Probabilistic Seismic Risk Assessment in Manizales, Colombia: Quantifying Losses for Insurance Purposes

The objective of a probabilistic seismic hazard analysis (PSHA) is to provide a long-term overview of the relationships between strong ground motion levels and their occurrence frequencies. For a fully probabilistic seismic risk assessment, a hazard representation by means of stochastic events (event-based approach) is required. A PSHA allows the user to obtain feasible earthquakes with different locations, magnitudes, and occurrence frequencies, associated to all the seismic sources of interest. For this case, the PSHA was developed using the latest version of the program CRISIS, a well-known and accepted tool that implements state-of-the-art methodologies to perform PSHA (Ordaz et al. 2013). This program is also the seismic hazard module of the CAPRA platform and an input data provider for the R-System, used for the case of buildings.

The PSHA was developed at local level considering more than 35 seismic sources for which, in all cases, a geometric area model was used for their representation accounting for variations in dip angles and depths (Salgado-Gálvez et al. 2013, 2016b). Regionally developed and calibrated ground motion prediction equations (GMPE) were assigned to each seismic source (Bernal 2014), thus allowing the consideration of different hazard intensities such as spectral accelerations, peak ground velocity (PGV), and peak ground displacement (PGD). As a result of the PSHA a total of 1275 stochastic events were obtained, each of them with a set of hazard intensities associated to several spectral ordinates ranging between 0.0 and 4.0 s, and for ground accelerations, velocities, and displacements. Since a Poissonian process is assumed to occur for the loss generation process, the stochastic event set needs to be mutually exclusive and collectively exhaustive; individual events need to have an associated occurrence frequency and data of at least the first two statistical moments for each hazard intensity measure.

Due to the characteristics of the soils in Manizales, and also because of the local nature of the analysis, as well as its high level of resolution, it is relevant to take into account the dynamic soil response. This was done by using spectral transfer functions based on the results of the recently updated and harmonized seismic microzonation developed for the city [also developed within the framework of the disaster risk management and reduction project, see Bernal et al. (2015, 2017)] that provides results at 377 different locations.

The development of exposure databases has always constituted a great challenge in modeling disaster risk because, in most cases, in order to complete the identification and characterization process, data from several sources need to be combined, leading to a process where assumptions and simplifications are usually required. However, the exposure databases developed for these case studies can be considered an exception. Given the previous continuous efforts of the water and sewage company, as well as of the city (cadastral office), in gathering, arranging, classifying, and updating the information related to the exposed assets, all relevant parameters for a seismic risk assessment (ranging from age to structural system, construction material, location, and geometry) could be obtained directly from the official sources (only specific field visits to validate and review very specific issues were needed). The economic appraisal of each element included in the exposure databases was done using official data in a task where close collaboration was received from specialists working in the water utilities company of Manizales and the cadastral office of the city.

The exposure databases for the water and sewage network, as well as for the public and private buildings, required the development of specific tasks such as the identification of the relevant structural parameters that are related to the expected behavior of each asset when subjected to earthquake hazard intensities. Because of the evident differences between the type of components, and since the use of the results was not directly linked or combined, these activities were performed in a separate manner.

The economic appraisal of the whole water and sewage network was developed in terms of the replacement cost, that is the amount of money required to repair or rebuild the damaged element and take it back to the exact same conditions it was in before it was damaged; in this case, the value for each element does not simply correspond to its monetary cost (for example, 5 m of concrete pipe), but also to other costs associated to activities of installation, such as breaking and repairing streets at different depths. All values were obtained directly from the water utilities company and those data can be said to be complete, reliable, and updated. Overall, pipelines for primary and secondary water and sewage networks account for most of the exposed value with around 80%, followed by the water storage tanks that account for approximately 7% of the total exposed value of the water and sewage network in Manizales.

