations (between 75% and 95%) for all spectral ordinates. It reaches a maximum around 0.2 g at the 0.1 s spectral acceleration, while the Mw 6.7 earthquake reaches 0.95 g at the same spectral ordinate.

3.2. Exposure and Seismic Vulnerability Evaluation

Since there is no layer of building footprints available providing a geographical basis for the exposure database, we had to create them using the orthophoto obtained from the drone flight. It has been georeferenced with control points generated from the available layer of parcels. The accuracy of the orthophoto has been estimated to be 20 cm, which is sufficient for the purpose of this study.

Initially, we attempted the extraction of building footprints semi-automatically, by means of an object-based analysis (OBIA) in Arcgis Pro. The OBIA segmentation step has not given satisfactory results, with most of the buildings being over-segmented due to the complexity of their shapes and the heterogeneity of their covering materials (particularly those with clay tiles, which constitute the majority in the city; Figure 6).

As an alternative, we have conducted a classification by pixel. For this purpose, a vector layer containing 150 random points was generated in the GIS environment. Of these, 94 points fell on building roofs, where polygons were subsequently delineated to collect the sample data. Since the objective was to identify roof types, the following classes were defined: grey asbestos, red asbestos, metal, and clay tile. In addition, 20 samples of vegetated areas were included. The number of samples per class is proportional to their presence in the study area. Subsequently, a testing dataset was created to verify the classification a posteriori. This was achieved in a similar way, by generating a vector layer with 90 random points, of which 68 fell on building roofs or vegetation. Both asbestos roof types were merged in a single class, as colour has no influence on the building’s seismic performance. This testing dataset was reserved as unseen data. These numbers are shown in

Table 2. Number of samples in the training and testing datasets.

	Nr. Samples
Class	Training	Testing
Grey Asbestos	16 pol. (1,029,905 pix.)	16
Red Asbestos	12 pol. (522,707 pix.)
Metal	14 pol. (611,631 pix.)	16
Clay Tile	32 pol. (2,956,151 pix.)	24
Vegetation	20 pol. (2,418,297 pix.)	12
Total	94 pol. (7,538,691 pix.)	68

.

The classifier used was a Support Vector Machine [47], and a confusion matrix was derived in the testing phase, from which the main accuracy measures described in [48] were computed. As a result, an average F1-score of 80.7% was obtained, with similar values for recall, precision, and overall accuracy. The F1-score is a measure of accuracy that is calculated as the harmonic mean between the user’s accuracy and the producer’s accuracy. It is a very comprehensive metric that balances the cost of false positive and false negative errors. It is considered the most suitable for comparing models when samples are unbalanced, which is the case in the present study. The classification outcome can be regarded as highly satisfactory based on the accuracy metrics obtained.

In addition, Cohen’s kappa coefficient was calculated to compare the observed accuracy with that expected by chance, providing a more robust measure of classification agreement [49]. In this study, a kappa of 0.74 was achieved, indicating substantial agreement.

Table 3. Confusion matrix computed in the testing phase, and the main average accuracy measures.

		Predicted
		Vegetation	Asbestos	Clay Tile	Metal	total
Real Class	Vegetation	10	0	2	0	12
	Asbestos	1	13	2	0	16
	Clay Tile	0	3	17	4	24
	Metal	0	0	1	15	16
	TOTAL	11	16	22	19	68
	Average accuracy measures
	Overall Accuracy	Recall	Precision	F1-Score	Kappa
	80.9%	80.9%	81.0%	80.7%	0.74

The bold numbers along the diagonal of the confusion matrix represent the true positive (TP) classifications for each class.

presents the confusion matrix, followed by the main average accuracy measures and the Cohen’s kappa coefficient.

In parallel, as the segmentation could not generate building footprints automatically, we digitised the building footprints manually (e.g., [39,50,51]), using the point data of the population census as a reference. This resulted in a total of 262 buildings, which took less than 2 h of work. Finally, by means of a zonal query, the mode of the classified pixels was assigned to each footprint to obtain a classification of the roofing material in the study area. The result shows that there are 133 buildings with clay tile roofs (51%), 61 have metallic roofs (23%), 46 have grey asbestos (18%), and 22 have red asbestos (8%).

