In its current version, RAPID-N includes the most frequently used damage classifications and fragility curves for storage tanks available in the scientific literature. For this case study, RAPID-N automatically selected for each plant unit a fragility curve and a corresponding damage classification that represented the best fit with the available data. Where plant-unit characteristics required for the calculations were missing, these data were estimated using RAPID-N’s built-in property estimation framework. For the 15 storage tanks included in the damage analysis, RAPID-N utilized the fragility curves defined by O’Rourke and So (2000) for atmospheric cylindrical tanks, and those of Moschonas et al. (2014) for pressurized spherical tanks. The different damage states used and their definitions are summarized in Table 10.3. The damage states of O’Rourke and So are similar to HAZUS damage states commonly used for seismic damage assessment (FEMA, 2010).

O’Rourke and So (2000) provide four different fragility curve sets for atmospheric storage tanks applicable for fill levels <50% and ≥50%, and height/diameter (H/D) ratios <0.7 and ≥0.7. For pressurized tanks, Moschonas et al. (2014) provide two fragility curve sets for column support with and without diagonal braces. Using this information, three different fragility curves were selected by RAPID-N to assess the probability of the different damage categories. The curves and the corresponding damage probabilities as a function of damage state are summarized in Table 10.4. The damage parameters are provided for discrete rather than cumulative damage categories and show the probability of a certain level of damage as shown in Fig. 10.4.

The damage data in Table 10.4 were subsequently used in the Natech risk analysis. As outlined in Chapter 8, the expected damage states and their probabilities were linked to consequence scenarios via risk state definitions, and the consequences were then estimated by using scenario-specific, dynamically generated consequence models formed by RAPID-N. For some damage states, the defined physical damage does not imply loss of containment. For some others, no explicit information is provided by the damage definition about substance amount possibly released although physical damage leading to a release is indicated. Since historical data about release rates due to seismic damage are scarce, possible release scenarios were defined based on expert judgment as percentage of tank volume released for each damage state. Similarly, the conditional probability of release given a specific degree of physical damage was assigned to each damage state since loss of containment is not always expected, especially if physical damage is limited. Risk states used for the case study are summarized in Table 10.5.

Since the consequence-analysis methods and equations currently available in RAPID-N are based on US EPA’s Risk Management Program (RMP) guidance for offsite consequence analysis methodology (US EPA, 1999), a number of simplifying assumptions were made for the risk analysis. For example, for all scenarios the atmospheric stability was assumed to be neutral (corresponding to Pasquill stability class D) with a wind speed of 3 m/s, which is the alternative scenario according to RMP. The direction of the wind was not taken into account as it is not included in the RMP methodology. The ambient temperature was assumed to be 25°C and the humidity 50%. For surface roughness, rural topography (flat terrain) was considered as the facility is mainly surrounded by sea and crop fields as shown in Fig. 10.2. Hazmat releases due to the seismic forces were assumed to occur at ground level. In addition, for unbounded liquid spills (i.e., not contained within a dike or in case of dike overflow) a minimum pool depth of 10 cm created by a substance release was assumed. While the RMP methodology adopts a pool depth of 1 cm, it was found to yield unrealistic pool sizes, in particular if the amount of released substance is high. For relative probabilities of ignition and explosion for flammable substances, estimates by Cox et al. (1990) were used. The chemical properties of the hazardous materials used for the consequence analysis are summarized in Table 10.6.

For damage states DS2 and DS3, some methanol is predicted to be released which forms a pool that ignites but stays confined in the tank’s dike area. In the case of DS4, the amount of released substance increases significantly. However, because the dike’s storage capacity is big enough to hold the methanol spill, the end-point distance does not increase and the thermal effects of the pool fire stay mitigated. For the worst-case damage scenario DS5, the whole tank volume is released due to the earthquake damage. The dike cannot hold the spilled methanol, causing a dike overflow and the spreading of the methanol beyond the dike perimeter. Assuming that the spill creates an unbounded pool with a minimum depth of 10 cm outside the dike area, the resulting pool area increases more than twofold. Consequently, the end-point distance for heat radiation also shows a substantial increase.