For the case of the public and private building portfolio, the base information to assemble the exposure database was delivered by the Secretary of Finance and the Risk Management Unit (Unidad de Gestión del Riesgo) of the city administration, and was complemented with additional parameters from aerial images, existing maps, and field visits. The first exposure database for private buildings in Manizales, developed for purposes related to seismic risk assessment in 2004, included representative variables such as age, number of stories, structural system, location, and socioeconomic level. The latter is specified in terms of strata, from 1 to 6, in which strata 1 and 2 group inhabitants with the lowest economic capacity and are therefore granted a property tax exemption, whereas strata 3–6 group property tax payers. At the time the project was conducted, more than 85,816 dwellings were included and, for the purpose of the analysis, they were split into three different categories as explained in detail in Marulanda et al. (2014, 2016).

One of the tasks within the disaster risk management project developed in Manizales was related to the update and completion of the exposure database to take into account not only improvements related to the content and attributes of the exposure database, but also the consideration of new buildings that have been built since 2004. For the purpose of this analysis more than 113,064 dwellings were considered. Figure 1 shows the geographical distribution of the identified structural systems in the city. Non-engineered materials and structural systems such as adobe and other earthen structures (bahareque) can be easily identified in the oldest and historical areas of the city (Cumanday, San José, and La Macarena). Further details about the assembly of the buildings exposure database of Manizales can be found in Gonzalez (2014).

The vulnerability function approach accounts for a continuous, quantitative, and probabilistic representation of the physical damage of the assets for different levels of hazard intensities. For each asset included in the exposure databases, a unique vulnerability function is assigned in order to estimate the losses associated with different hazard intensity levels, for each of the considered stochastic events. Due to the great variety of types of elements, vulnerability functions in terms of several hazard intensities were developed, given that different intensities show good correlation with the damage of different types of elements. For example, vulnerability functions were developed in terms of peak ground acceleration (PGA) for the case of buried water storage tanks, in terms of spectral accelerations for administrative buildings and elevated water storage tanks, and in terms of PGV for buried pipelines. For the latter, the recommendations provided by the American Lifeline Alliance—ALA (2001) were followed in order to obtain the repair rates, RR (the number of damages to be repaired by km).

For other types of components such as administrative buildings and water storage tanks, the development of the vulnerability models included the consideration and combination of different approaches, such as analytical and empirical analyses combined with expert criteria. For all types of components, loss is considered to be a random variable for which the variation of the first two statistical moments for different seismic demand levels are described by means of vulnerability functions (Miranda 1999; Ordaz 2000). Figure 2 shows the vulnerability functions (expected values of the loss) used for the different water storage tank classifications (left), and some of the vulnerability functions used for the building portfolio (right).

Once the input data for hazard, exposure, and vulnerability was ready, the loss estimation was made following the fully probabilistic and event-based procedure proposed by Ordaz (2000). This procedure has been implemented in the CAPRA-GIS program, which corresponds to the risk module of the CAPRA platform, in the case of the water and sewage network, and the R System (ERN 2014), in the case of the portfolio of buildings. The main outcome of a state-of-the-art probabilistic risk assessment corresponds to the LEC, which contains several loss values together with their associated exceedance frequencies. The same methodological approach and procedure was followed to estimate seismic losses in the two different exposure databases and, therefore, results in terms of the same risk metrics have been obtained. Under this approach, the LEC is calculated using 50 different loss values ranging between 0 and 80% of the total exposed value (logarithmically spaced) using the following expression:where ν(l) is the loss exceedance rate (1/year), Pr(L > l|Event _i ) is the loss probability exceedance conditional to the occurrence of the ith event, and F _A (Event _i ) is the annual occurrence frequency of the ith event. Once the LEC is calculated, other risk metrics such as the AAL and the PML (for any return period of interest) can be obtained; additionally, loss exceedance probabilities for different exposure timeframes can be computed. Equation 1 shows that an event-based approach is required for the hazard representation because the associated occurrence frequencies are required along with the expected hazard intensities. More details on the loss assessment methodology can be found in Ordaz (2000) and Salgado-Gálvez et al. (2014b).