On the other hand, by observation of the spherical images, it was possible to provide each footprint with other relevant attributes, such as building material and number of floors. The construction material could be identified by observing the thickness of the walls, the arrangement of the windows, and the sides of the buildings, as in many cases there was a lack of cladding material. The approximate average time for photointerpretation was estimated to be less than 5 min per building in this study area. Expert knowledge is recommended to accomplish this task effectively. All the buildings in the study area are masonry structures, distributed as follows: 113 adobe buildings (43%), 73 solid brick block buildings (28%), and 77 concrete block buildings (29%). In terms of the number of stories, 224 buildings are single-storey (85%), 37 have two stories (14%), and only two buildings have three stories (1%). All of them are residential and of normal importance, which allows us to adopt vulnerability models from the literature developed for this type of construction.

Accordingly, based on the attributes included in the exposure database, the buildings were classified into three typologies following the models proposed by [34], who provide a comprehensive set of vulnerability models for masonry structures: M2, which are adobe dwellings with one or two floors; M5, which represents unreinforced masonry buildings with wooden floors; and M7, which are reinforced masonry dwellings made with mortar blocks. Each of these typologies has been subdivided into groups according to the building height, being called low-rise (_L) when the buildings have one or two floors and mid-rise (_M) when they have three. Finally, they have been assigned the vulnerability model proposed by [34], which represents the behaviour of each type of building in the event of an earthquake by means of a capacity spectrum and a set of fragility curves. The authors provide capacity and fragility curves for low-quality masonry structures that distinguish between different floor materials. This distinction is essential in our study, as it allows us to separate buildings with wooden floors (M5), which are more ductile, from those with concrete floors (M7). Figure 7 shows the spatial distribution of the seismic vulnerability of the historic centre of Nejapa.

Regarding the exposed population, the number of persons per dwelling has been estimated based on the layer of census sections with population estimates for 2022 and considering the number of digitised dwellings in each section. For the historic centre of Nejapa, there is an estimate of 776 persons, which results in an average of 3 persons per household.

3.3. Earthquake Damage and Loss Assessment

The damage distribution is shown in Figure 8. For the Mw 6.7 earthquake, a total of 185 dwellings would suffer complete damage (71%) and 77 would undergo moderate damage (29%). For the Mw 5.0 earthquake, damage is much reduced, leaving 43% of the housing stock with slight damage (113 dwellings) and 57% with no damage (149 dwellings).

The two simulated scenarios correspond to those obtained from the seismic hazard disaggregation. The Mw 5.5 event is the more probable one and, in light of these results, would not cause severe damage in the city of Nejapa. However, the Mw 6.7 event would result in a high level of severe damage, leading to a disastrous situation that would be difficult to manage for the emergency response agencies. The increase in damage in this extreme scenario can be attributed to the difference of more than one unit in moment magnitude, which leads to an increase of almost one order of magnitude in the short-period (0.1–0.2 s) spectral acceleration (from 0.2 g to 1 g, Figure 5). This is a crucial factor in the study area, where buildings show high vulnerability—about 70% are built with low-quality masonry—and are predominantly low-rise structures, which are more affected by vibrations associated with short-period ground motions.

Based on these damage results for the Mw 6.7 scenario, a simple estimate of the associated human and economic losses has been made. It was found that about 530 people would be homeless, which represents 68% of the population in the historic centre of Nejapa (the study area). Three scenarios have been simulated at different times to vary the level of household occupancy: at 2:00 a.m., at 10:00 a.m., and at 7:00 p.m. Figure 9 shows the toll of human losses for each scenario and the average value. The expected number of fatalities varies between 38 and 67 (with an average of 54). Serious casualties range from 88 to 155 (with an average of 126) and light and uninjured casualties range from 329 to 426 (the average is 370).

Economic losses have been calculated assuming a construction cost of 550 USD/m² in the San Salvador Metropolitan Area (architect of the El Salvador Metropolitan Area Master Urban Plan, personal communication, 2021). For the simulation of the Mw 6.7 earthquake, it has been obtained that the cost of reconstruction of the buildings that would be destroyed is USD 14,650,000, to which must be added USD 763,000 in repair costs of damaged buildings. In the case of the Mw 5.0 earthquake, there would be no dwellings to rebuild, but there would be a repair cost of USD 168,000.