Fig. 10.5 indicates that depending on the level of damage to T-1, some other tanks on-site are predicted to fall within the heat radiation zones. Since the thermal-radiation intensity criteria are defined for humans, escalation effects (e.g., domino accidents) are not expected at this level, although some physical damage to other units is probable. Heat intensities that can significantly affect storage tanks are in the range of 9.5–38 kW/m², and intensity limits suggested for the quantitative risk analysis of domino effects are 15 and 45 kW/m² for atmospheric and pressurized storage tanks, respectively (Cozzani et al., 2006). Additional analyses with RAPID-N for these end-point intensities showed that among the tanks adjacent to T-1, the pressurized storage tanks containing ammonia (S-1 and S-2) are located outside the 45 kW/m² end-point distance estimated as 11 m for DS2–DS4 and 29 m for DS5. However, the atmospheric storage tank containing methanol (T-2) is inside the 15 kW/m² end-point distance calculated as 19 m for DS2–DS4 and 51 m for DS5. It is also adjacent to the higher 45 kW/m² end-point distance. RAPID-N therefore gives an indication of the other plant units potentially at risk from heat impingement if T-1 undergoes damage and the released methanol ignites. While not providing a detailed quantitative estimate of this risk, RAPID-N can nonetheless highlight areas of concern due to potential domino effects.

A dense-plume model was used to simulate the atmospheric dispersion of acrylonitrile, which is heavier than air. Reference Table 19 (dense gas, 60-min release, rural conditions, atmospheric stability D, wind speed 3 m/s) of the RMP guidance was utilized to determine the end-point distances. In accordance with the definition of damage state DS1, no release occurs although there is some seismic loading. For damage states DS2–DS5, RAPID-N predicts the release of acrylonitrile and evaporating pool formation, which results in an end-point distance of 2.4 km with occurrence probabilities ranging between 1.5 × 10⁻¹ and 2.2 × 10⁻³. The released amount of substance shows a pronounced increase for DS4 and DS5, but the end-point distance is not affected. Similar to the analysis for tank T-1, the dike’s holding capacity is sufficient to keep the substance confined within the dike, this time even for the worst-case scenario, and hence evaporation is limited. This demonstrates the importance of measures to contain the spill. The results of the analysis for tank FR-1 are summarized in Table 10.8 while the impact areas are shown in Fig. 10.6.

When subjected to the seismic forces of the Istanbul earthquake scenario, RAPID-N predicts releases from all tanks. Consequently, the maximum end-point distances increase significantly to a maximum value of 13.5 km with the atmospheric dispersion of ammonia for the worst-case scenario of complete release of content from tank S-2. Fig. 10.7 shows that in this case also urban areas at the shore of Izmit Bay opposite the case-study installation could be at risk of exposure if the wind blows in this direction. Fig. 10.8 provides a close-up of the expected Natech end-point distances in the vicinity of the installation. The end-point circles are concentrated in an area of 2.4 km around the tank farm, making this the most critical region for suffering toxic effects.

For tanks with similar dimensions and structural characteristics (e.g., tanks FR-1 to FR-6) RAPID-N calculates similar earthquake damage probabilities and related Natech event probabilities. The end-point distances are also similar and differ only due to the substance type (e.g., acrylonitrile vs. vinyl acetate) because in most of the cases the containment dikes were large enough to hold the released material for all damage states and hence they reduced the pool surface area, which restricted evaporation. The effect of H/D ratio on the damage probability can be clearly seen in Table 10.4 between floating-roof (H/D < 0.7) and fixed-roof tanks (H/D ≥ 0.7, except T-5). Increasing the H/D ratio shifts the higher damage probabilities from lower damage states (DS1–DS2) to higher damage states (DS3–DS4), although the difference between the estimated probability values is less than one order of magnitude. Nevertheless, in this case study the Natech probabilities for fixed-roof tanks were found to be much smaller than those for the other atmospheric tanks. Since the fixed-roof tanks contain flammable substances that ignite, the release probability needs to be multiplied with the conditional ignition probability, which decreases the final consequence probability for these tanks.

For the conditions of the Istanbul earthquake, RAPID-N predicts releases from all six fixed-roof tanks containing methanol and acetic acid. As already mentioned in Section 10.3.2, the end-point distances for heat radiation from the pool fires originating from these tanks envelop other tank locations. The potential for domino effects is therefore evident and should be considered for a more accurate assessment of the Natech risk from these tanks. It should also be noted that although an increased minimum pool depth is utilized to better simulate pool dimensions in case of dike overflows, adjoining dikes as in the case of this facility may result in smaller pool dimensions if the overflow runs to other dikes. A more detailed analysis is suggested for such scenarios.