For the case of the water and sewage network, additional analyses were considered for particular feasible earthquakes in which the approach from the temporal perspective can be considered as deterministic, but which is still probabilistic from the loss estimation perspective (since the uncertainties associated with the strong ground motion intensities and the vulnerability representation are still accounted for). In that case, the expected loss for the particular scenario is estimated using Eq. 1 where the occurrence frequency of the considered event is assumed to be equal to 1.0.

Table 1 shows the risk results for the components of the water utilities company in terms of the relative AAL. From Table 1 it is evident that the component with the highest relative AAL corresponds to the water storage tanks with around 6.6‰ (a considerably high value bearing in mind that the AAL of the whole network is 0.65‰). From these results, it is also worth noting that, although pipelines concentrate the vast majority of the exposed value, their AAL is low when compared, in relative terms, to any other component. Furthermore, it is important to see that the administrative buildings have an AAL that by no means can be considered as negligible with the aggravating fact that those buildings are to be considered as critical components. This is because aside from the physical losses, any important damage on the buildings can gravely affect the operation of the network in the aftermath of an earthquake, not only from a purely operational point of view but also from the emergency management perspective (Salgado-Gálvez 2014). For the case of the buried pipelines it is also possible to develop risk maps in terms of average annual repair rates (by km) such as the one shown in Fig. 3 for the sewage network of Manizales. More details about this study can be found in Salgado-Gálvez et al. (2017).

In addition to the fully probabilistic seismic risk assessment, an event capable of causing losses with a return period of 1000 years occurring in the vicinity of the city was considered (event with magnitude 7.3, with an epicenter located 25 km southwest of Manizales). In this case, instead of a loss exceedance rate, what is directly obtained is the mean damage ratio (MDR) of each component of the network, which is approximately 3.5% of the total exposed value. As in the fully probabilistic assessment, risk is mostly concentrated in the water storage tanks and the administrative buildings. Repair rates for pipelines with different materials and joint characteristics were compared with published post-earthquake survey data of water and sewage systems (Ballantyne et al. 1991, 2002), finding congruent results in terms of RR. Several of the tools and programs used for this study have been previously tested and validated with particular damage results for earthquakes in Lorca, Spain (Salgado-Gálvez et al. 2016a).

For the building portfolio, three separate seismic risk assessments were performed. The first included only the buildings that are exempt from property taxes (socioeconomic strata 1 and 2), the second included only the buildings subjected to property tax payments (socioeconomic strata 3–6), and the third included all 113,064 dwellings.

Figure 4 shows the geographical distribution of the relative (to the exposed value) AAL for the public and private buildings of Manizales. The highest values, around 2.25%, correspond to dwellings with non-engineered building classes; this value of 2.25% can be understood as losing the total exposed value of each asset on average every 40 years (Marulanda 2014).

Table 2 shows the summary of the risk metrics (AAL and PML) for the analyzed portfolios considering different deductible levels, in all cases using the methodological approach to account for limits and deductibles proposed by Ordaz (2000). The higher the deductible level in a traditional earthquake insurance scheme, the lower the expected losses in terms of either AAL or PML, for any return period.

The value of the pure premium, in 2014, for private properties was set equal to 2.86‰ without deductible, and to 1.63‰ after applying a 3% deductible. Despite the significant increase in the number of existing properties in the city in a period shorter than 10 years (from 85,816 to 113,064), the decrease of the pure risk premium value in the city (with 3% deductible) from 1.98 to 1.63‰ is particularly remarkable. It is rarely possible to measure the seismic risk of a city in different time periods with the same metrics, and even more so to reveal a reduction of risk which, in the case of Manizales, could be expected due to the efforts made in risk management. In recent years a significant number of buildings have been retrofitted in Manizales, particularly public sector buildings. This can explain the new pure risk premium value of 1.62‰ (with 3% of deductible) for these properties. This value is practically the same if compared to the value estimated for the private buildings. However, the main reason why the seismic risk of the city has been decreasing relates to the significant increase in construction of new buildings that effectively take into account all the earthquake resistance requirements established in the Colombian earthquake resistant building code—NSR-10 (Marulanda et al. 2016).