3.4. Dissemination of Results

All the information collected, as well as the results obtained with the different software, has been integrated and managed in ArcGis Pro v2.8. We implemented a dashboard (Figure 10, [52]) to facilitate very fast queries to the databases, crossing different tables to display combined data at the same time (for example, building typologies and heights), or to visualise different selections that have helped to detect inconsistencies in the data and to correct them very efficiently. The dashboard is also useful to prepare graphs and diagrams for the analysis of data and results. This control panel, together with all the data and results generated, has been shared with the municipality of Nejapa, Civil Protection, and the Hazards and Natural Resources Observatory of El Salvador so that they can make this study useful, defining DRR measures based on the results. Therefore, the purpose of creating and disseminating open data in El Salvador has been fulfilled.

4. Discussion and Main Conclusions

With this work, we have shown how it is possible to carry out a complete seismic risk study, in spite of not having initial data on the exposure and vulnerability of the building. The solution consisted of collecting data with a 360° camera and a UAV and analysing the captured terrestrial photographs, as well as the orthophoto, MDT, and MDS generated, to obtain the necessary information on the buildings in the study area. The data collection and analysis work were carried out in a short time, less than a week. This working procedure offers important advantages over the traditional method of field data collection, such as improved physical safety, reduced logistical complexity, and greater comfort, all of which contribute positively to overall performance and replicability.

In terms of the results obtained, it has been observed that an earthquake of Mw 6.7 on the Guaycume fault, about 5 km from the historic centre of Nejapa, would be catastrophic and very destructive, with dozens of victims and two thirds of the population displaced from their homes. In fact, this scenario would be very similar to the one that followed the earthquake that devastated several cities in El Salvador in the early 20th century (e.g., [53]). An earthquake generated in the same location, but much more frequent, of Mw 5.0, would only cause minor damage to buildings and no casualties. While economic losses were estimated, these figures should be interpreted cautiously due to the simplified methodology used; future studies should incorporate more detailed economic modelling for greater accuracy.

The results obtained in the present study could be used to propose specific disaster risk reduction (DRR) strategies for Nejapa. A cost–benefit analysis of the implementation of structural retrofitting for highly vulnerable structures should be a priority [54]. In addition, low-cost, easy-to-implement measures should be promoted, including anchoring or removal of prone-to-fall non-structural external elements and contents [55]. Furthermore, working directly with the population to raise awareness among families about the importance of safe construction practices in seismic areas is suggested. This could be achieved through social participation activities in local communities, dissemination of information through accessible communication channels, and the organisation of drills and preparedness exercises in collaboration with public authorities.

It should be recognised that, due to limited human resources and time constraints, the study area covered is relatively small. Future work should aim to expand the analysis to larger urban areas and incorporate diverse building typologies and urban configurations. Given that El Salvador is currently undertaking a country-wide data collection program to update its digital urban cartography and census, as well as to modernise its cadastral system, these data sources would be highly valuable for replicating the techniques presented in this study in the generation of exposure and vulnerability databases at a local scale within a national coverage program. However, to fully leverage these developments, it would be highly desirable that, in parallel with the modernisation of its urban spatial data, El Salvador develops a policy to facilitate open data sharing between public institutions and researchers. Such a policy would benefit both sides: for analysts, access to official data would make studies faster and more accurate; and for institutions, the transfer of results would contribute to national development.

Therefore, while this study provides a solid foundation for rapid seismic risk assessment using remote sensing and photogrammetry, future research should focus on scaling the methodology, improving data integration, and fostering institutional collaboration to support resilient and sustainable urban development.

Apart from risk management, census and cadastre data would be useful in other areas, such as water and waste management, habitability and sustainability studies, urban planning and smart cities, the fight against climate change, the generation of digital twins, etc. These issues are priorities in the United Nations Agenda for Sustainable Development, specifically in SDGs 6, 7, 9, 11, and 13. Far from being a drawback to a nation’s interests, data sharing has been shown in other countries to be a path to development, by enabling the combined advancement of knowledge, research, and technology.